Compare commits

..
Author SHA1 Message Date
“haonan”andCursor f7e08f461b feat(record): capture interactions in embedded frames
Track recording per document so injectable child frames are captured while partial frame failures remain visible in exported traces.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-17 14:41:02 +08:00
“haonan” 61a0e0d218 fix(record): keep committed empty fill steps in exported traces 2026-08-17 11:02:10 +08:00
“haonan” 67b7227abf fix(popup): point record prompt at the ./trace directory bundle 2026-08-13 19:55:34 +08:00
“haonan” 02d1d69379 docs(ux): clarify version mismatches and update record skill docs
Surface actionable update guidance in the popup and connection errors, document
the trace v3 directory bundle in SKILL.md.
2026-08-13 19:40:54 +08:00
“haonan” 025751ef03 feat(cli): write record output as a trace directory bundle
Persist `trace.json` plus per-state page VOM files under `--output`, tolerate
raced session teardown on stop, and clarify version-mismatch recovery hints.
2026-08-13 19:38:05 +08:00
“haonan” 13a61b2688 feat(extension): record user flows as trace v3 page-state observations
Capture settled page observations into a states/steps trace, match targets by
geometry with capture fallbacks, and reuse VOM rendering with richer refs and
optional value redaction for record bundles.
2026-08-13 19:34:37 +08:00
“haonan” 1508839bf9 feat(protocol): bump to 1.1 and redefine record traces as state-linked v3
Replace page-indexed record traces with states/steps + FillCommit/hover/scroll,
and raise daemon/extension protocol floors together so old 1.0 peers are rejected.
2026-08-13 19:31:35 +08:00
174 changed files with 7983 additions and 21198 deletions
-105
View File
@@ -1,105 +0,0 @@
name: Release dsh plugin
on:
push:
tags:
- dsh-plugin-v*
workflow_dispatch:
permissions:
contents: read
concurrency:
group: release-dsh-plugin-${{ github.ref }}
cancel-in-progress: false
env:
PACKAGE_NAME: "@wxg-prc-cpg/browser-skill-dsh-plugin"
PACKAGE_DIR: packages/dsh-plugin-browserskill
jobs:
resolve:
runs-on: ubuntu-latest
outputs:
version: ${{ steps.version.outputs.version }}
tag: ${{ steps.version.outputs.tag }}
steps:
- uses: actions/checkout@v6
- name: Resolve plugin version
id: version
shell: bash
run: |
set -euo pipefail
if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
VERSION="$(node -p "require('./${{ env.PACKAGE_DIR }}/package.json').version")"
else
VERSION="${GITHUB_REF_NAME#dsh-plugin-v}"
fi
VERSION="${VERSION#v}"
echo "version=${VERSION}" >> "$GITHUB_OUTPUT"
echo "tag=dsh-plugin-v${VERSION}" >> "$GITHUB_OUTPUT"
guard:
needs: resolve
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Verify tag matches package.json name and version
shell: bash
run: |
set -euo pipefail
PKG_NAME="$(node -p "require('./${{ env.PACKAGE_DIR }}/package.json').name")"
PKG_VERSION="$(node -p "require('./${{ env.PACKAGE_DIR }}/package.json').version")"
RELEASE_VERSION="${{ needs.resolve.outputs.version }}"
if [ "$PKG_NAME" != "$PACKAGE_NAME" ]; then
echo "package.json name (${PKG_NAME}) does not match expected ${PACKAGE_NAME}"
exit 1
fi
if [ "$PKG_VERSION" != "$RELEASE_VERSION" ]; then
echo "${PACKAGE_DIR}/package.json version (${PKG_VERSION}) does not match release version (${RELEASE_VERSION})"
exit 1
fi
echo "Version guard passed: ${PKG_NAME}@${RELEASE_VERSION}"
publish:
needs: [resolve, guard]
runs-on: ubuntu-latest
# Secret NPM_TOKEN is stored on the "npm-publish" GitHub Environment.
environment: npm-publish
permissions:
contents: read
steps:
- uses: actions/checkout@v6
- uses: pnpm/action-setup@v6
- uses: actions/setup-node@v6
with:
node-version: 22
cache: pnpm
registry-url: https://registry.npmjs.org
scope: "@wxg-prc-cpg"
- name: Install dependencies
run: pnpm install --frozen-lockfile
- name: Typecheck plugin
run: pnpm --filter "${{ env.PACKAGE_NAME }}" typecheck
- name: Test plugin
run: pnpm --filter "${{ env.PACKAGE_NAME }}" test
- name: Publish to npm
working-directory: ${{ env.PACKAGE_DIR }}
env:
NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
run: |
set -euo pipefail
if [ -z "${NODE_AUTH_TOKEN:-}" ]; then
echo "::error::Environment secret NPM_TOKEN is not configured"
exit 1
fi
pnpm publish --access public --no-git-checks
echo "Published ${PACKAGE_NAME}@${{ needs.resolve.outputs.version }}"
+4
View File
@@ -17,3 +17,7 @@ dist/
# Environment
.env.local
.env.*.local
.pnpm-store/
# Local recording output
/trace/
-3
View File
@@ -3,6 +3,3 @@
**/.output/**
**/target/**
**/node_modules/**
# Generated client CSS (tailwind compile output); built by
# packages/dsh-plugin-browserskill/scripts/build-client-css.mjs.
**/dsh-plugin-browserskill/src/client/bsk-ui.nomodule.css
+5 -9
View File
@@ -34,17 +34,13 @@ where only `extension connected` fails is expected; go to Step 3.
## 3. Open the extension install page
If `extension connected` is `FAIL` (`0 browsers connected`), the user likely
has not installed or enabled the browser extension yet. Open the install page
matching the user's browser —
[Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi)
for Chrome and other Chromium browsers,
[Edge Add-ons](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg)
for Microsoft Edge — then tell the user:
has not installed or enabled the browser extension yet. Open the
[Chrome Web Store install page](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi)
in the user's browser, then tell the user:
> Install **BrowserSkill** from the
> [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi)
> (or [Edge Add-ons](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg)
> on Edge), open the popup, and wait until it turns green. Reply when done.
> [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi),
> open the popup, and wait until it turns green. Reply when done.
If opening the page fails, give the user the same link. Then run `bsk doctor`
once more. All `ok`/`na` → tell the user it's ready.
+5 -37
View File
@@ -13,7 +13,7 @@
</p>
**BrowserSkill** connects Cursor, Claude Code, Codex, OpenClaw, CodeBuddy,
WorkBuddy, Pi, Hermes Agent, DeepSeek Harness, and other AI agents to your already logged-in
WorkBuddy, Pi, Hermes Agent, and other shell-capable AI agents to your already logged-in
browser.
Need the agent to touch a tab you already have open? It must borrow that tab
@@ -67,8 +67,7 @@ Set up browser-skill on this machine by following https://raw.githubusercontent.
<br>
Install the CLI, then install the extension from the [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi)
or [Edge Add-ons](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg).
Install the CLI, then install the extension from the [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi).
#### 1. Install the `bsk` CLI
@@ -92,14 +91,7 @@ bsk --version
#### 2. Install the browser extension
Install BrowserSkill from your browser's store:
| Browser | Store listing |
| --- | --- |
| Chrome | [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi) |
| Microsoft Edge | [Edge Add-ons](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg) |
On other Chromium-based browsers, install the Chrome Web Store build.
Install BrowserSkill from the [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi).
#### 3. Install the skill
@@ -131,8 +123,7 @@ internal variants and install paths.
Other shell-capable agent harnesses are supported too. Copy
[`skill/SKILL.md`](skill/SKILL.md) into your harness's skills directory as
`browser-skill/SKILL.md` to install the skill manually. DeepSeek Harness uses a
dedicated plugin instead — see [DeepSeek Harness plugin](#deepseek-harness-plugin).
`browser-skill/SKILL.md` to install the skill manually.
</details>
@@ -142,26 +133,6 @@ Start a new Agent session and write a prompt that needs the browser, for example
/browser-skill open example.com and summarize what is on the page.
```
## DeepSeek Harness plugin
Using [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (`dsh`)?
BrowserSkill ships a first-class dsh plugin on npm as
[`@wxg-prc-cpg/browser-skill-dsh-plugin`](https://www.npmjs.com/package/@wxg-prc-cpg/browser-skill-dsh-plugin).
It injects native `browser_*` tools (no shelling out to `bsk`) and a live Web UI
overlay of each Agent Window.
Add it to a dsh profile, then start that profile:
```sh
dsh plugin --profile web add @wxg-prc-cpg/browser-skill-dsh-plugin
dsh --profile web
```
The plugin carries its own copy of the skill, so `bsk install-skill` is not needed
for dsh — but the `bsk` CLI and the browser extension are still prerequisites. See
the [plugin README](packages/dsh-plugin-browserskill/README.md) for the tool list,
configuration, and the observation overlay.
## How It Works
BrowserSkill is a local bridge between your agent harness and your browser.
@@ -195,9 +166,7 @@ flowchart TB
The agent never talks to the browser directly. It asks the `bsk` CLI to perform a
browser task; the local daemon routes that request to the extension; the
extension runs it in an Agent Window. DeepSeek Harness takes the same path
through the [plugin](#deepseek-harness-plugin): the agent calls injected
`browser_*` tools, and the plugin invokes `bsk` on its behalf.
extension runs it in an Agent Window.
## For Developers
@@ -207,7 +176,6 @@ The repository is a Cargo + pnpm workspace:
- `crates/bsk-protocol` — shared wire types and JSON schemas
- `apps/extension` — browser extension
- `packages/ui` and `packages/i18n` — shared extension UI support
- `packages/dsh-plugin-browserskill` — DeepSeek Harness plugin (`@wxg-prc-cpg/browser-skill-dsh-plugin`)
## License
+7 -32
View File
@@ -12,7 +12,7 @@
<a href="README.md">English</a> · 中文
</p>
**BrowserSkill** 把 Cursor、Claude Code、Codex、OpenClaw、CodeBuddy、WorkBuddy、Pi、Hermes Agent、DeepSeek Harness 等 AI Agent 连接到你已登录的浏览器。
**BrowserSkill** 把 Cursor、Claude Code、Codex、OpenClaw、CodeBuddy、WorkBuddy、Pi、Hermes Agent 等支持 Shell 的 AI Agent 连接到你已登录的浏览器。
需要 Agent 操作你已打开的标签页?必须显式借用该标签,任务结束后归还,其余浏览器窗口不受影响。
@@ -54,8 +54,7 @@ BrowserSkill 由两个本地运行组件组成:`bsk` CLI/daemon 和浏览器
<br>
先安装 CLI,再从 [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi)
或 [Edge 加载项商店](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg) 安装浏览器扩展。
先安装 CLI,再从 [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi) 安装浏览器扩展。
#### 1. 安装 `bsk` CLI
@@ -65,11 +64,8 @@ BrowserSkill 由两个本地运行组件组成:`bsk` CLI/daemon 和浏览器
curl -fsSL https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.sh | sh
```
**Windows**(PowerShell,安装到 `~/.local/bin`):
```powershell
irm https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.ps1 | iex
```
**Windows**:从 [最新 CLI release](https://github.com/Tencent/BrowserSkill/releases/latest)
下载 `bsk-v<version>-x86_64-pc-windows-msvc.zip`,解压后将 `bsk.exe` 加入 `PATH`。
验证二进制:
@@ -79,14 +75,7 @@ bsk --version
#### 2. 安装浏览器扩展
在对应浏览器的商店安装 BrowserSkill:
| 浏览器 | 商店页面 |
| --- | --- |
| Chrome | [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi) |
| Microsoft Edge | [Edge 加载项商店](https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg) |
其他基于 Chromium 的浏览器,安装 Chrome Web Store 版本即可。
从 [Chrome Web Store](https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi) 安装 BrowserSkill。
#### 3. 安装 skill
@@ -113,7 +102,7 @@ bsk install-skill
用 <kbd>Space</kbd> 选择需要安装的 Agent harness,然后按 <kbd>Enter</kbd> 安装 skill。运行 `bsk install-skill --list` 可查看 internal 变体及安装路径。
其他支持 Shell 的 Agent harness 也可使用 BrowserSkill,但需手动将 [`skill/SKILL.md`](skill/SKILL.md) 复制到对应 skills 目录下的 `browser-skill/SKILL.md`。DeepSeek Harness 走独立插件,见 [DeepSeek Harness 插件](#deepseek-harness-插件)。
其他支持 Shell 的 Agent harness 也可使用 BrowserSkill,但需手动将 [`skill/SKILL.md`](skill/SKILL.md) 复制到对应 skills 目录下的 `browser-skill/SKILL.md`。
</details>
@@ -123,19 +112,6 @@ bsk install-skill
/browser-skill open example.com and summarize what is on the page.
```
## DeepSeek Harness 插件
在用 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)(`dsh`)?BrowserSkill 提供了官方 dsh 插件,已发布到 npm:[`@wxg-prc-cpg/browser-skill-dsh-plugin`](https://www.npmjs.com/package/@wxg-prc-cpg/browser-skill-dsh-plugin)。它会注入原生 `browser_*` 工具(无需再通过 Shell 调用 `bsk`),并在 Web UI 中实时观察每个 Agent Window。
把它装进某个 dsh profile,然后启动该 profile:
```sh
dsh plugin --profile web add @wxg-prc-cpg/browser-skill-dsh-plugin
dsh --profile web
```
插件自带 skill,所以在 dsh 下无需执行 `bsk install-skill`;但 `bsk` CLI 和浏览器扩展仍是前置条件。工具清单、配置项与观察浮层见[插件 README](packages/dsh-plugin-browserskill/README.md)。
## 工作原理
BrowserSkill 是 Agent 运行时与浏览器之间的本地桥接层。
@@ -167,7 +143,7 @@ flowchart TB
style UserWindows fill:#f8fafc,stroke:#cbd5e1,color:#334155
```
Agent 不直接与浏览器通信。它通过 `bsk` CLI 下发浏览器任务;本地 daemon 把请求路由到扩展;扩展在 Agent Window 中执行。DeepSeek Harness 走同一条链路,只是经由 [插件](#deepseek-harness-插件):Agent 调用注入的 `browser_*` 工具,由插件代为执行 `bsk`。
Agent 不直接与浏览器通信。它通过 `bsk` CLI 下发浏览器任务;本地 daemon 把请求路由到扩展;扩展在 Agent Window 中执行。
## 面向开发者
@@ -177,7 +153,6 @@ Agent 不直接与浏览器通信。它通过 `bsk` CLI 下发浏览器任务;
- `crates/bsk-protocol` — 共享协议类型与 JSON Schema
- `apps/extension` — 浏览器扩展
- `packages/ui` 和 `packages/i18n` — 扩展 UI 共享支持
- `packages/dsh-plugin-browserskill` — DeepSeek Harness 插件(`@wxg-prc-cpg/browser-skill-dsh-plugin`)
## 许可证
@@ -1,5 +1,5 @@
import { describe, expect, it, vi } from "vitest";
import { type CdpDebuggee, type CdpDebuggerApi, ChromiumCdp } from "../chromium-cdp";
import { type CdpDebuggerApi, ChromiumCdp } from "../chromium-cdp";
function fakeChromeEvent<TArgs extends unknown[]>() {
const listeners = new Set<(...args: TArgs) => void>();
@@ -14,7 +14,7 @@ function fakeChromeEvent<TArgs extends unknown[]>() {
}
function fakeApi() {
const onEvent = fakeChromeEvent<[CdpDebuggee, string, unknown]>();
const onEvent = fakeChromeEvent<[chrome.debugger.Debuggee, string, unknown]>();
const onDetach = fakeChromeEvent<[chrome.debugger.Debuggee, string]>();
const api: CdpDebuggerApi = {
attach: vi.fn(async () => {}),
@@ -29,135 +29,6 @@ function fakeApi() {
}
describe("ChromiumCdp", () => {
it("discovers multiple iframe targets and recursively routes nested OOPIF commands", async () => {
const { api, onEvent } = fakeApi();
(api.sendCommand as ReturnType<typeof vi.fn>).mockImplementation(
async (
target: chrome.debugger.Debuggee & { sessionId?: string },
method: string,
params?: { frameId?: string },
) => {
if (method === "DOM.getFrameOwner") {
return { backendNodeId: params?.frameId === "nested" ? 300 : 200 };
}
if (method !== "Page.getFrameTree") return {};
if (target.sessionId === "right-session") {
return {
frameTree: {
frame: { id: "right", url: "https://right.test" },
childFrames: [
{ frame: { id: "nested", parentId: "right", url: "https://nested.test" } },
],
},
};
}
if (target.sessionId === "nested-session") {
return { frameTree: { frame: { id: "nested", url: "https://nested.test" } } };
}
return {
frameTree: {
frame: { id: "main", url: "https://app.test" },
childFrames: [
{ frame: { id: "left", parentId: "main", url: "https://left.test" } },
{
frame: { id: "right", parentId: "main", url: "https://right.test" },
childFrames: [
{ frame: { id: "nested", parentId: "right", url: "https://nested.test" } },
],
},
],
},
};
},
);
const cdp = new ChromiumCdp(api);
await cdp.ensureAttached(4);
onEvent.fire({ tabId: 4 }, "Target.attachedToTarget", { sessionId: "right-session" });
onEvent.fire({ tabId: 4 } as chrome.debugger.Debuggee, "Target.attachedToTarget", {
sessionId: "nested-session",
});
const graph = await cdp.getFrameGraph(4);
expect(graph.frames).toHaveLength(4);
expect(graph.frames.find((frame) => frame.frameId === "left")?.target).toEqual({ tabId: 4 });
expect(graph.frames.find((frame) => frame.frameId === "right")?.target.sessionId).toBe(
"right-session",
);
expect(graph.frames.find((frame) => frame.frameId === "nested")?.target.sessionId).toBe(
"nested-session",
);
expect(graph.frames.find((frame) => frame.frameId === "right")?.ownerBackendNodeId).toBe(200);
expect(graph.frames.find((frame) => frame.frameId === "nested")?.ownerBackendNodeId).toBe(300);
expect(api.sendCommand).toHaveBeenCalledWith(
{ tabId: 4, sessionId: "right-session" },
"DOM.getFrameOwner",
{ frameId: "nested" },
);
expect(api.sendCommand).toHaveBeenCalledWith(
{ tabId: 4, sessionId: "nested-session" },
"Target.setAutoAttach",
expect.objectContaining({ flatten: true }),
);
});
it("waits for recursively attached iframe targets without a depth limit", async () => {
const { api, onEvent } = fakeApi();
const depth = 6;
const scheduled = new Set<string>();
const nestedTree = (index: number): Record<string, unknown> => ({
frame: {
id: index === 0 ? "main" : `frame-${index}`,
...(index > 0 ? { parentId: index === 1 ? "main" : `frame-${index - 1}` } : {}),
},
...(index < depth ? { childFrames: [nestedTree(index + 1)] } : {}),
});
(api.sendCommand as ReturnType<typeof vi.fn>).mockImplementation(
async (target: CdpDebuggee, method: string, params?: { frameId?: string }) => {
if (method === "Target.setAutoAttach") {
const parentIndex = target.sessionId
? Number(target.sessionId.replace("session-", ""))
: 0;
const nextIndex = parentIndex + 1;
const key = `${target.sessionId ?? "root"}:${nextIndex}`;
if (nextIndex <= depth && !scheduled.has(key)) {
scheduled.add(key);
setTimeout(() => {
onEvent.fire(
{ tabId: 4, ...(target.sessionId ? { sessionId: target.sessionId } : {}) },
"Target.attachedToTarget",
{
sessionId: `session-${nextIndex}`,
targetInfo: { type: "iframe" },
},
);
}, 5);
}
return {};
}
if (method === "Page.getFrameTree") {
if (!target.sessionId) return { frameTree: nestedTree(0) };
const index = Number(target.sessionId.replace("session-", ""));
return { frameTree: nestedTree(index) };
}
if (method === "DOM.getFrameOwner") {
return { backendNodeId: Number(params?.frameId?.replace("frame-", "")) + 100 };
}
return {};
},
);
const cdp = new ChromiumCdp(api);
const graph = await cdp.getFrameGraph(4);
expect(graph.frames).toHaveLength(depth + 1);
for (let index = 1; index <= depth; index += 1) {
expect(
graph.frames.find((frame) => frame.frameId === `frame-${index}`)?.target.sessionId,
).toBe(`session-${index}`);
}
});
it("coalesces concurrent attach calls for the same tab", async () => {
const { api } = fakeApi();
let releaseAttach!: () => void;
@@ -1,136 +0,0 @@
import { describe, expect, it } from "vitest";
import { buildFrameGraph, type CdpFrameTreeNode, type CdpFrameTreeSource } from "../frame-graph";
describe("buildFrameGraph", () => {
it("keeps sibling frames distinct and routes nested OOPIFs to their child sessions", () => {
const graph = buildFrameGraph([
{
target: { tabId: 4 },
tree: {
frame: { id: "main", url: "https://app.test" },
childFrames: [
{ frame: { id: "left", parentId: "main", url: "https://left.test" } },
{
frame: { id: "right", parentId: "main", url: "https://right.test" },
childFrames: [
{ frame: { id: "nested", parentId: "right", url: "https://nested.test" } },
],
},
],
},
},
{
target: { tabId: 4, sessionId: "right-session" },
tree: {
frame: { id: "right", url: "https://right.test" },
childFrames: [{ frame: { id: "nested", parentId: "right", url: "https://nested.test" } }],
},
},
{
target: { tabId: 4, sessionId: "nested-session" },
tree: { frame: { id: "nested", url: "https://nested.test" } },
},
]);
expect(graph?.frames).toHaveLength(4);
expect(graph?.frames.find((frame) => frame.frameId === "left")?.target).toEqual({ tabId: 4 });
expect(graph?.frames.find((frame) => frame.frameId === "right")).toMatchObject({
parentFrameId: "main",
target: { tabId: 4, sessionId: "right-session" },
});
expect(graph?.frames.find((frame) => frame.frameId === "nested")).toMatchObject({
parentFrameId: "right",
target: { tabId: 4, sessionId: "nested-session" },
});
});
it("propagates each target boundary to its same-process descendants", () => {
const sources: CdpFrameTreeSource[] = [
{
target: { tabId: 4 },
tree: {
frame: { id: "main" },
childFrames: [
{
frame: { id: "oopif-a", parentId: "main" },
childFrames: [
{
frame: { id: "same-process-b", parentId: "oopif-a" },
childFrames: [
{
frame: { id: "oopif-c", parentId: "same-process-b" },
childFrames: [{ frame: { id: "same-process-d", parentId: "oopif-c" } }],
},
],
},
],
},
],
},
},
{
target: { tabId: 4, sessionId: "session-a" },
tree: {
frame: { id: "oopif-a" },
childFrames: [
{
frame: { id: "same-process-b", parentId: "oopif-a" },
childFrames: [{ frame: { id: "oopif-c", parentId: "same-process-b" } }],
},
],
},
},
{
target: { tabId: 4, sessionId: "session-c" },
tree: {
frame: { id: "oopif-c" },
childFrames: [{ frame: { id: "same-process-d", parentId: "oopif-c" } }],
},
},
];
const graph = buildFrameGraph(sources);
expect(graph?.frames.map((frame) => [frame.frameId, frame.target.sessionId])).toEqual([
["main", undefined],
["oopif-a", "session-a"],
["same-process-b", "session-a"],
["oopif-c", "session-c"],
["same-process-d", "session-c"],
]);
expect(graph?.frames.find((frame) => frame.frameId === "oopif-a")?.parentFrameId).toBe("main");
expect(graph?.frames.find((frame) => frame.frameId === "oopif-c")?.parentFrameId).toBe(
"same-process-b",
);
const reordered = buildFrameGraph([sources[2], sources[1], sources[0]]);
expect(reordered).toEqual(graph);
});
it("walks deeply nested frame trees without consuming the JavaScript call stack", () => {
const depth = 5_000;
const root: CdpFrameTreeNode = { frame: { id: "frame-0" } };
let parent = root;
for (let index = 1; index <= depth; index += 1) {
const child: CdpFrameTreeNode = {
frame: { id: `frame-${index}`, parentId: `frame-${index - 1}` },
};
parent.childFrames = [child];
parent = child;
}
const graph = buildFrameGraph([{ target: { tabId: 4 }, tree: root }]);
expect(graph?.frames).toHaveLength(depth + 1);
expect(graph?.frames.at(-1)?.frameId).toBe(`frame-${depth}`);
});
it("does not loop when an in-memory frame tree contains an object cycle", () => {
const root: CdpFrameTreeNode = { frame: { id: "main" } };
root.childFrames = [root];
expect(buildFrameGraph([{ target: { tabId: 4 }, tree: root }])?.frames).toEqual([
{ frameId: "main", target: { tabId: 4 } },
]);
});
});
+12 -211
View File
@@ -31,15 +31,6 @@ import type {
NetworkEntryKind,
NetworkResult,
} from "@/transport/types";
import {
buildFrameGraph,
type CdpFrameGraph,
type CdpFrameTreeNode,
type CdpFrameTreeSource,
type CdpTarget,
} from "./frame-graph";
export type CdpDebuggee = chrome.debugger.Debuggee & { sessionId?: string };
/**
* Minimal slice of `chrome.debugger` the rest of the extension
@@ -47,15 +38,21 @@ export type CdpDebuggee = chrome.debugger.Debuggee & { sessionId?: string };
* fake without monkey-patching the real `chrome` global.
*/
export interface CdpDebuggerApi {
attach(target: CdpDebuggee, requiredVersion: string): Promise<void>;
detach(target: CdpDebuggee): Promise<void>;
sendCommand(target: CdpDebuggee, method: string, commandParams?: object): Promise<unknown>;
attach(target: chrome.debugger.Debuggee, requiredVersion: string): Promise<void>;
detach(target: chrome.debugger.Debuggee): Promise<void>;
sendCommand(
target: chrome.debugger.Debuggee,
method: string,
commandParams?: object,
): Promise<unknown>;
/**
* Fires for every CDP event (`Page.lifecycleEvent`, `DOM.documentUpdated`,
* …). The first callback argument is the source debuggee; the second
* is the CDP method name; the third is the payload.
*/
onEvent: chrome.events.Event<(source: CdpDebuggee, method: string, params: unknown) => void>;
onEvent: chrome.events.Event<
(source: chrome.debugger.Debuggee, method: string, params: unknown) => void
>;
/**
* Fires when Chrome unilaterally detaches us — most commonly because
* the tab navigated to a chrome:// URL or the user clicked
@@ -113,8 +110,6 @@ const MAX_CONSOLE_STACK_FRAMES = 20;
const MAX_NETWORK_BUFFER = 200;
const MAX_NETWORK_FIELD_LENGTH = 4096;
const MAX_NETWORK_REQUEST_META = 1024;
const FRAME_DISCOVERY_TIMEOUT_MS = 1000;
const FRAME_DISCOVERY_QUIET_MS = 20;
interface ParsedDialogOpening {
type: JavaScriptDialogType;
@@ -136,27 +131,6 @@ interface NetworkRequestMeta {
truncated: boolean;
}
interface FrameDiscoveryState {
sessions: Set<string>;
pending: Set<Promise<void>>;
generation: number;
}
async function settleBeforeDeadline(promises: Promise<void>[], deadline: number): Promise<boolean> {
if (promises.length === 0) return true;
const remaining = deadline - Date.now();
if (remaining <= 0) return false;
let timer: ReturnType<typeof setTimeout> | undefined;
const timedOut = await Promise.race([
Promise.allSettled(promises).then(() => false),
new Promise<boolean>((resolve) => {
timer = setTimeout(() => resolve(true), remaining);
}),
]);
if (timer) clearTimeout(timer);
return !timedOut;
}
/**
* Wrapper around `chrome.debugger` that owns the "attach once per
* tabId" cache and exposes typed `send<T>()`.
@@ -175,12 +149,10 @@ export class ChromiumCdp {
private readonly networkSequences = new Map<number, number>();
private readonly networkDomainsEnabledTabs = new Set<number>();
private readonly networkRequestMeta = new Map<number, Map<string, NetworkRequestMeta>>();
private readonly frameDiscovery = new Map<number, FrameDiscoveryState>();
private detachSubscription: { dispose(): void } | null = null;
private dialogSubscription: { dispose(): void } | null = null;
private consoleSubscription: { dispose(): void } | null = null;
private networkSubscription: { dispose(): void } | null = null;
private frameTargetSubscription: { dispose(): void } | null = null;
constructor(api: CdpDebuggerApi = chromeDebuggerApi) {
this.api = api;
@@ -188,7 +160,6 @@ export class ChromiumCdp {
this.bindDialogHandler();
this.bindConsoleHandler();
this.bindNetworkHandler();
this.bindFrameTargetHandler();
}
/** Attach to `tabId` if we haven't already in this driver. */
@@ -206,9 +177,6 @@ export class ChromiumCdp {
await this.enableConsoleDomains(tabId);
await this.enableNetworkDomainBestEffort(tabId);
this.attachedTabs.add(tabId);
await this.enableFrameDiscovery({ tabId }).catch((err) => {
console.debug("[bsk cdp] frame discovery unavailable", { tabId, err });
});
} catch (err) {
// A CDP domain enable failed after the raw attach succeeded
// (e.g. `Page.enable` rejects because the tab just navigated to
@@ -254,73 +222,6 @@ export class ChromiumCdp {
}
}
async sendToTarget<T = unknown>(target: CdpTarget, method: string, params?: object): Promise<T> {
if (!this.attachedTabs.has(target.tabId)) {
await this.ensureAttached(target.tabId);
}
try {
return (await this.api.sendCommand(target, method, params ?? {})) as T;
} catch (err) {
throw normalizeError(err);
}
}
async getFrameGraph(tabId: number): Promise<CdpFrameGraph> {
await this.ensureAttached(tabId);
await this.enableFrameDiscovery({ tabId }).catch(() => {});
await this.drainFrameAttachTasks(tabId);
const sources: CdpFrameTreeSource[] = [];
const root = await this.sendToTarget<{ frameTree?: CdpFrameTreeNode }>(
{ tabId },
"Page.getFrameTree",
{},
);
if (root.frameTree) sources.push({ target: { tabId }, tree: root.frameTree });
const sessions = [...(this.frameDiscovery.get(tabId)?.sessions ?? [])];
const childTrees = await Promise.all(
sessions.map(async (sessionId): Promise<CdpFrameTreeSource | null> => {
const target = { tabId, sessionId };
try {
const reply = await this.sendToTarget<{ frameTree?: CdpFrameTreeNode }>(
target,
"Page.getFrameTree",
{},
);
return reply.frameTree ? { target, tree: reply.frameTree } : null;
} catch {
return null;
}
}),
);
for (const source of childTrees) {
if (source) sources.push(source);
}
const graph = buildFrameGraph(sources);
if (!graph) throw new Error("Page.getFrameTree returned no root frame");
const frameById = new Map(graph.frames.map((frame) => [frame.frameId, frame]));
await Promise.all(
graph.frames.map(async (frame) => {
if (!frame.parentFrameId) return;
const parent = frameById.get(frame.parentFrameId);
if (!parent) return;
try {
const owner = await this.sendToTarget<{ backendNodeId?: number }>(
parent.target,
"DOM.getFrameOwner",
{ frameId: frame.frameId },
);
if (owner.backendNodeId !== undefined) frame.ownerBackendNodeId = owner.backendNodeId;
} catch {
// The frame may have navigated between tree capture and owner lookup.
}
}),
);
return graph;
}
/** Return a cursor marking the current dialog sequence for `tabId`. */
dialogCursor(tabId: number): DialogCursor {
return this.dialogSequences.get(tabId) ?? 0;
@@ -431,7 +332,6 @@ export class ChromiumCdp {
this.clearDialogState(tabId);
this.clearConsoleState(tabId);
this.clearNetworkState(tabId);
this.clearFrameState(tabId);
try {
await this.api.detach({ tabId });
} catch (err) {
@@ -454,7 +354,7 @@ export class ChromiumCdp {
}
/** Subscribe to all CDP events. Returned disposable removes the listener. */
onEvent(handler: (source: CdpDebuggee, method: string, params: unknown) => void): {
onEvent(handler: (source: chrome.debugger.Debuggee, method: string, params: unknown) => void): {
dispose(): void;
} {
this.api.onEvent.addListener(handler);
@@ -478,7 +378,6 @@ export class ChromiumCdp {
this.networkSequences.clear();
this.networkDomainsEnabledTabs.clear();
this.networkRequestMeta.clear();
this.frameDiscovery.clear();
await Promise.all(
tabs.map(async (tabId) => {
try {
@@ -529,104 +428,9 @@ export class ChromiumCdp {
}
}
private async enableFrameDiscovery(target: CdpTarget): Promise<void> {
await this.api.sendCommand(target, "Target.setAutoAttach", {
autoAttach: true,
waitForDebuggerOnStart: false,
flatten: true,
filter: [{ type: "iframe", exclude: false }],
});
}
private bindFrameTargetHandler(): void {
if (this.frameTargetSubscription) return;
const listener = (source: CdpDebuggee, method: string, params: unknown) => {
const tabId = source.tabId;
if (typeof tabId !== "number") return;
const raw = (params ?? {}) as Record<string, unknown>;
const sessionId = typeof raw.sessionId === "string" ? raw.sessionId : undefined;
if (!sessionId) return;
if (method === "Target.detachedFromTarget") {
const state = this.frameDiscovery.get(tabId);
if (state?.sessions.delete(sessionId)) state.generation += 1;
return;
}
if (method !== "Target.attachedToTarget") return;
const targetInfo = raw.targetInfo as { type?: string } | undefined;
if (targetInfo?.type && targetInfo.type !== "iframe") return;
const state = this.frameDiscoveryState(tabId);
if (state.sessions.has(sessionId)) return;
state.sessions.add(sessionId);
state.generation += 1;
const task = this.initializeFrameTarget({ tabId, sessionId });
state.pending.add(task);
void task.finally(() => {
state.pending.delete(task);
});
};
this.api.onEvent.addListener(listener);
this.frameTargetSubscription = {
dispose: () => this.api.onEvent.removeListener(listener),
};
}
private async initializeFrameTarget(target: CdpTarget): Promise<void> {
try {
await this.enableFrameDiscovery(target);
} catch (err) {
const state = this.frameDiscovery.get(target.tabId);
if (state?.sessions.delete(target.sessionId as string)) state.generation += 1;
console.debug("[bsk cdp] child frame target initialization failed", { target, err });
}
}
private async drainFrameAttachTasks(tabId: number): Promise<void> {
const deadline = Date.now() + FRAME_DISCOVERY_TIMEOUT_MS;
while (Date.now() < deadline) {
const state = this.frameDiscovery.get(tabId);
const generation = state?.generation ?? 0;
if (state?.pending.size && !(await settleBeforeDeadline([...state.pending], deadline))) break;
// Target.setAutoAttach may enqueue the next attachedToTarget event after
// its command promise settles. Wait for a short quiet window, then finish
// only if neither the state object nor its generation changed.
const quietTime = Math.min(FRAME_DISCOVERY_QUIET_MS, deadline - Date.now());
if (quietTime <= 0) break;
await new Promise<void>((resolve) => setTimeout(resolve, quietTime));
const current = this.frameDiscovery.get(tabId);
if (
current === state &&
(current?.generation ?? 0) === generation &&
!current?.pending.size
) {
return;
}
}
console.debug("[bsk cdp] frame discovery did not reach quiescence before timeout", { tabId });
}
private frameDiscoveryState(tabId: number): FrameDiscoveryState {
const existing = this.frameDiscovery.get(tabId);
if (existing) return existing;
const created: FrameDiscoveryState = {
sessions: new Set(),
pending: new Set(),
generation: 0,
};
this.frameDiscovery.set(tabId, created);
return created;
}
private clearFrameState(tabId: number): void {
this.frameDiscovery.delete(tabId);
}
private bindDialogHandler(): void {
if (this.dialogSubscription) return;
const listener = (source: CdpDebuggee, method: string, params: unknown) => {
const listener = (source: chrome.debugger.Debuggee, method: string, params: unknown) => {
if (method !== "Page.javascriptDialogOpening") return;
const tabId = source.tabId;
if (typeof tabId !== "number") return;
@@ -810,7 +614,6 @@ export class ChromiumCdp {
this.clearDialogState(source.tabId);
this.clearConsoleState(source.tabId);
this.clearNetworkState(source.tabId);
this.clearFrameState(source.tabId);
}
};
this.api.onDetach.addListener(listener);
@@ -829,8 +632,6 @@ export class ChromiumCdp {
this.consoleSubscription = null;
this.networkSubscription?.dispose();
this.networkSubscription = null;
this.frameTargetSubscription?.dispose();
this.frameTargetSubscription = null;
}
/** Detach tabs only when no other live session has claimed them. */
@@ -1,107 +0,0 @@
export interface CdpTarget {
tabId: number;
sessionId?: string;
}
export interface CdpFrame {
frameId: string;
parentFrameId?: string;
ownerBackendNodeId?: number;
url?: string;
target: CdpTarget;
}
export interface CdpFrameGraph {
rootFrameId: string;
frames: CdpFrame[];
}
export interface CdpFrameTreeNode {
frame: {
id: string;
parentId?: string;
url?: string;
};
childFrames?: CdpFrameTreeNode[];
}
export interface CdpFrameTreeSource {
target: CdpTarget;
tree: CdpFrameTreeNode;
}
export function cdpTargetKey(target: CdpTarget): string {
return `${target.tabId}:${target.sessionId ?? "root"}`;
}
function mergeFrame(
frames: Map<string, CdpFrame>,
order: string[],
node: CdpFrameTreeNode,
target: CdpTarget,
): void {
const existing = frames.get(node.frame.id);
if (!existing) order.push(node.frame.id);
const parentFrameId = node.frame.parentId || existing?.parentFrameId;
const url = node.frame.url || existing?.url;
frames.set(node.frame.id, {
frameId: node.frame.id,
...(parentFrameId ? { parentFrameId } : {}),
...(url ? { url } : {}),
target,
});
}
function walkFrameTree(
source: CdpFrameTreeSource,
targetByRootFrameId: ReadonlyMap<string, CdpTarget>,
frames: Map<string, CdpFrame>,
order: string[],
): void {
const stack: Array<{ node: CdpFrameTreeNode; inheritedTarget: CdpTarget }> = [
{ node: source.tree, inheritedTarget: source.target },
];
const expanded = new WeakSet<CdpFrameTreeNode>();
while (stack.length > 0) {
const current = stack.pop();
if (!current || expanded.has(current.node)) continue;
expanded.add(current.node);
const target = targetByRootFrameId.get(current.node.frame.id) ?? current.inheritedTarget;
mergeFrame(frames, order, current.node, target);
const children = current.node.childFrames ?? [];
for (let index = children.length - 1; index >= 0; index -= 1) {
stack.push({ node: children[index], inheritedTarget: target });
}
}
}
export function buildFrameGraph(sources: CdpFrameTreeSource[]): CdpFrameGraph | null {
const rootSource = sources.find((source) => source.target.sessionId === undefined) ?? sources[0];
if (!rootSource) return null;
const frames = new Map<string, CdpFrame>();
const order: string[] = [];
const targetByRootFrameId = new Map(
sources.map((source) => [source.tree.frame.id, source.target] as const),
);
// The top-level source is authoritative if malformed input reports the same
// frame as both a root target and a child target.
targetByRootFrameId.set(rootSource.tree.frame.id, rootSource.target);
walkFrameTree(rootSource, targetByRootFrameId, frames, order);
for (const source of sources) {
if (source === rootSource) continue;
walkFrameTree(source, targetByRootFrameId, frames, order);
}
return {
rootFrameId: rootSource.tree.frame.id,
frames: order.flatMap((frameId) => {
const frame = frames.get(frameId);
return frame ? [frame] : [];
}),
};
}
@@ -8,7 +8,6 @@ import {
useRef,
useState,
} from "react";
import type { HelpHighlightRect } from "@/lib/help-bridge";
import logoUrl from "../../assets/logo.png";
export interface HelpRequestData {
@@ -20,10 +19,6 @@ export interface HelpRequestData {
displayMode?: "full" | "compact";
/** CSS selectors to scroll to + flash-highlight. */
selectors: string[];
/** Pre-resolved top-viewport rectangles, used for cross-frame targets. */
rects?: HelpHighlightRect[];
/** Re-resolve cross-frame rectangles after scroll, resize, or layout changes. */
refreshRects?: () => Promise<HelpHighlightRect[] | undefined>;
onContinue: (note: string) => void;
onCancel: () => void;
}
@@ -54,7 +49,6 @@ const PANEL_WIDTH = 420;
const FALLBACK_PANEL_H = 180;
const FALLBACK_PANEL_H_COLLAPSED = 44;
const EMPTY_SELECTORS: string[] = [];
const EMPTY_RECTS: HelpHighlightRect[] = [];
function clamp(value: number, min: number, max: number): number {
return Math.max(min, Math.min(max, value));
@@ -185,8 +179,6 @@ export function HelpRequestOverlay({ request }: Props) {
const { t } = useTranslation("extension");
const isCompact = request?.displayMode === "compact";
const effectiveSelectors = isCompact ? EMPTY_SELECTORS : (request?.selectors ?? EMPTY_SELECTORS);
const [liveRects, setLiveRects] = useState<HelpHighlightRect[] | null>(null);
const effectiveRects = isCompact ? EMPTY_RECTS : (liveRects ?? request?.rects ?? EMPTY_RECTS);
const [note, setNote] = useState("");
const [collapsed, setCollapsed] = useState(false);
const [boxes, setBoxes] = useState<Box[]>([]);
@@ -209,7 +201,6 @@ export function HelpRequestOverlay({ request }: Props) {
// Reset the note and collapse state whenever a new request appears.
useEffect(() => {
setNote("");
setLiveRects(null);
setCollapsed(false);
placementRef.current = null;
setDragPos(null);
@@ -218,32 +209,6 @@ export function HelpRequestOverlay({ request }: Props) {
dragStartRef.current = null;
}, [request?.id]);
useEffect(() => {
if (!request || isCompact || !request.refreshRects) return;
let disposed = false;
let resolving = false;
const refresh = async () => {
if (resolving) return;
resolving = true;
try {
const rects = await request.refreshRects?.();
if (!disposed && rects) setLiveRects(rects);
} finally {
resolving = false;
}
};
const onViewportChange = () => void refresh();
window.addEventListener("scroll", onViewportChange, true);
window.addEventListener("resize", onViewportChange);
const interval = window.setInterval(onViewportChange, 500);
return () => {
disposed = true;
window.removeEventListener("scroll", onViewportChange, true);
window.removeEventListener("resize", onViewportChange);
window.clearInterval(interval);
};
}, [request, isCompact]);
// Scroll the first matched target into view once per distinct request id.
useLayoutEffect(() => {
if (!request) return;
@@ -266,7 +231,7 @@ export function HelpRequestOverlay({ request }: Props) {
return;
}
const update = () => {
const nextBoxes = [...measure(effectiveSelectors), ...effectiveRects];
const nextBoxes = measure(effectiveSelectors);
setBoxes((prev) => (boxesEqual(prev, nextBoxes) ? prev : nextBoxes));
};
update();
@@ -278,7 +243,7 @@ export function HelpRequestOverlay({ request }: Props) {
window.removeEventListener("resize", update);
window.clearInterval(interval);
};
}, [request, effectiveSelectors, effectiveRects]);
}, [request, effectiveSelectors]);
const onHeaderPointerDown = useCallback(
(e: ReactPointerEvent<HTMLDivElement>) => {
@@ -340,7 +305,7 @@ export function HelpRequestOverlay({ request }: Props) {
return;
}
if (userMovedRef.current) return;
const measured = boxes.length > 0 ? boxes : [...measure(effectiveSelectors), ...effectiveRects];
const measured = boxes.length > 0 ? boxes : measure(effectiveSelectors);
const anchor = unionBox(measured);
const banner = bannerRef.current;
if (!anchor || !banner) {
@@ -364,7 +329,7 @@ export function HelpRequestOverlay({ request }: Props) {
}
const nextPos = placePanel(anchor, panelW, panelH, placementRef.current.placement);
setPanelPos((prev) => (panelPosEqual(prev, nextPos) ? prev : nextPos));
}, [request, boxes, collapsed, isCompact, effectiveSelectors, effectiveRects]);
}, [request, boxes, collapsed, isCompact, effectiveSelectors]);
if (!request) return null;
+63 -38
View File
@@ -1,4 +1,5 @@
import { useTranslation } from "@browser-skill/i18n/react";
import type { FrameCaptureFailure, FrameCaptureStatus } from "@/transport/types";
import {
type PointerEvent as ReactPointerEvent,
useCallback,
@@ -21,6 +22,8 @@ export interface RecordRequestData {
type Props = {
request: RecordRequestData | null;
captureStatus?: FrameCaptureStatus;
captureFailures?: FrameCaptureFailure[];
};
const VIEWPORT_MARGIN = 16;
@@ -52,7 +55,11 @@ function clampDragPos(
};
}
export function RecordOverlay({ request }: Props) {
export function RecordOverlay({
request,
captureStatus = "complete",
captureFailures = [],
}: Props) {
const { t } = useTranslation("extension");
const [show, setShow] = useState(false);
const [elapsedSeconds, setElapsedSeconds] = useState(0);
@@ -154,6 +161,7 @@ export function RecordOverlay({ request }: Props) {
const introVisible = elapsedSeconds * 1_000 < INTRO_DURATION_MS;
const positioned = dragPos !== null;
const partialCapture = captureStatus === "partial" || captureFailures.length > 0;
const defaultTransform = show
? "translateX(-50%) translateY(0)"
: "translateX(-50%) translateY(8px)";
@@ -208,8 +216,9 @@ export function RecordOverlay({ request }: Props) {
data-slot="record-overlay-status"
style={{
display: "flex",
alignItems: "center",
gap: 8,
flexDirection: "column",
alignItems: "flex-start",
gap: 2,
minWidth: 0,
flex: 1,
borderRadius: 9999,
@@ -218,41 +227,57 @@ export function RecordOverlay({ request }: Props) {
paddingRight: 4,
}}
>
<span
data-slot="record-overlay-indicator"
style={{
width: 9,
height: 9,
borderRadius: "50%",
backgroundColor: "#ef4444",
flexShrink: 0,
animation: "bsk-rec-pulse 1.4s ease-in-out infinite",
}}
/>
<span
data-slot="record-overlay-label"
style={{
display: "block",
flex: "0 1 auto",
fontSize: 16,
fontWeight: 500,
color: "#333",
whiteSpace: "nowrap",
userSelect: "none",
overflow: "hidden",
textOverflow: "ellipsis",
width: introVisible ? "auto" : 48,
maxWidth: introVisible ? 280 : 48,
fontVariantNumeric: introVisible ? undefined : "tabular-nums",
textAlign: introVisible ? "left" : "center",
margin: 0,
padding: 0,
minWidth: 0,
transition: "max-width 180ms ease-out",
}}
>
{introVisible ? t("recordOverlay.recording") : formatElapsed(elapsedSeconds)}
</span>
<div style={{ display: "flex", alignItems: "center", gap: 8, minWidth: 0, width: "100%" }}>
<span
data-slot="record-overlay-indicator"
style={{
width: 9,
height: 9,
borderRadius: "50%",
backgroundColor: "#ef4444",
flexShrink: 0,
animation: "bsk-rec-pulse 1.4s ease-in-out infinite",
}}
/>
<span
data-slot="record-overlay-label"
style={{
display: "block",
flex: "0 1 auto",
fontSize: 16,
fontWeight: 500,
color: "#333",
whiteSpace: "nowrap",
userSelect: "none",
overflow: "hidden",
textOverflow: "ellipsis",
width: introVisible ? "auto" : 48,
maxWidth: introVisible ? 280 : 48,
fontVariantNumeric: introVisible ? undefined : "tabular-nums",
textAlign: introVisible ? "left" : "center",
margin: 0,
padding: 0,
minWidth: 0,
transition: "max-width 180ms ease-out",
}}
>
{introVisible ? t("recordOverlay.recording") : formatElapsed(elapsedSeconds)}
</span>
</div>
{partialCapture ? (
<span
data-slot="record-overlay-partial-warning"
style={{
fontSize: 12,
lineHeight: "16px",
color: "#b45309",
whiteSpace: "normal",
maxWidth: 280,
}}
>
{t("recordOverlay.partialCapture")}
</span>
) : null}
</div>
<button
type="button"
@@ -1,6 +1,6 @@
import { i18n } from "@browser-skill/i18n";
import { I18nextProvider } from "@browser-skill/i18n/react";
import { cleanup, fireEvent, render, screen, waitFor } from "@testing-library/react";
import { cleanup, fireEvent, render, screen } from "@testing-library/react";
import { createElement } from "react";
import { afterEach, describe, expect, it, vi } from "vitest";
import { type HelpRequestData, HelpRequestOverlay } from "../HelpRequestOverlay";
@@ -38,36 +38,6 @@ describe("HelpRequestOverlay", () => {
expect(screen.getByText("Please complete the captcha")).toBeTruthy();
});
it("renders explicit viewport rectangles for cross-frame targets", () => {
const { container } = renderOverlay(
baseRequest({ rects: [{ top: 30, left: 40, width: 120, height: 50 }] }),
);
const highlight = container.querySelector<HTMLElement>("[data-slot='help-highlight']");
expect(highlight?.style.top).toBe("30px");
expect(highlight?.style.left).toBe("40px");
expect(highlight?.style.width).toBe("120px");
expect(highlight?.style.height).toBe("50px");
});
it("re-resolves cross-frame rectangles after the viewport changes", async () => {
const refreshRects = vi.fn(async () => [{ top: 80, left: 90, width: 140, height: 60 }]);
const { container } = renderOverlay(
baseRequest({
rects: [{ top: 30, left: 40, width: 120, height: 50 }],
refreshRects,
}),
);
window.dispatchEvent(new Event("scroll"));
await waitFor(() => {
const highlight = container.querySelector<HTMLElement>("[data-slot='help-highlight']");
expect(highlight?.style.top).toBe("80px");
expect(highlight?.style.left).toBe("90px");
});
expect(refreshRects).toHaveBeenCalledTimes(1);
});
it("keeps the inactive render stable", () => {
const { container, rerender } = renderOverlay(null);
@@ -178,4 +178,30 @@ describe("RecordOverlay", () => {
expect(pill.style.bottom).toBe("auto");
expect(pill.style.transform).toBe("none");
});
it("shows a persistent partial-capture warning when frame capture is incomplete", () => {
const { container } = render(
createElement(
I18nextProvider,
{ i18n },
createElement(RecordOverlay, {
request: { id: "rec-1", onFinish: vi.fn() },
captureStatus: "partial",
captureFailures: [
{
reason: "not_injectable",
frame_id: 2,
url: "chrome://settings",
},
],
}),
),
);
const warning = container.querySelector("[data-slot='record-overlay-partial-warning']");
expect(warning).toBeTruthy();
expect(warning?.textContent).toBe(
i18n.t("recordOverlay.partialCapture", { ns: "extension" }),
);
});
});
@@ -1,64 +0,0 @@
import { describe, expect, it, vi } from "vitest";
import { createHelpRequestData } from "../help-request";
describe("createHelpRequestData", () => {
it("gives live delivery and recovery the same rectangle refresh mechanism", async () => {
const finish = vi.fn();
const query = vi.fn(async () => ({
active: true,
request: {
requestId: "help-1",
prompt: "complete the challenge",
selectors: [],
rects: [{ top: 80, left: 90, width: 140, height: 60 }],
timeoutMs: 1_000,
},
}));
const request = createHelpRequestData(
{
requestId: "help-1",
prompt: "complete the challenge",
selectors: [],
rects: [{ top: 10, left: 20, width: 30, height: 40 }],
timeoutMs: 1_000,
},
{ finish, query },
);
await expect(request.refreshRects?.()).resolves.toEqual([
{ top: 80, left: 90, width: 140, height: 60 },
]);
request.onContinue(" done ");
request.onCancel();
expect(query).toHaveBeenCalledOnce();
expect(finish).toHaveBeenNthCalledWith(1, "help-1", "continued", " done ");
expect(finish).toHaveBeenNthCalledWith(2, "help-1", "cancelled");
});
it("ignores rectangle replies for a replaced request", async () => {
const request = createHelpRequestData(
{
requestId: "help-1",
prompt: "complete the challenge",
selectors: [],
timeoutMs: 1_000,
},
{
finish: vi.fn(),
query: vi.fn(async () => ({
active: true,
request: {
requestId: "help-2",
prompt: "new challenge",
selectors: [],
rects: [{ top: 1, left: 2, width: 3, height: 4 }],
timeoutMs: 1_000,
},
})),
},
);
await expect(request.refreshRects?.()).resolves.toBeUndefined();
});
});
@@ -1,5 +1,5 @@
import { beforeEach, describe, expect, it, vi } from "vitest";
import { RECORD_STOP, type RecordStepPayload } from "@/lib/record-bridge";
import { RECORD_START, RECORD_STOP, type RecordStepPayload } from "@/lib/record-bridge";
import { handleRecordContentMessage, startRecordCapture } from "../record-capture";
vi.stubGlobal("chrome", {
@@ -96,6 +96,65 @@ describe("handleRecordContentMessage stop/cancel", () => {
expect(dispose).not.toHaveBeenCalled();
expect(onStop).not.toHaveBeenCalled();
});
it("keeps a failed STOP retryable and redelivers its recorded step", async () => {
document.body.innerHTML = `<label for="retry">Draft</label><input id="retry" />`;
const sendMessage = vi.mocked(chrome.runtime.sendMessage);
sendMessage.mockReset();
sendMessage.mockRejectedValueOnce(new Error("service worker unavailable"));
sendMessage.mockResolvedValueOnce(undefined);
let activeRequestId: string | null = null;
let capture: ReturnType<typeof startRecordCapture> | null = null;
const onStop = vi.fn();
const dispatch = (
message:
| { type: typeof RECORD_START; requestId: string; startedAtMs?: number }
| { type: typeof RECORD_STOP; requestId: string },
sendResponse?: (response: unknown) => void,
) =>
handleRecordContentMessage(
message,
{
activeRequestId,
capture,
setActiveRequestId: (id) => {
activeRequestId = id;
},
setCapture: (next) => {
capture = next;
},
onStart: vi.fn(),
onStop,
},
sendResponse,
);
dispatch({ type: RECORD_START, requestId: "rec-retry" });
const input = document.querySelector("input")!;
input.dispatchEvent(new FocusEvent("focusin", { bubbles: true }));
input.value = "dirty final value";
input.dispatchEvent(new Event("input", { bubbles: true }));
const firstResponse = vi.fn();
expect(dispatch({ type: RECORD_STOP, requestId: "rec-retry" }, firstResponse)).toBe(true);
await vi.waitFor(() =>
expect(firstResponse).toHaveBeenCalledWith({
ok: false,
error: "failed to deliver one or more recorded steps",
}),
);
expect(activeRequestId).toBe("rec-retry");
expect(onStop).not.toHaveBeenCalled();
const retryResponse = vi.fn();
expect(dispatch({ type: RECORD_STOP, requestId: "rec-retry" }, retryResponse)).toBe(true);
await vi.waitFor(() => expect(retryResponse).toHaveBeenCalledWith({ ok: true }));
expect(sendMessage).toHaveBeenCalledTimes(2);
expect(activeRequestId).toBeNull();
expect(onStop).toHaveBeenCalledTimes(1);
});
});
describe("record-capture semantic", () => {
@@ -228,6 +287,7 @@ describe("record-capture semantic", () => {
expect(steps[0]).toMatchObject({
op: "hover",
target: { role: "button", name: "Open user navigation menu" },
geometry: { tag: "button", rect: { x: 900, y: 8, w: 32, h: 32 } },
});
expect(steps[1]).toMatchObject({
op: "click",
@@ -1554,3 +1614,79 @@ describe("record-capture semantic", () => {
expect(steps.map((s) => s.op)).toEqual(["click"]);
});
});
describe("record-capture child frame", () => {
let steps: RecordStepPayload[];
beforeEach(() => {
steps = [];
document.body.innerHTML = `
<label for="child-q">Child query</label>
<input id="child-q" name="child-q" />
<button type="button" aria-label="Child search">Search</button>
<select aria-label="Child status">
<option value="on">On</option>
<option value="off">Off</option>
</select>
`;
});
it("stamps a document token on start and clears it on dispose", () => {
const capture = startRecordCapture("rec-child-token", (step) => steps.push(step), {
frameMode: "child",
});
const token = document.documentElement.getAttribute("data-bsk-record-document");
expect(token).toBeTruthy();
click(document.querySelector("button")!);
expect(steps[0]?.documentToken).toBe(token);
expect(steps[0]?.frame_url).toBeTruthy();
capture.dispose();
expect(document.documentElement.hasAttribute("data-bsk-record-document")).toBe(false);
});
it("records click, fill, select, and press in child frame mode", () => {
const capture = startRecordCapture("rec-child-ops", (step) => steps.push(step), {
frameMode: "child",
});
const input = document.querySelector("input")!;
const select = document.querySelector("select")!;
input.dispatchEvent(new FocusEvent("focusin", { bubbles: true }));
input.value = "child value";
input.dispatchEvent(new Event("input", { bubbles: true }));
input.dispatchEvent(new FocusEvent("focusout", { bubbles: true }));
select.value = "off";
select.dispatchEvent(new Event("change", { bubbles: true }));
input.dispatchEvent(new FocusEvent("focusin", { bubbles: true }));
input.dispatchEvent(new KeyboardEvent("keydown", { key: "Enter", bubbles: true }));
click(document.querySelector("button")!);
capture.dispose();
expect(steps.map((s) => s.op)).toEqual(
expect.arrayContaining(["fill", "select", "press", "click"]),
);
expect(steps.every((s) => s.documentToken)).toBe(true);
});
it("commits the last fill when dispose runs in child frame mode", () => {
const capture = startRecordCapture("rec-child-fill-stop", (step) => steps.push(step), {
frameMode: "child",
});
const input = document.querySelector("input")!;
input.dispatchEvent(new FocusEvent("focusin", { bubbles: true }));
input.value = "pending child fill";
input.dispatchEvent(new Event("input", { bubbles: true }));
capture.dispose();
expect(steps.some((s) => s.op === "fill" && s.value === "pending child fill")).toBe(true);
});
it("does not emit navigate steps for child frame navigations", () => {
const capture = startRecordCapture("rec-child-nav", (step) => steps.push(step), {
frameMode: "child",
});
history.pushState({}, "", "/child-frame-path");
capture.dispose();
expect(steps.filter((s) => s.op === "navigate")).toHaveLength(0);
});
});
@@ -1,31 +0,0 @@
import type { HelpQueryResponse, HelpRequestMessage } from "@/lib/help-bridge";
import type { HelpRequestData } from "./HelpRequestOverlay";
export interface HelpRequestDataDeps {
finish(requestId: string, outcome: "continued" | "cancelled", note?: string): void;
query(): Promise<HelpQueryResponse | undefined>;
}
/** Build the content-side representation shared by live delivery and recovery. */
export function createHelpRequestData(
message: Omit<HelpRequestMessage, "type">,
deps: HelpRequestDataDeps,
): HelpRequestData {
return {
id: message.requestId,
prompt: message.prompt,
...(message.title ? { title: message.title } : {}),
...(message.displayMode ? { displayMode: message.displayMode } : {}),
selectors: message.selectors,
rects: message.rects,
refreshRects: async () => {
const response = await deps.query();
return response?.active && response.request?.requestId === message.requestId
? response.request.rects
: undefined;
},
onContinue: (note: string) =>
deps.finish(message.requestId, "continued", note.trim() ? note : undefined),
onCancel: () => deps.finish(message.requestId, "cancelled"),
};
}
+141 -35
View File
@@ -1,9 +1,9 @@
import {
type CaptureTargetDescriptor,
describeEventTarget,
describeTarget,
resolveClickableElement,
resolveHoverElement,
type TargetDescriptor,
} from "@/lib/describe-target";
import {
evaluateHoverTrigger,
@@ -21,12 +21,18 @@ import {
RECORD_STEP,
RECORD_STOP,
type RecordCancelMessage,
type RecordFrameMode,
type RecordStartAck,
type RecordStartMessage,
type RecordStepPayload,
type RecordStopAck,
type RecordStopMessage,
} from "@/lib/record-bridge";
import {
clearRecordDocumentToken,
createRecordDocumentToken,
stampRecordDocumentToken,
} from "@/lib/record-document-token";
import { shouldRecordPress } from "@/lib/trace-reducer";
import {
closestHoverSurfaceCandidate,
@@ -39,17 +45,26 @@ import {
const pendingStepSends = new Map<string, Set<Promise<boolean>>>();
const failedStepDeliveries = new Set<string>();
const failedStepPayloads = new Map<string, RecordStepPayload[]>();
const pendingStopFlushes = new Map<string, Promise<RecordStopAck>>();
const knownRecordRequests = new Set<string>();
function sendRecordStep(requestId: string, step: RecordStepPayload): void {
function deliverRecordStep(requestId: string, step: RecordStepPayload): Promise<boolean> {
const payload = { type: RECORD_STEP, requestId, step };
const pending = Promise.resolve(chrome.runtime.sendMessage(payload)).then(
return Promise.resolve(chrome.runtime.sendMessage(payload)).then(
() => true,
() => {
failedStepDeliveries.add(requestId);
const failed = failedStepPayloads.get(requestId) ?? [];
failed.push(step);
failedStepPayloads.set(requestId, failed);
return false;
},
);
}
function sendRecordStep(requestId: string, step: RecordStepPayload): void {
const pending = deliverRecordStep(requestId, step);
const sends = pendingStepSends.get(requestId) ?? new Set<Promise<boolean>>();
sends.add(pending);
pendingStepSends.set(requestId, sends);
@@ -65,14 +80,15 @@ export interface RecordCaptureController {
interface FillSession {
element: FillableElement;
target: TargetDescriptor;
target: CaptureTargetDescriptor;
baselineValue: string;
lastValue: string;
pendingCommit?: "enter" | "suggestion" | "blur";
}
interface HoverCandidate {
element: Element;
target: TargetDescriptor;
target: CaptureTargetDescriptor;
recordedAt: number;
score: number;
eligible: boolean;
@@ -157,6 +173,28 @@ function nearbyFillableFromSearchChrome(target: Element): FillableElement | null
return null;
}
function captureGeometry(el: Element): RecordStepPayload["geometry"] {
const rect = el.getBoundingClientRect();
const style = window.getComputedStyle(el);
return {
rect: { x: rect.x, y: rect.y, w: rect.width, h: rect.height },
scrollX: window.scrollX,
scrollY: window.scrollY,
position: style.position,
tag: el.tagName.toLowerCase(),
};
}
function geometryForEventTarget(
target: EventTarget | null,
): RecordStepPayload["geometry"] | undefined {
if (!(target instanceof Element)) return undefined;
const clickable = target.closest(
'a,button,input,select,textarea,[role="button"],[role="link"],[role="menuitem"],[contenteditable="true"]',
);
return captureGeometry(clickable ?? target);
}
function fillableValue(el: FillableElement): string {
if (el instanceof HTMLInputElement || el instanceof HTMLTextAreaElement) {
return el.value;
@@ -232,7 +270,7 @@ function collectHoverTriggerLabelText(root: Element): string {
return normalizeLabelText(text);
}
function compactHoverTargetName(el: Element, desc: TargetDescriptor): TargetDescriptor {
function compactHoverTargetName(el: Element, desc: CaptureTargetDescriptor): CaptureTargetDescriptor {
if (!desc.name) return desc;
const fullText = normalizeLabelText(el.textContent ?? "");
const compactName = collectHoverTriggerLabelText(el);
@@ -250,7 +288,7 @@ function compactHoverTargetName(el: Element, desc: TargetDescriptor): TargetDesc
return desc;
}
function isWeakHoverTarget(target: TargetDescriptor): boolean {
function isWeakHoverTarget(target: CaptureTargetDescriptor): boolean {
return !target.role && !target.name && target.tag === "div";
}
@@ -261,9 +299,9 @@ function looksLikeAvatarElement(el: Element): boolean {
function normalizeHoverTarget(
el: Element,
desc: TargetDescriptor,
desc: CaptureTargetDescriptor,
decision: HoverTriggerDecision,
): TargetDescriptor {
): CaptureTargetDescriptor {
if (desc.role === "img" && !desc.name && looksLikeAvatarElement(el)) {
return { ...desc, name: "image" };
}
@@ -277,7 +315,7 @@ function normalizeHoverTarget(
return desc;
}
function hoverTriggerSignals(el: Element, desc: TargetDescriptor) {
function hoverTriggerSignals(el: Element, desc: CaptureTargetDescriptor) {
if (!(el instanceof HTMLElement)) return null;
const style = hoverTriggerStyle(el);
return {
@@ -370,12 +408,37 @@ function scheduleInputCompletionCommit(
setTimeout(syncAndCommit, 0);
}
export interface RecordCaptureOptions {
frameMode?: RecordFrameMode;
}
export function startRecordCapture(
_requestId: string,
sendStep: (step: RecordStepPayload) => void,
options: RecordCaptureOptions = {},
): RecordCaptureController {
const frameMode = options.frameMode ?? "top";
const emitNavigate = frameMode === "top";
const documentToken = createRecordDocumentToken();
stampRecordDocumentToken(documentToken);
const capturedAtMs = () => Date.now();
const resolveTopPageUrl = (): string => {
try {
return window.top?.location.href ?? location.href;
} catch {
return location.href;
}
};
const emitStep = (step: RecordStepPayload) => {
sendStep({ page_url: location.href, ...step });
sendStep({
page_url: resolveTopPageUrl(),
frame_url: location.href,
documentToken,
capturedAtMs: capturedAtMs(),
...step,
});
};
const hoverSurfaceStateMap = (states: HoverSurfaceState[]): Map<Element, string> =>
new Map(states.map((state) => [state.element, state.signature]));
@@ -411,14 +474,17 @@ export function startRecordCapture(
emitStep({
op: "fill",
target: session.target,
geometry: captureGeometry(session.element),
value,
commit: session.pendingCommit ?? "blur",
...(isPassword ? { redacted: true } : {}),
});
};
const commitFillSession = () => {
const commitFillSession = (commit: "enter" | "suggestion" | "blur" = "blur") => {
if (!fillSession || composing) return;
const session = fillSession;
session.pendingCommit = commit;
fillSession = null;
emitFill(session);
committedValues.set(session.element, session.lastValue);
@@ -443,6 +509,7 @@ export function startRecordCapture(
};
const emitNavigateIfChanged = (causedByAction?: boolean) => {
if (!emitNavigate) return;
if (location.href === lastUrl) return;
commitFillSession();
lastUrl = location.href;
@@ -624,6 +691,7 @@ export function startRecordCapture(
emitStep({
op: "hover",
target: hover.target,
geometry: captureGeometry(hover.element),
});
emittedHoverElements.add(hover.element);
};
@@ -678,6 +746,7 @@ export function startRecordCapture(
emitStep({
op: "click",
target,
geometry: geometryForEventTarget(eventTarget(event)),
expects_navigation: true,
});
};
@@ -779,7 +848,7 @@ export function startRecordCapture(
scheduleInputCompletionCommit(
sessionElement,
syncFillSessionValue,
commitFillSession,
() => commitFillSession("suggestion"),
(el) => fillSession?.element === el,
);
return;
@@ -839,6 +908,7 @@ export function startRecordCapture(
emitStep({
op: "select",
target: desc,
geometry: captureGeometry(target),
values,
labels,
expects_navigation: true,
@@ -879,7 +949,7 @@ export function startRecordCapture(
};
}
if (fillable) {
commitFillSession();
commitFillSession(event.key === "Enter" ? "enter" : "blur");
}
const desc = describeEventTarget(target);
if (!desc && !event.key) return;
@@ -887,7 +957,7 @@ export function startRecordCapture(
emitStep({
op: "press",
key: event.key,
...(desc ? { target: desc } : {}),
...(desc ? { target: desc, geometry: geometryForEventTarget(target) } : {}),
...(modifiers.length ? { modifiers } : {}),
expects_navigation: event.key === "Enter",
});
@@ -906,9 +976,12 @@ export function startRecordCapture(
const urlObserver = new MutationObserver(() => emitNavigateIfChanged());
urlObserver.observe(document, { subtree: true, childList: true });
const onUrlEvent = () => emitNavigateIfChanged();
// Wrap: passing commitFillSession directly would forward the DOM event as
// the `commit` argument and stamp it onto the recorded fill step.
const onPageHide = () => commitFillSession();
window.addEventListener("hashchange", onUrlEvent);
window.addEventListener("popstate", onUrlEvent);
window.addEventListener("pagehide", commitFillSession);
window.addEventListener("pagehide", onPageHide);
const originalPushState = history.pushState;
const originalReplaceState = history.replaceState;
@@ -924,6 +997,7 @@ export function startRecordCapture(
return {
dispose() {
commitFillSession();
clearRecordDocumentToken();
document.removeEventListener("click", onClick, true);
document.removeEventListener("mouseover", onMouseOver, true);
document.removeEventListener("focusin", onFocusIn, true);
@@ -936,7 +1010,7 @@ export function startRecordCapture(
urlObserver.disconnect();
window.removeEventListener("hashchange", onUrlEvent);
window.removeEventListener("popstate", onUrlEvent);
window.removeEventListener("pagehide", commitFillSession);
window.removeEventListener("pagehide", onPageHide);
history.pushState = originalPushState;
history.replaceState = originalReplaceState;
},
@@ -956,8 +1030,8 @@ export function handleRecordContentMessage(
capture: RecordCaptureController | null;
setActiveRequestId(id: string | null): void;
setCapture(capture: RecordCaptureController | null): void;
onStart(requestId: string, startedAtMs?: number): void;
onStop(): void;
onStart?(requestId: string, startedAtMs?: number): void;
onStop?(): void;
},
sendResponse?: (response: RecordStartAck | RecordStopAck) => void,
): boolean {
@@ -965,15 +1039,23 @@ export function handleRecordContentMessage(
if (!knownRecordRequests.has(message.requestId)) {
knownRecordRequests.add(message.requestId);
failedStepDeliveries.delete(message.requestId);
failedStepPayloads.delete(message.requestId);
}
state.capture?.dispose();
const frameMode = message.frameMode ?? "top";
state.setCapture(
startRecordCapture(message.requestId, (step) => {
sendRecordStep(message.requestId, step);
}),
startRecordCapture(
message.requestId,
(step) => {
sendRecordStep(message.requestId, step);
},
{ frameMode },
),
);
state.setActiveRequestId(message.requestId);
state.onStart(message.requestId, message.startedAtMs);
if (message.showOverlay !== false && frameMode === "top") {
state.onStart?.(message.requestId, message.startedAtMs);
}
sendResponse?.({ ok: true });
return sendResponse !== undefined;
}
@@ -988,31 +1070,55 @@ export function handleRecordContentMessage(
state.capture?.dispose();
state.setCapture(null);
const finishStop = () => {
state.onStop();
state.onStop?.();
state.setActiveRequestId(null);
};
if (isRecordStopMessage(message) && sendResponse) {
const existingFlush = pendingStopFlushes.get(message.requestId);
if (existingFlush) {
void existingFlush.then(sendResponse);
return true;
}
const failedBeforeStop = [...(failedStepPayloads.get(message.requestId) ?? [])];
if (failedBeforeStop.length > 0) {
failedStepPayloads.delete(message.requestId);
failedStepDeliveries.delete(message.requestId);
}
const pending = [...(pendingStepSends.get(message.requestId) ?? [])];
void Promise.all(pending).then((delivered) => {
finishStop();
const succeeded = delivered.every(Boolean) && !failedStepDeliveries.has(message.requestId);
const flush = Promise.all(pending).then(async (delivered): Promise<RecordStopAck> => {
const retried = await Promise.all(
failedBeforeStop.map((step) => deliverRecordStep(message.requestId, step)),
);
const succeeded =
delivered.every(Boolean) &&
retried.every(Boolean) &&
!failedStepDeliveries.has(message.requestId);
if (succeeded) {
failedStepDeliveries.delete(message.requestId);
failedStepPayloads.delete(message.requestId);
knownRecordRequests.delete(message.requestId);
finishStop();
}
return succeeded
? { ok: true }
: {
ok: false,
error: "failed to deliver one or more recorded steps",
};
});
pendingStopFlushes.set(message.requestId, flush);
void flush.then(sendResponse).finally(() => {
if (pendingStopFlushes.get(message.requestId) === flush) {
pendingStopFlushes.delete(message.requestId);
}
sendResponse(
succeeded
? { ok: true }
: {
ok: false,
error: "failed to deliver one or more recorded steps",
},
);
});
return true;
}
if (isRecordCancelMessage(message)) {
failedStepDeliveries.delete(message.requestId);
failedStepPayloads.delete(message.requestId);
pendingStopFlushes.delete(message.requestId);
knownRecordRequests.delete(message.requestId);
}
finishStop();
+23 -1
View File
@@ -33,6 +33,9 @@ import {
defaultBorrowChromeWindows,
requestBorrowConfirmation,
} from "@/tools/borrow-confirmation";
import {
type RecordCaptureStatusMessage,
} from "@/lib/record-bridge";
import { ToolDispatcher } from "@/tools/dispatcher";
import {
attachRecordFinishListener,
@@ -172,8 +175,27 @@ export default defineBackground(() => {
dispatcher.start();
const recordDeps = {
tabsApi: chrome.tabs,
cdp,
sendToTab: (tabId: number, msg: Parameters<typeof chrome.tabs.sendMessage>[1]) =>
chrome.tabs.sendMessage(tabId, msg),
sendToDocument: (
tabId: number,
documentId: string,
msg: Parameters<typeof chrome.tabs.sendMessage>[1],
) => chrome.tabs.sendMessage(tabId, msg, { documentId }),
getAllFrames: async (tabId: number) => {
if (!chrome.webNavigation?.getAllFrames) return [];
const frames = await chrome.webNavigation.getAllFrames({ tabId });
return (frames ?? []).map((frame) => ({
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
parentFrameId: frame.parentFrameId,
}));
},
broadcastCaptureStatus: async (tabId: number, msg: RecordCaptureStatusMessage) => {
await chrome.tabs.sendMessage(tabId, msg, { frameId: 0 });
},
bypassOverlay: async (tabId: number, enabled: boolean) => {
try {
await chrome.tabs.sendMessage(tabId, {
@@ -188,7 +210,7 @@ export default defineBackground(() => {
// Message listeners stay up (cheap; fire only for record message types).
// Tab / webNavigation observation attaches lazily while a recording is
// active — see ensureBrowserObservationListeners in tools/record.ts.
attachRecordStepListener();
attachRecordStepListener(recordDeps);
attachRecordFinishListener(recordDeps);
attachRecordQueryListener(recordDeps);
if (typeof chrome.notifications?.onClicked?.addListener === "function") {
+44 -16
View File
@@ -7,7 +7,6 @@ import { BorrowConfirmationOverlay } from "@/content/BorrowConfirmationOverlay";
import { ControlOverlay } from "@/content/ControlOverlay";
import { createCaptureSuppressController } from "@/content/capture-suppress";
import { HelpRequestOverlay } from "@/content/HelpRequestOverlay";
import { createHelpRequestData } from "@/content/help-request";
import overlayCss from "@/content/overlay.css?inline";
import { OverlayController, shouldShowAgentControlOverlay } from "@/content/overlay-controller";
import { RecordOverlay } from "@/content/RecordOverlay";
@@ -45,12 +44,15 @@ import {
} from "@/lib/overlay-bridge";
import { sendInterrupt } from "@/lib/overlay-interrupt-client";
import {
isRecordCaptureStatusMessage,
RECORD_FINISH,
RECORD_QUERY,
type RecordCaptureStatusMessage,
type RecordQueryResponse,
type RecordStartAck,
type RecordStopAck,
} from "@/lib/record-bridge";
import type { FrameCaptureFailure, FrameCaptureStatus } from "@/transport/types";
import type {
BorrowCancelMessage,
BorrowRequestMessage,
@@ -71,6 +73,8 @@ export default defineContentScript({
const overlays = new OverlayController();
let recordCapture: RecordCaptureController | null = null;
let activeRecordRequestId: string | null = null;
let recordCaptureStatus: FrameCaptureStatus = "complete";
let recordCaptureFailures: FrameCaptureFailure[] = [];
let reactRoot: ReactDOM.Root | null = null;
let overlayHost: HTMLElement | null = null;
let overlayContainer: HTMLElement | null = null;
@@ -189,7 +193,11 @@ export default defineContentScript({
requests: overlayState.borrowRequests,
}),
React.createElement(HelpRequestOverlay, { request: overlayState.activeHelp }),
React.createElement(RecordOverlay, { request: overlayState.activeRecord }),
React.createElement(RecordOverlay, {
request: overlayState.activeRecord,
captureStatus: recordCaptureStatus,
captureFailures: recordCaptureFailures,
}),
React.createElement(ControlOverlay, {
visible: controlOverlayVisible,
interrupting: overlayState.interrupting,
@@ -261,7 +269,8 @@ export default defineContentScript({
| CaptureSuppressMessage
| OverlayAgentOverlayResetMessage
| OverlayAgentStateMessage
| OverlayAutomationBypassMessage,
| OverlayAutomationBypassMessage
| RecordCaptureStatusMessage,
_sender: chrome.runtime.MessageSender,
sendResponse: (response: BorrowResponseMessage | HelpAckMessage | CaptureSuppressAck) => void,
) => {
@@ -269,6 +278,13 @@ export default defineContentScript({
return captureSuppress.handleMessage(message, sendResponse);
}
if (isRecordCaptureStatusMessage(message)) {
recordCaptureStatus = message.status;
recordCaptureFailures = message.failures ?? [];
renderAll();
return false;
}
if (isRecordContentMessage(message)) {
const needsAsync = handleRecordContentMessage(
message,
@@ -296,6 +312,8 @@ export default defineContentScript({
},
onStop: () => {
overlays.clearAgentRecordRequest(activeRecordRequestId ?? undefined);
recordCaptureStatus = "complete";
recordCaptureFailures = [];
renderAll();
},
},
@@ -345,7 +363,16 @@ export default defineContentScript({
if (isHelpRequestMessage(message)) {
const helpMsg = message as HelpRequestMessage;
const previousHelp = mountHelpRequest(helpMsg);
const previousHelp = overlays.setAgentHelpRequest({
id: helpMsg.requestId,
prompt: helpMsg.prompt,
...(helpMsg.title ? { title: helpMsg.title } : {}),
...(helpMsg.displayMode ? { displayMode: helpMsg.displayMode } : {}),
selectors: helpMsg.selectors,
onContinue: (note: string) =>
void sendHelpFinish(helpMsg.requestId, "continued", note.trim() ? note : undefined),
onCancel: () => void sendHelpFinish(helpMsg.requestId, "cancelled"),
});
if (previousHelp && previousHelp.id !== helpMsg.requestId) {
void sendHelpFinish(previousHelp.id, "cancelled");
}
@@ -398,18 +425,17 @@ export default defineContentScript({
});
}
function mountHelpRequest(helpMsg: Omit<HelpRequestMessage, "type">) {
return overlays.setAgentHelpRequest(
createHelpRequestData(helpMsg, {
finish: (requestId, outcome, note) => {
void sendHelpFinish(requestId, outcome, note);
},
query: () =>
chrome.runtime.sendMessage({ type: HELP_QUERY }) as Promise<
HelpQueryResponse | undefined
>,
}),
);
function mountHelpRequest(helpMsg: Omit<HelpRequestMessage, "type">): void {
overlays.setAgentHelpRequest({
id: helpMsg.requestId,
prompt: helpMsg.prompt,
...(helpMsg.title ? { title: helpMsg.title } : {}),
...(helpMsg.displayMode ? { displayMode: helpMsg.displayMode } : {}),
selectors: helpMsg.selectors,
onContinue: (note: string) =>
void sendHelpFinish(helpMsg.requestId, "continued", note.trim() ? note : undefined),
onCancel: () => void sendHelpFinish(helpMsg.requestId, "cancelled"),
});
}
async function queryActiveHelpWithRetry(): Promise<void> {
@@ -442,6 +468,8 @@ export default defineContentScript({
) {
const requestId = recordQuery.requestId;
const startedAtMs = recordQuery.startedAtMs;
recordCaptureStatus = recordQuery.captureStatus ?? "complete";
recordCaptureFailures = recordQuery.captureFailures ?? [];
overlays.setAgentRecordRequest({
id: requestId,
...(typeof startedAtMs === "number" ? { startedAtMs } : {}),
@@ -100,6 +100,24 @@ describe("App", () => {
expect(screen.queryByText("录制你的操作,供 Agent 参考")).toBeNull();
});
it("renders localized copy for a version failure instead of protocol details", () => {
mockUseConnectionState.mockReturnValue({
snapshot: {
...baseSnapshot,
lastError:
"[handshake] daemon rejected handshake: version_too_old — peer protocol 1.0 is below local min_compatible_protocol 1.1",
},
statusState: "disconnected",
setLabel,
setConnectionEnabled,
});
render(<App />);
const error = screen.getByText(/bsk 与 BrowserSkill 扩展版本不兼容/);
expect(error.textContent).not.toContain("min_compatible_protocol");
});
it("shows single-line compact metadata and copies the instance id", async () => {
mockUseConnectionState.mockReturnValue({
snapshot: {
@@ -200,6 +218,13 @@ describe("App", () => {
expect(copied).toContain("--browser 03c3e47f");
expect(copied).toContain('--purpose "发布 wiki 文档"');
expect(copied).not.toMatch(/bsk record start[^\n]*--url/);
expect(copied).toContain("./trace");
expect(copied).toContain("trace.json");
expect(copied).toContain("pages/");
expect(copied).toContain("总结我做了什么");
expect(copied).not.toContain("不要重跑");
expect(copied).not.toContain("pages + steps");
expect(copied).not.toContain("@eN");
// Button label stays static; a transient toast confirms the copy.
expect(copyButton.textContent).toContain("复制录制指令");
await waitFor(() => expect(screen.getByRole("status").textContent).toContain("已复制"));
+5 -2
View File
@@ -83,6 +83,9 @@ export function App() {
const daemonProtocol = snapshot.handshake?.protocol_version ?? "—";
const extensionVersion = snapshot.extensionVersion || "—";
const instanceId = snapshot.instanceId || "—";
const errorMessage = snapshot.lastError?.includes("version_too_old")
? t("popup.versionIncompatibleError")
: snapshot.lastError;
const copyInstanceId = async () => {
if (!snapshot.instanceId || !navigator.clipboard?.writeText) return;
@@ -252,12 +255,12 @@ export function App() {
</div>
</section>
{snapshot.lastError && (
{errorMessage && (
<div
className="rounded-lg border border-destructive/25 bg-destructive/10 px-3 py-2 text-xs leading-snug text-destructive"
data-slot="popup-error"
>
{snapshot.lastError}
{errorMessage}
</div>
)}
@@ -0,0 +1,77 @@
import {
handleRecordContentMessage,
isRecordContentMessage,
type RecordCaptureController,
} from "@/content/record-capture";
import {
RECORD_QUERY,
type RecordQueryResponse,
} from "@/lib/record-bridge";
/** Capture-only content script for child frames (no React overlay). */
export default defineContentScript({
matches: ["<all_urls>"],
runAt: "document_end",
allFrames: true,
matchAboutBlank: true,
matchOriginAsFallback: true,
main(ctx) {
if (window.top === window) return;
let recordCapture: RecordCaptureController | null = null;
let activeRecordRequestId: string | null = null;
const onMessage = (
message: unknown,
_sender: chrome.runtime.MessageSender,
sendResponse: (response: unknown) => void,
) => {
if (isRecordContentMessage(message)) {
const needsAsync = handleRecordContentMessage(
message,
{
activeRequestId: activeRecordRequestId,
capture: recordCapture,
setActiveRequestId: (id) => {
activeRecordRequestId = id;
},
setCapture: (capture) => {
recordCapture = capture;
},
},
sendResponse as never,
);
return needsAsync;
}
return false;
};
async function queryActiveRecord(): Promise<void> {
try {
const recordQuery = (await chrome.runtime.sendMessage({
type: RECORD_QUERY,
})) as RecordQueryResponse | undefined;
if (
recordQuery?.active &&
typeof recordQuery.requestId === "string" &&
activeRecordRequestId === null
) {
// Background rearm sends RECORD_START; nothing to mount locally.
}
} catch (err) {
console.debug("[bsk record-frame] record query failed", err);
}
}
chrome.runtime.onMessage.addListener(onMessage);
void queryActiveRecord();
ctx.onInvalidated(() => {
chrome.runtime.onMessage.removeListener(onMessage);
recordCapture?.dispose();
recordCapture = null;
activeRecordRequestId = null;
});
},
});
@@ -27,19 +27,19 @@ function handshake(
describe("computeConnectedState (protocol-based compat)", () => {
it("returns connected when protocol strings match", () => {
expect(computeConnectedState(handshake("1.0", "1.0"), MIN_COMPATIBLE_PROTOCOL)).toEqual({
expect(computeConnectedState(handshake("1.1", "1.1"), MIN_COMPATIBLE_PROTOCOL)).toEqual({
kind: "connected",
});
});
it("returns version_skew when daemon protocol minor is newer", () => {
expect(computeConnectedState(handshake("1.1", "1.0"))).toEqual({
expect(computeConnectedState(handshake("1.2", "1.1"))).toEqual({
kind: "version_skew",
});
});
it("returns version_skew when daemon protocol string differs but floor is satisfied", () => {
expect(computeConnectedState(handshake("1", "1.0"))).toEqual({
expect(computeConnectedState(handshake("1.1.0", "1.1"))).toEqual({
kind: "version_skew",
});
});
@@ -48,16 +48,19 @@ describe("computeConnectedState (protocol-based compat)", () => {
const result = computeConnectedState(handshake("2.0", "1.0"));
expect(result.kind).toBe("rejected");
if (result.kind === "rejected") {
expect(result.reason).toContain("protocol-major mismatch");
expect(result.reason).toContain("bsk CLI");
expect(result.reason).toContain("BrowserSkill extension");
expect(result.reason).not.toContain("protocol");
}
});
it("rejects when extension is below daemon min_compatible_protocol", () => {
const result = computeConnectedState(handshake("1.0", "1.5"));
const result = computeConnectedState(handshake("1.1", "1.5"));
expect(result.kind).toBe("rejected");
if (result.kind === "rejected") {
expect(result.reason).toContain("min_compatible_protocol");
expect(result.reason).toContain("1.5");
expect(result.reason).toContain("BrowserSkill extension is out of date");
expect(result.reason).toContain("Update and reload the extension");
expect(result.reason).not.toContain("min_compatible_protocol");
}
});
@@ -65,7 +68,7 @@ describe("computeConnectedState (protocol-based compat)", () => {
const result = computeConnectedState({
server: "browser-skill-daemon",
version: "0.1.0",
protocol_version: "1.0",
protocol_version: "1.1",
min_compatible_peer: "0.1.0",
});
expect(result).toEqual({ kind: "connected" });
@@ -75,16 +78,23 @@ describe("computeConnectedState (protocol-based compat)", () => {
const result = computeConnectedState(handshake("1.0", "1.0"), "1.1");
expect(result.kind).toBe("rejected");
if (result.kind === "rejected") {
expect(result.reason).toContain("below extension min_compatible_protocol");
expect(result.reason).toContain("bsk CLI is out of date");
expect(result.reason).toContain("bsk daemon restart");
expect(result.reason).not.toContain("min_compatible_protocol");
}
});
it("rejects a pre-Trace-v3 daemon using the configured extension floor", () => {
const result = computeConnectedState(handshake("1.0", "1.0"));
expect(result.kind).toBe("rejected");
});
it("rejects malformed daemon min_compatible_protocol with a daemon-floor reason", () => {
const result = computeConnectedState(handshake("1.0", "not-a-protocol"));
const result = computeConnectedState(handshake("1.1", "not-a-protocol"));
expect(result.kind).toBe("rejected");
if (result.kind === "rejected") {
expect(result.reason).toContain("daemon min_compatible_protocol");
expect(result.reason).toContain("not-a-protocol");
expect(result.reason).toContain("could not verify version compatibility");
expect(result.reason).not.toContain("min_compatible_protocol");
}
});
@@ -242,11 +252,11 @@ describe("ConnectionController connectionEnabled", () => {
const second = transport.send.mock.calls[1]?.[0] as { id: string };
expect(second.id).not.toBe(first.id);
transport.emitMessage({ id: first.id, result: handshake("1.0", "1.0") });
transport.emitMessage({ id: first.id, result: handshake("1.1", "1.1") });
await Promise.resolve();
expect(controller.snapshot().state).not.toBe("connected");
transport.emitMessage({ id: second.id, result: handshake("1.0", "1.0") });
transport.emitMessage({ id: second.id, result: handshake("1.1", "1.1") });
await vi.waitFor(() => expect(controller.snapshot().state).toBe("connected"));
});
});
@@ -32,7 +32,6 @@ describe("help-bridge", () => {
prompt: "log in",
displayMode: "compact",
selectors: ["#login"],
rects: [{ top: 10, left: 20, width: 100, height: 40 }],
timeoutMs: 1000,
}),
).toBe(true);
@@ -0,0 +1,172 @@
import type { RenderedRef } from "@browser-skill/vom";
import { describe, expect, it } from "vitest";
import type { CapturedNode } from "@/tools/vom/capture";
import { matchTarget } from "../match-target";
function node(backendNodeId: number, overrides: Partial<CapturedNode> = {}): CapturedNode {
return {
backendNodeId,
parentBackendNodeId: null,
tag: "button",
attrs: {},
rect: { x: 10, y: 20, w: 100, h: 30 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
...overrides,
};
}
describe("matchTarget", () => {
const refs: RenderedRef[] = [
{ ref: "e1", backendNodeId: 42, role: "button", name: "发布", line: 3 },
];
it("matches a unique node by viewport coordinates", () => {
const target = matchTarget({
geometry: {
rect: { x: 10, y: 20, w: 100, h: 30 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
},
captured: [node(42)],
refs,
fallback: { tag: "button", name: "发布" },
});
expect(target).toEqual({ ref: "e1", role: "button", name: "发布" });
});
it("matches a non-fixed node after the top frame scrolls", () => {
const target = matchTarget({
geometry: {
rect: { x: 10, y: 20, w: 100, h: 30 },
scrollX: 40,
scrollY: 300,
position: "static",
tag: "button",
},
captured: [
node(42, {
documentRect: { x: 50, y: 320, w: 100, h: 30 },
localRect: { x: 10, y: 20, w: 100, h: 30 },
rect: { x: 10, y: 20, w: 100, h: 30 },
}),
],
refs,
fallback: { tag: "button", name: "发布" },
});
expect(target).toEqual({ ref: "e1", role: "button", name: "发布" });
});
it("matches a static node by document coordinates when scroll changed after observation", () => {
const target = matchTarget({
geometry: {
rect: { x: 10, y: 50, w: 100, h: 30 },
scrollX: 0,
scrollY: 250,
position: "static",
tag: "button",
},
captured: [
node(42, {
documentRect: { x: 10, y: 300, w: 100, h: 30 },
localRect: { x: 10, y: 200, w: 100, h: 30 },
rect: { x: 10, y: 200, w: 100, h: 30 },
}),
],
refs,
fallback: { tag: "button", name: "发布" },
});
expect(target).toEqual({ ref: "e1", role: "button", name: "发布" });
});
it("returns unmatched when multiple candidates match", () => {
const target = matchTarget({
geometry: {
rect: { x: 10, y: 20, w: 100, h: 30 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
},
captured: [node(42), node(43)],
refs,
});
expect(target.unmatched).toBe(true);
});
it("uses viewport coordinates for fixed elements", () => {
const target = matchTarget({
geometry: {
rect: { x: 5, y: 5, w: 80, h: 24 },
scrollX: 100,
scrollY: 200,
position: "fixed",
tag: "button",
},
captured: [node(42, { position: "fixed", rect: { x: 5, y: 5, w: 80, h: 24 } })],
refs,
});
expect(target.ref).toBe("e1");
});
it("isolates geometry matching to the owner iframe backend node", () => {
const sharedRect = { x: 10, y: 20, w: 100, h: 30 };
const iframeRefs: RenderedRef[] = [
{ ref: "e1", backendNodeId: 42, role: "button", name: "Top", line: 1 },
{ ref: "e2", backendNodeId: 99, role: "button", name: "Iframe", line: 2 },
];
const target = matchTarget({
geometry: {
rect: sharedRect,
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
ownerFrameBackendNodeId: 13,
},
captured: [
node(42, { rect: sharedRect, documentRect: sharedRect }),
node(99, {
rect: sharedRect,
documentRect: sharedRect,
ownerFrameBackendNodeId: 13,
}),
],
refs: iframeRefs,
fallback: { tag: "button", name: "Iframe" },
});
expect(target).toEqual({ ref: "e2", role: "button", name: "Iframe" });
});
it("matches iframe static elements using frame scroll and document coordinates", () => {
const iframeRefs: RenderedRef[] = [
{ ref: "e2", backendNodeId: 99, role: "button", name: "Iframe", line: 2 },
];
const target = matchTarget({
geometry: {
rect: { x: 10, y: 50, w: 100, h: 30 },
scrollX: 0,
scrollY: 120,
position: "static",
tag: "button",
ownerFrameBackendNodeId: 13,
},
captured: [
node(99, {
ownerFrameBackendNodeId: 13,
documentRect: { x: 10, y: 170, w: 100, h: 30 },
localRect: { x: 10, y: 50, w: 100, h: 30 },
rect: { x: 10, y: 50, w: 100, h: 30 },
}),
],
refs: iframeRefs,
fallback: { tag: "button", name: "Iframe" },
});
expect(target).toEqual({ ref: "e2", role: "button", name: "Iframe" });
});
});
@@ -0,0 +1,116 @@
import { beforeEach, describe, expect, it } from "vitest";
import type { CapturedNode } from "@/tools/vom/capture";
import type { DraftTraceStep } from "@/transport/types";
import { resetStateIdCounterForTests } from "../record-constants";
import {
applyTargetMatching,
buildTraceV3,
createObservationState,
rememberStepOnPage,
} from "../record-observation";
import { registerObservation } from "../trace-reducer";
const URL = "https://example.com/login";
function capturedInput(): CapturedNode {
return {
backendNodeId: 42,
parentBackendNodeId: null,
tag: "input",
attrs: {},
rect: { x: 20, y: 40, w: 200, h: 30 },
localRect: { x: 20, y: 40, w: 200, h: 30 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
};
}
function finalizedFillBody(value: string, redactValues: boolean): string {
const obs = createObservationState({ redactValues });
const rawVomText = '@vom 1\ntextbox "Password" value="••••••" [ref=e1]';
const stateId = registerObservation(obs.stateRegistry, { url: URL, rawVomText });
obs.lastSettled = {
stateId,
captured: [capturedInput()],
matchNodes: [capturedInput()],
documentTokenOwners: new Map(),
refs: [{ ref: "e1", backendNodeId: 42, role: "textbox", name: "Password", line: 1 }],
url: URL,
vomText: rawVomText,
};
const draft: DraftTraceStep = {
op: "fill",
target: { unmatched: true },
value,
geometry: {
rect: { x: 20, y: 40, w: 200, h: 30 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "input",
},
};
applyTargetMatching(obs, draft, 1);
draft.postStateId = stateId;
rememberStepOnPage(obs, draft, 1);
return (
buildTraceV3({
obs,
steps: [draft],
startedAt: "2026-08-12T00:00:00.000Z",
stoppedBy: "user_finish",
bskVersion: "test",
}).states[0]?.body ?? ""
);
}
describe("record observation annotations", () => {
beforeEach(() => {
resetStateIdCounterForTests();
});
it("omits a fill literal from finalized state bodies when values are redacted", () => {
const secret = "hunter2-private";
const body = finalizedFillBody(secret, true);
expect(body).toContain("step 1: fill");
expect(body).not.toContain(secret);
});
it("keeps fill details in ordinary recording annotations", () => {
const value = "ordinary text";
const body = finalizedFillBody(value, false);
expect(body).toContain(`step 1: fill: ${JSON.stringify(value)}`);
});
});
describe("applyTargetMatching without a settled observation", () => {
it("keeps the capture role/name instead of a bare unmatched target", () => {
const obs = createObservationState();
const draft: DraftTraceStep = {
op: "hover",
target: { unmatched: true },
captureTarget: { tag: "button", role: "button", name: "新建" },
geometry: {
rect: { x: 900, y: 8, w: 60, h: 32 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
},
};
applyTargetMatching(obs, draft, 1);
expect(draft.preStateId).toBeUndefined();
expect(draft.target).toEqual({
role: "button",
name: "新建",
unmatched: true,
});
});
});
@@ -12,12 +12,42 @@ describe("recording-step-buffer", () => {
expect(buffer.steps).toEqual([
{
op: "click",
target: { tag: "button", role: "button", name: "发布" },
target: { unmatched: true },
captureTarget: { tag: "button", role: "button", name: "发布" },
},
]);
expect(buffer.pendingNavigation).toBe(true);
});
it("keeps the hovered element description as a capture fallback", () => {
const buffer = { steps: [], pendingNavigation: false };
appendRecordedPayload(buffer, {
op: "hover",
target: { tag: "button", role: "button", name: "新建" },
geometry: {
rect: { x: 900, y: 8, w: 60, h: 32 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
},
});
expect(buffer.steps).toEqual([
{
op: "hover",
target: { unmatched: true },
captureTarget: { tag: "button", role: "button", name: "新建" },
geometry: {
rect: { x: 900, y: 8, w: 60, h: 32 },
scrollX: 0,
scrollY: 0,
position: "static",
tag: "button",
},
},
]);
});
it("annotates navigated_to on action-caused navigation instead of wait_for_navigation", () => {
const buffer = {
steps: [
@@ -71,10 +101,41 @@ describe("recording-step-buffer", () => {
currentUrl: "https://example.com/a",
pendingNavigation: false,
};
observeRecordedNavigation(buffer, "https://example.com/b", false);
const result = observeRecordedNavigation(buffer, "https://example.com/b", false);
expect(result).toEqual({ kind: "appended", index: 0 });
expect(buffer.steps[0]).toMatchObject({
op: "navigate",
url: "https://example.com/b",
});
});
it("asks the recorder to coalesce redirect hops instead of emitting each one", () => {
const buffer = {
steps: [],
currentUrl: "https://passport.example/login",
pendingNavigation: false,
};
const hop1 = observeRecordedNavigation(
buffer,
"https://passport.example/callback",
false,
"link",
["server_redirect"],
);
expect(hop1).toEqual({
kind: "coalesce_redirect",
url: "https://passport.example/callback",
});
expect(buffer.steps).toEqual([]);
const hop2 = observeRecordedNavigation(buffer, "https://app.example/dashboard", false, "link", [
"client_redirect",
]);
expect(hop2).toEqual({
kind: "coalesce_redirect",
url: "https://app.example/dashboard",
});
expect(buffer.steps).toEqual([]);
expect(buffer.currentUrl).toBe("https://app.example/dashboard");
});
});
@@ -1,6 +1,11 @@
import { describe, expect, it } from "vitest";
import type { DraftTraceStep } from "@/transport/types";
import { reduceTraceSteps, resolveTraceStartUrl, shouldRecordPress } from "../trace-reducer";
import { resetStateIdCounterForTests } from "@/lib/record-constants";
import {
reduceTraceSteps,
registerObservation,
resolveTraceStartUrl,
shouldRecordPress,
} from "../trace-reducer";
describe("shouldRecordPress", () => {
it("keeps Enter and Escape", () => {
@@ -17,86 +22,85 @@ describe("shouldRecordPress", () => {
});
});
describe("reduceTraceSteps", () => {
it("builds steps with pages dictionary and page id references", () => {
const drafts: DraftTraceStep[] = [
{
op: "navigate",
url: "https://example.com/search?q=hello&utm_source=x",
page_url: "https://example.com/search?q=hello&utm_source=x",
},
{
op: "fill",
target: { tag: "input", role: "textbox", name: "搜索", name_attr: "q" },
value: "browser skill",
page_url: "https://example.com/search?q=hello&utm_source=x",
},
{
op: "press",
key: "Enter",
target: { tag: "input", role: "textbox", name: "搜索", name_attr: "q" },
navigated_to: "https://example.com/results/42",
page_url: "https://example.com/search?q=hello&utm_source=x",
},
{
op: "click",
target: { tag: "button", role: "button", name: "发布" },
navigated_to: "https://example.com/p/99",
page_url: "https://example.com/results/42",
},
{
op: "press",
key: "a",
page_url: "https://example.com/p/99",
},
];
const { pages, steps } = reduceTraceSteps(
drafts,
"https://example.com/search?q=hello&utm_source=x",
);
expect(JSON.stringify(steps)).not.toContain("parameters");
expect(JSON.stringify(steps)).not.toContain("intent");
expect(JSON.stringify(steps)).not.toContain("summary");
expect(steps.map((s) => s.op)).toEqual(["navigate", "fill", "press", "click"]);
expect(pages.map((p) => p.url)).toEqual([
"https://example.com/search?q=hello&utm_source=x",
"https://example.com/results/42",
"https://example.com/p/99",
]);
expect(steps[0]).toMatchObject({
id: 1,
op: "navigate",
page: "p1",
to: "https://example.com/search?q=hello&utm_source=x",
describe("reduceTraceSteps v3", () => {
it("builds steps with state/result references and deduped states", () => {
resetStateIdCounterForTests();
const registry = new Map();
const body = "@vom 1\nL1 page";
const s1 = registerObservation(registry, {
url: "https://example.com/search",
rawVomText: body,
});
const s2 = registerObservation(registry, {
url: "https://example.com/results/42",
rawVomText: "@vom 1\nL1 results",
});
const { states, steps } = reduceTraceSteps(
[
{
op: "navigate",
url: "https://example.com/search",
preStateId: s1,
postStateId: s1,
},
{
op: "fill",
target: { ref: "e1", role: "textbox", name: "搜索" },
value: "browser skill",
commit: "enter",
preStateId: s1,
postStateId: s1,
},
{
op: "press",
key: "Enter",
target: { ref: "e1", role: "textbox", name: "搜索" },
preStateId: s1,
postStateId: s2,
navigated_to: "https://example.com/results/42",
},
{
op: "click",
target: { ref: "e2", role: "button", name: "发布" },
preStateId: s2,
postStateId: s2,
},
],
registry,
);
expect(states).toHaveLength(2);
expect(steps.map((s) => s.op)).toEqual(["navigate", "fill", "press", "click"]);
expect(steps[1]).toMatchObject({
op: "fill",
page: "p1",
value: "browser skill",
state: s1,
result: { state: s1 },
commit: "enter",
});
expect(steps[2]).toMatchObject({
op: "press",
key: "Enter",
page: "p1",
effect: { navigated_to: "p2" },
});
expect(steps[3]).toMatchObject({
op: "click",
page: "p2",
effect: { navigated_to: "p3" },
state: s1,
result: { state: s2 },
});
expect(JSON.stringify(steps)).not.toContain("page");
expect(JSON.stringify(steps)).not.toContain("effect");
});
it("collapses consecutive navigations", () => {
const { steps } = reduceTraceSteps([
{ op: "navigate", url: "https://a.example/redirect1" },
{ op: "navigate", url: "https://a.example/final" },
]);
resetStateIdCounterForTests();
const registry = new Map();
const s1 = registerObservation(registry, {
url: "https://a.example/redirect1",
rawVomText: "a",
});
const { steps } = reduceTraceSteps(
[
{ op: "navigate", url: "https://a.example/redirect1", preStateId: s1, postStateId: s1 },
{ op: "navigate", url: "https://a.example/final", preStateId: s1, postStateId: s1 },
],
registry,
);
expect(steps).toHaveLength(1);
expect(steps[0]).toMatchObject({
op: "navigate",
@@ -104,73 +108,124 @@ describe("reduceTraceSteps", () => {
});
});
it("keeps committed empty fill steps", () => {
const { steps } = reduceTraceSteps(
it("keeps the origin state and cause of the hop that started a redirect chain", () => {
resetStateIdCounterForTests();
const registry = new Map();
const start = registerObservation(registry, {
url: "https://example.com/",
rawVomText: "start",
});
const hop = registerObservation(registry, {
url: "https://iwiki.woa.com/",
rawVomText: "hop",
});
const landing = registerObservation(registry, {
url: "https://iwiki.woa.com/dashboard",
rawVomText: "landing",
});
const { steps, stepIdByDraftId } = reduceTraceSteps(
[
{
op: "fill",
target: { tag: "input", role: "textbox", name: "Search query" },
value: "",
page_url: "https://example.com/search",
op: "navigate",
url: "https://iwiki.woa.com/",
preStateId: start,
postStateId: hop,
transitionType: "typed",
},
{
op: "navigate",
url: "https://iwiki.woa.com/dashboard",
preStateId: hop,
postStateId: landing,
transitionType: "link",
},
],
"https://example.com/search",
registry,
);
expect(steps).toEqual([
expect.objectContaining({
op: "fill",
value: "",
}),
]);
expect(steps).toHaveLength(1);
expect(steps[0]).toMatchObject({
op: "navigate",
id: 1,
state: start,
result: { state: landing },
to: "https://iwiki.woa.com/dashboard",
cause: "user_typed",
});
// Both drafts fold into step 1, so anything tagged with either draft id
// must resolve to it.
expect(stepIdByDraftId.get(1)).toBe(1);
expect(stepIdByDraftId.get(2)).toBe(1);
});
it("maps select navigated_to onto effect.navigated_to (page id)", () => {
const { pages, steps } = reduceTraceSteps(
it("maps draft ids to published step ids across dropped drafts", () => {
resetStateIdCounterForTests();
const registry = new Map();
const s1 = registerObservation(registry, { url: "https://a.example/", rawVomText: "a" });
const { steps, stepIdByDraftId } = reduceTraceSteps(
[
{
op: "select",
target: { tag: "select", role: "combobox", name: "分类" },
values: ["tech"],
labels: ["技术"],
navigated_to: "https://example.com/list?cat=tech",
page_url: "https://example.com/list",
},
// Dropped: bare typing press.
{ op: "press", key: "a", preStateId: s1, postStateId: s1 },
{ op: "click", target: { ref: "e2" }, preStateId: s1, postStateId: s1 },
],
"https://example.com/list",
registry,
);
expect(pages.map((p) => p.url)).toEqual([
"https://example.com/list",
"https://example.com/list?cat=tech",
]);
expect(steps[0]).toMatchObject({
op: "select",
page: "p1",
selection: [{ value: "tech", label: "技术" }],
effect: { navigated_to: "p2" },
});
expect(steps.map((step) => step.op)).toEqual(["click"]);
expect(stepIdByDraftId.get(1)).toBeUndefined();
expect(stepIdByDraftId.get(2)).toBe(1);
});
it("keeps committed empty fills", () => {
resetStateIdCounterForTests();
const registry = new Map();
const s1 = registerObservation(registry, { url: "https://a.example/", rawVomText: "a" });
const { steps } = reduceTraceSteps(
[
{ op: "fill", target: { ref: "e1", role: "textbox", name: "Search" }, value: "", preStateId: s1, postStateId: s1 },
{ op: "click", target: { ref: "e2", role: "button", name: "Apply" }, preStateId: s1, postStateId: s1 },
],
registry,
);
expect(steps.map((step) => step.op)).toEqual(["fill", "click"]);
expect(steps[0]).toMatchObject({ op: "fill", value: "" });
});
it("keeps hover steps before menu clicks", () => {
resetStateIdCounterForTests();
const registry = new Map();
const state = registerObservation(registry, {
url: "https://example.com/app",
rawVomText: '@e1 button "Account"\n@e2 link "Profile"',
});
const { steps } = reduceTraceSteps(
[
{
op: "hover",
target: { tag: "span", role: "button", name: "Account" },
target: { ref: "e1", role: "button", name: "Account" },
page_url: "https://example.com/app",
preStateId: state,
postStateId: state,
},
{
op: "click",
target: { tag: "a", role: "link", name: "Profile" },
target: { ref: "e2", role: "link", name: "Profile" },
page_url: "https://example.com/app",
preStateId: state,
postStateId: state,
},
],
"https://example.com/app",
registry,
);
expect(steps.map((s) => s.op)).toEqual(["hover", "click"]);
expect(steps[0]).toMatchObject({
op: "hover",
page: "p1",
state,
result: { state },
target: { name: "Account" },
});
});
@@ -178,7 +233,7 @@ describe("reduceTraceSteps", () => {
it("resolveTraceStartUrl prefers explicit start URL", () => {
expect(
resolveTraceStartUrl(
[{ op: "navigate", url: "https://example.com/other" }],
[{ op: "navigate", url: "https://example.com/other", preStateId: "s1", postStateId: "s1" }],
"https://example.com/start",
),
).toBe("https://example.com/start");
@@ -313,6 +313,15 @@ export type HandshakeVerdict =
| { kind: "version_skew" }
| { kind: "rejected"; reason: string };
const UPDATE_BOTH_VERSION_MESSAGE =
"The installed bsk CLI and BrowserSkill extension are incompatible. Update both, reload the extension, then run `bsk daemon restart`.";
const UPDATE_BSK_VERSION_MESSAGE =
"The installed bsk CLI is out of date for this BrowserSkill extension. Update bsk, then run `bsk daemon restart`.";
const UPDATE_EXTENSION_VERSION_MESSAGE =
"This BrowserSkill extension is out of date for the installed bsk CLI. Update and reload the extension, then try again.";
const INVALID_VERSION_MESSAGE =
"BrowserSkill could not verify version compatibility. Update bsk and the BrowserSkill extension, reload the extension, then run `bsk daemon restart`.";
/**
* Symmetric extension-side **protocol** compat check.
*
@@ -339,19 +348,19 @@ export function computeConnectedState(
if (daemonMajor === null || ourMajor === null || daemonMajor !== ourMajor) {
return {
kind: "rejected",
reason: `protocol-major mismatch (daemon protocol v${handshake.protocol_version}, extension protocol v${PROTOCOL_VERSION})`,
reason: UPDATE_BOTH_VERSION_MESSAGE,
};
}
if (compareProtocol(handshake.protocol_version, localMinCompatibleProtocol) === null) {
return {
kind: "rejected",
reason: `daemon protocol v${handshake.protocol_version} is unparseable`,
reason: INVALID_VERSION_MESSAGE,
};
}
if (compareProtocol(handshake.protocol_version, localMinCompatibleProtocol)! < 0) {
return {
kind: "rejected",
reason: `daemon protocol v${handshake.protocol_version} below extension min_compatible_protocol ${localMinCompatibleProtocol}`,
reason: UPDATE_BSK_VERSION_MESSAGE,
};
}
const peerFloor = handshake.min_compatible_protocol;
@@ -359,13 +368,13 @@ export function computeConnectedState(
if (compareProtocol(PROTOCOL_VERSION, peerFloor) === null) {
return {
kind: "rejected",
reason: `daemon min_compatible_protocol ${peerFloor} is unparseable`,
reason: INVALID_VERSION_MESSAGE,
};
}
if (compareProtocol(PROTOCOL_VERSION, peerFloor)! < 0) {
return {
kind: "rejected",
reason: `extension protocol v${PROTOCOL_VERSION} below daemon min_compatible_protocol ${peerFloor}`,
reason: UPDATE_EXTENSION_VERSION_MESSAGE,
};
}
}
+5 -4
View File
@@ -6,7 +6,8 @@
* `{ "tag": "div" }` / “点击div” is useless and must not be recorded.
*/
export interface TargetDescriptor {
/** Content-script capture descriptor before VOM geometric matching. */
export interface CaptureTargetDescriptor {
role?: string;
name?: string;
tag: string;
@@ -247,7 +248,7 @@ export function resolveHoverElement(target: Element): Element | null {
* (visible name), or at least a form `name_attr` for checkbox/radio.
* Recording “点击div” with no name fails this bar.
*/
export function isMeaningfulClickTarget(target: TargetDescriptor): boolean {
export function isMeaningfulClickTarget(target: CaptureTargetDescriptor): boolean {
const name = target.name?.trim();
if (name && isActionableLabel(name)) return true;
if (
@@ -323,7 +324,7 @@ function nearbyLabelText(el: Element): string | undefined {
return undefined;
}
export function describeTarget(el: Element): TargetDescriptor {
export function describeTarget(el: Element): CaptureTargetDescriptor {
const tag = el.tagName.toLowerCase();
const role = inferRole(el);
const name = accessibleName(el);
@@ -350,7 +351,7 @@ export function describeTarget(el: Element): TargetDescriptor {
};
}
export function describeEventTarget(target: EventTarget | null): TargetDescriptor | null {
export function describeEventTarget(target: EventTarget | null): CaptureTargetDescriptor | null {
if (!(target instanceof Element)) return null;
const clickable = resolveClickableElement(target);
if (!clickable) return null;
@@ -0,0 +1,62 @@
import type { Step } from "@/transport/types";
import { OBSERVATION_FILE_VERSION } from "./record-constants";
export interface ObservationAnnotation {
stepId: number;
op: Step["op"];
line: number;
stateId: string;
detail?: string;
}
function formatAnnotation({ stepId, op, detail }: ObservationAnnotation): string {
const suffix = detail ? `: ${detail}` : "";
return ` ⟵ step ${stepId}: ${op}${suffix}`;
}
/** Serialize a page observation file (front matter + VOM body + step annotations). */
export function formatObservationFile(input: {
stateId: string;
url: string;
title?: string;
stepsHere: number[];
body: string;
annotations?: ObservationAnnotation[];
}): string {
const lines: string[] = [
`# bsk-observation ${OBSERVATION_FILE_VERSION}`,
`state: ${input.stateId}`,
`url: ${input.url}`,
];
if (input.title) lines.push(`title: ${input.title}`);
if (input.stepsHere.length > 0) {
lines.push(`steps_here: [${input.stepsHere.join(", ")}]`);
}
lines.push("---");
const bodyLines = input.body.split("\n");
const annotationMap = new Map<number, ObservationAnnotation[]>();
for (const ann of input.annotations ?? []) {
const bucket = annotationMap.get(ann.line) ?? [];
bucket.push(ann);
annotationMap.set(ann.line, bucket);
}
for (let i = 0; i < bodyLines.length; i += 1) {
let line = bodyLines[i] ?? "";
const anns = annotationMap.get(i);
if (anns) {
for (const ann of anns) {
line += formatAnnotation(ann);
}
}
lines.push(line);
}
return `${lines.join("\n")}\n`;
}
/** Hash input: VOM body **before** annotations are inserted. */
export function observationBodyForHash(vomText: string): string {
return vomText;
}
-19
View File
@@ -16,13 +16,6 @@ export const HELP_ACK = "bsk-help-ack";
export const HELP_QUERY = "bsk-help-query";
export const HELP_FINISH = "bsk-help-finish";
export interface HelpHighlightRect {
top: number;
left: number;
width: number;
height: number;
}
export interface HelpRequestMessage {
type: typeof HELP_REQUEST;
requestId: string;
@@ -33,8 +26,6 @@ export interface HelpRequestMessage {
displayMode?: "full" | "compact";
/** CSS selectors to scroll to + flash-highlight (may be empty). */
selectors: string[];
/** Top-viewport rectangles for targets that cannot be represented by a root selector. */
rects?: HelpHighlightRect[];
timeoutMs: number;
}
@@ -83,16 +74,6 @@ export function isHelpRequestMessage(msg: unknown): msg is HelpRequestMessage {
(m.displayMode === undefined || m.displayMode === "full" || m.displayMode === "compact") &&
Array.isArray(m.selectors) &&
m.selectors.every((selector) => typeof selector === "string") &&
(m.rects === undefined ||
(Array.isArray(m.rects) &&
m.rects.every(
(rect) =>
typeof rect === "object" &&
rect !== null &&
["top", "left", "width", "height"].every((key) =>
Number.isFinite((rect as Record<string, unknown>)[key]),
),
))) &&
typeof m.timeoutMs === "number"
);
}
+102
View File
@@ -0,0 +1,102 @@
import type { RenderedRef } from "@browser-skill/vom";
import type { CapturedNode } from "@/tools/vom/capture";
import type { CaptureGeometry, TargetDescriptor } from "@/transport/types";
import type { CaptureTargetDescriptor } from "./describe-target";
import { GEOM_MATCH_TOLERANCE_PX } from "./record-constants";
export interface MatchTargetInput {
geometry: CaptureGeometry;
captured: CapturedNode[];
refs: RenderedRef[];
fallback?: CaptureTargetDescriptor;
}
function withinTolerance(a: number, b: number): boolean {
return Math.abs(a - b) <= GEOM_MATCH_TOLERANCE_PX;
}
function rectsMatch(
a: { x: number; y: number; w: number; h: number },
b: { x: number; y: number; w: number; h: number },
): boolean {
return (
withinTolerance(a.x, b.x) &&
withinTolerance(a.y, b.y) &&
withinTolerance(a.w, b.w) &&
withinTolerance(a.h, b.h)
);
}
function nodeViewportRect(
node: CapturedNode,
): { x: number; y: number; w: number; h: number } | null {
return node.localRect ?? node.rect;
}
function matchingRects(
geometry: CaptureGeometry,
node: CapturedNode,
): {
target: { x: number; y: number; w: number; h: number };
node: { x: number; y: number; w: number; h: number } | null;
} {
const viewportPosition = geometry.position === "fixed" || geometry.position === "sticky";
if (viewportPosition || !node.documentRect) {
return { target: geometry.rect, node: nodeViewportRect(node) };
}
return {
target: {
x: geometry.rect.x + geometry.scrollX,
y: geometry.rect.y + geometry.scrollY,
w: geometry.rect.w,
h: geometry.rect.h,
},
node: node.documentRect,
};
}
function tagMatches(geometryTag: string, nodeTag: string): boolean {
return geometryTag.toLowerCase() === nodeTag.toLowerCase();
}
/** Best description available when the element cannot be located in the VOM. */
export function fallbackDescriptor(fallback?: CaptureTargetDescriptor): TargetDescriptor {
if (!fallback) {
return { unmatched: true };
}
return {
...(fallback.role ? { role: fallback.role } : {}),
...(fallback.name ? { name: fallback.name } : {}),
unmatched: true,
};
}
/** Locate the interacted element in the last settled observation by geometry. */
export function matchTarget(input: MatchTargetInput): TargetDescriptor {
const ownerFrame = input.geometry.ownerFrameBackendNodeId ?? null;
const candidates = input.captured.filter((node) => {
if (ownerFrame !== (node.ownerFrameBackendNodeId ?? null)) return false;
if (!tagMatches(input.geometry.tag, node.tag)) return false;
const rects = matchingRects(input.geometry, node);
if (!rects.node) return false;
return rectsMatch(rects.target, rects.node);
});
if (candidates.length !== 1) {
return fallbackDescriptor(input.fallback);
}
const backendNodeId = candidates[0]!.backendNodeId;
const refEntry = input.refs.find((r) => r.backendNodeId === backendNodeId);
if (!refEntry) {
return fallbackDescriptor(input.fallback);
}
return {
ref: refEntry.ref,
...(refEntry.role ? { role: refEntry.role } : {}),
...(refEntry.name ? { name: refEntry.name } : {}),
...(refEntry.ctx ? { ctx: refEntry.ctx } : {}),
};
}
+92
View File
@@ -0,0 +1,92 @@
// Deciding *when* a page is worth observing.
//
// A fixed delay cannot serve both a modal that toggles in one frame and a
// route change that renders for a second. Instead of guessing, ask the page:
// a document that has stopped mutating and is no longer loading is done
// reacting to whatever the user just did.
import type { CdpRunner } from "@/tools/shared";
import { SETTLE_MAX_MS, SETTLE_MIN_MS, SETTLE_POLL_MS, SETTLE_QUIET_MS } from "./record-constants";
/**
* Installed once per document. Records only the timestamp of the last DOM
* change so each poll stays O(1) no matter how large the page is.
*/
const QUIET_PROBE = `(() => {
const scope = window;
let probe = scope.__bskRecordQuiet;
if (!probe) {
probe = { changedAt: Date.now() };
const observer = new MutationObserver(() => {
probe.changedAt = Date.now();
});
observer.observe(document.documentElement, {
subtree: true,
childList: true,
attributes: true,
characterData: true,
});
scope.__bskRecordQuiet = probe;
}
return { idleMs: Date.now() - probe.changedAt, readyState: document.readyState };
})()`;
export type SettleOutcome =
/** The page stopped changing on its own. */
| "quiet"
/** Still changing when the budget ran out; observe it as it is. */
| "timeout"
/** A newer action took over; this observation is no longer wanted. */
| "cancelled";
interface QuietProbe {
idleMs: number;
readyState: string;
}
function sleep(ms: number): Promise<void> {
return new Promise((resolve) => setTimeout(resolve, ms));
}
async function readQuietProbe(cdp: CdpRunner, tabId: number): Promise<QuietProbe | null> {
try {
const reply = await cdp.send<{ result?: { value?: unknown } }>(tabId, "Runtime.evaluate", {
expression: QUIET_PROBE,
returnByValue: true,
});
const value = reply.result?.value;
if (!value || typeof value !== "object") return null;
const { idleMs, readyState } = value as { idleMs?: unknown; readyState?: unknown };
if (typeof idleMs !== "number" || typeof readyState !== "string") return null;
return { idleMs, readyState };
} catch {
// An unreadable page is a page mid-swap; that is a reason to keep waiting,
// not a reason to give up.
return null;
}
}
/** Wait until the page has finished reacting, or until the budget runs out. */
export async function waitForPageSettled(
cdp: CdpRunner,
tabId: number,
options: { cancelled?: () => boolean } = {},
): Promise<SettleOutcome> {
const startedAt = Date.now();
const floor = startedAt + SETTLE_MIN_MS;
const deadline = startedAt + SETTLE_MAX_MS;
for (;;) {
if (options.cancelled?.()) return "cancelled";
await sleep(SETTLE_POLL_MS);
if (options.cancelled?.()) return "cancelled";
const probe = await readQuietProbe(cdp, tabId);
const now = Date.now();
if (now < floor) continue;
if (now >= deadline) return "timeout";
if (!probe) continue;
if (probe.readyState === "loading") continue;
if (probe.idleMs >= SETTLE_QUIET_MS) return "quiet";
}
}
+49 -4
View File
@@ -3,7 +3,8 @@
* service worker and a tab's content script.
*/
import type { TargetDescriptor } from "./describe-target";
import type { CaptureTargetDescriptor } from "./describe-target";
import type { FrameCaptureFailure, FrameCaptureStatus } from "@/transport/types";
export const RECORD_START = "bsk-record-start";
export const RECORD_STEP = "bsk-record-step";
@@ -11,6 +12,7 @@ export const RECORD_STOP = "bsk-record-stop";
export const RECORD_CANCEL = "bsk-record-cancel";
export const RECORD_FINISH = "bsk-record-finish";
export const RECORD_QUERY = "bsk-record-query";
export const RECORD_CAPTURE_STATUS = "bsk-record-capture-status";
export interface RecordStartAck {
ok: true;
@@ -27,8 +29,12 @@ export interface RecordQueryResponse {
requestId?: string;
/** Epoch ms when the recording began; see `RecordStartMessage.startedAtMs`. */
startedAtMs?: number;
captureStatus?: FrameCaptureStatus;
captureFailures?: FrameCaptureFailure[];
}
export type RecordFrameMode = "top" | "child";
export interface RecordStartMessage {
type: typeof RECORD_START;
requestId: string;
@@ -38,11 +44,15 @@ export interface RecordStartMessage {
* content-script remounts instead of restarting per page.
*/
startedAtMs?: number;
/** When true, the top-frame overlay should mount. Child frames omit this. */
showOverlay?: boolean;
/** Whether this document runs top-frame or child-frame capture semantics. */
frameMode?: RecordFrameMode;
}
export interface RecordStepPayload {
op: "click" | "hover" | "fill" | "press" | "select" | "navigate";
target?: TargetDescriptor;
op: "click" | "hover" | "fill" | "press" | "select" | "navigate" | "scroll";
target?: CaptureTargetDescriptor;
value?: string;
key?: string;
modifiers?: Array<"alt" | "ctrl" | "meta" | "shift">;
@@ -50,12 +60,31 @@ export interface RecordStepPayload {
labels?: string[];
url?: string;
redacted?: boolean;
/** Page URL when the step was captured. */
commit?: "enter" | "suggestion" | "blur";
/** Top-level tab URL when the step was captured. */
page_url?: string;
/** URL of the frame document where the interaction occurred. */
frame_url?: string;
/** Ephemeral token stamped on the active document root; not persisted in trace steps. */
documentToken?: string;
/** Epoch ms when the step was captured in the content script. */
capturedAtMs?: number;
/** Document geometry at capture time for background geometric matching. */
geometry?: {
rect: { x: number; y: number; w: number; h: number };
scrollX: number;
scrollY: number;
position: string;
tag: string;
ownerFrameBackendNodeId?: number | null;
};
/** Capture-only hint; never persisted unless converted to navigated_to. */
expects_navigation?: boolean;
/** Whether an observed URL change was synchronously caused by the action. */
navigation_caused_by_action?: boolean;
/** Raw webNavigation transition metadata for cause mapping. */
transitionType?: string;
transitionQualifiers?: string[];
}
export interface RecordStepMessage {
@@ -79,6 +108,12 @@ export interface RecordFinishMessage {
requestId: string;
}
export interface RecordCaptureStatusMessage {
type: typeof RECORD_CAPTURE_STATUS;
status: FrameCaptureStatus;
failures?: FrameCaptureFailure[];
}
export function isRecordStartMessage(msg: unknown): msg is RecordStartMessage {
if (typeof msg !== "object" || msg === null) return false;
const m = msg as Record<string, unknown>;
@@ -109,6 +144,16 @@ export function isRecordQueryMessage(msg: unknown): msg is RecordQueryMessage {
return m.type === RECORD_QUERY;
}
export function isRecordCaptureStatusMessage(msg: unknown): msg is RecordCaptureStatusMessage {
if (typeof msg !== "object" || msg === null) return false;
const m = msg as Record<string, unknown>;
return (
m.type === RECORD_CAPTURE_STATUS &&
(m.status === "complete" || m.status === "partial") &&
(m.failures === undefined || Array.isArray(m.failures))
);
}
export function isRecordStepMessage(msg: unknown): msg is RecordStepMessage {
if (typeof msg !== "object" || msg === null) return false;
const m = msg as Record<string, unknown>;
@@ -0,0 +1,97 @@
import type { FillCommit, NavigationCause } from "@/transport/types";
/**
* Floor before a post-action observation may be taken. A page usually has not
* started reacting in the first frames after an action, and capturing then
* would record the page as it was, not as the action left it.
*/
export const SETTLE_MIN_MS = 150;
/** How long the DOM must stop changing before a page counts as settled. */
export const SETTLE_QUIET_MS = 250;
/** Upper bound on waiting for a page to settle; animations never stop. */
export const SETTLE_MAX_MS = 2_000;
/** How often to ask the page whether it has gone quiet. */
export const SETTLE_POLL_MS = 60;
/** Minimum interval between consecutive observations on the same tab. */
export const OBSERVATION_MIN_INTERVAL_MS = 200;
/** Delay before retrying a capture that lost its execution context. */
export const CAPTURE_RETRY_DELAY_MS = 250;
/** Document-coordinate matching tolerance in CSS pixels. */
export const GEOM_MATCH_TOLERANCE_PX = 2;
/** Default max tokens per page observation when CLI omits `--max-page-tokens`. */
export const DEFAULT_MAX_PAGE_TOKENS = 3000;
/** VOM observation file format version (front matter header). */
export const OBSERVATION_FILE_VERSION = 1;
/** FNV-1a 64-bit offset basis. */
const FNV_OFFSET = 0xcbf29ce484222325n;
/** FNV-1a 64-bit prime. */
const FNV_PRIME = 0x100000001b3n;
/** Content-hash for state deduplication (sync, non-cryptographic). */
export function fnv1a64(text: string): string {
let hash = FNV_OFFSET;
for (let i = 0; i < text.length; i += 1) {
hash ^= BigInt(text.charCodeAt(i));
hash = (hash * FNV_PRIME) & 0xffffffffffffffffn;
}
return hash.toString(16).padStart(16, "0");
}
let nextStateSerial = 0;
/** Monotonic state id generator (`s1`, `s2`, …). */
export function nextStateId(): string {
nextStateSerial += 1;
return `s${nextStateSerial}`;
}
/** Test seam: reset the state id counter. */
export function resetStateIdCounterForTests(): void {
nextStateSerial = 0;
}
const REDIRECT_QUALIFIERS = new Set(["client_redirect", "server_redirect"]);
const TRANSITION_TO_CAUSE: Record<string, NavigationCause> = {
typed: "user_typed",
generated: "user_typed",
keyword: "user_typed",
keyword_generated: "user_typed",
link: "link",
form_submit: "form_submit",
reload: "reload",
auto_bookmark: "browser",
start_page: "browser",
};
export interface NavigationTransitionMeta {
transitionType?: string;
transitionQualifiers?: string[];
navigationActionPending?: boolean;
}
/** Map webNavigation metadata to protocol `NavigationCause`. Returns null for redirects. */
export function mapNavigationCause(meta: NavigationTransitionMeta): NavigationCause | null {
const qualifiers = meta.transitionQualifiers ?? [];
if (qualifiers.includes("forward_back")) return "history";
if (qualifiers.some((q) => REDIRECT_QUALIFIERS.has(q))) return null;
if (qualifiers.includes("from_address_bar")) return "user_typed";
const type = meta.transitionType ?? "";
const mapped = TRANSITION_TO_CAUSE[type];
if (mapped) return mapped;
if (!type && meta.navigationActionPending === false) return "script";
return "browser";
}
export const DEFAULT_FILL_COMMIT: FillCommit = "blur";
@@ -0,0 +1,24 @@
/** Attribute written on `<html>` while a document is actively recording. */
export const RECORD_DOCUMENT_TOKEN_ATTR = "data-bsk-record-document";
export function createRecordDocumentToken(): string {
if (typeof crypto !== "undefined" && typeof crypto.randomUUID === "function") {
return crypto.randomUUID();
}
return `doc-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`;
}
export function stampRecordDocumentToken(token: string): void {
document.documentElement.setAttribute(RECORD_DOCUMENT_TOKEN_ATTR, token);
}
export function clearRecordDocumentToken(): void {
document.documentElement.removeAttribute(RECORD_DOCUMENT_TOKEN_ATTR);
}
export function readRecordDocumentTokenFromAttrs(
attrs: Record<string, string> | undefined,
): string | undefined {
const token = attrs?.[RECORD_DOCUMENT_TOKEN_ATTR];
return token?.trim() ? token : undefined;
}
@@ -0,0 +1,250 @@
import {
RECORD_CANCEL,
RECORD_START,
RECORD_STOP,
type RecordCancelMessage,
type RecordStartAck,
type RecordStartMessage,
type RecordStopAck,
type RecordStopMessage,
} from "@/lib/record-bridge";
import type { FrameCaptureFailure, FrameCaptureFailureReason } from "@/transport/types";
export interface WebNavigationFrameInfo {
frameId: number;
documentId?: string;
url?: string;
parentFrameId: number;
}
export interface ArmedDocument {
tabId: number;
frameId: number;
documentId: string;
url?: string;
parentFrameId: number;
}
export interface RecordFrameSendDeps {
sendToDocument(
tabId: number,
documentId: string,
msg: RecordStartMessage | RecordStopMessage | RecordCancelMessage,
): Promise<unknown>;
getAllFrames(tabId: number): Promise<WebNavigationFrameInfo[]>;
}
const RECORD_START_RETRIES = 3;
const RECORD_START_RETRY_DELAY_MS = 500;
function sleep(ms: number): Promise<void> {
return new Promise((resolve) => setTimeout(resolve, ms));
}
export function isFrameUrlRestricted(url: string | undefined): boolean {
if (!url) return false;
const lower = url.toLowerCase();
return (
lower.startsWith("chrome://") ||
lower.startsWith("chrome-extension://") ||
lower.startsWith("edge://") ||
lower.startsWith("devtools://") ||
lower.startsWith("devtools:") ||
lower.startsWith("https://chrome.google.com/webstore")
);
}
function isRecordStartAck(response: unknown): response is RecordStartAck {
return (
typeof response === "object" &&
response !== null &&
"ok" in response &&
(response as RecordStartAck).ok === true
);
}
function isRecordStopAckOk(response: unknown): boolean {
return (
typeof response === "object" &&
response !== null &&
"ok" in response &&
(response as RecordStopAck).ok === true
);
}
export function makeFrameCaptureFailure(
reason: FrameCaptureFailureReason,
input: { frameId: number; documentId?: string; url?: string; detail?: string },
): FrameCaptureFailure {
return {
reason,
frame_id: input.frameId,
...(input.documentId ? { document_id: input.documentId } : {}),
...(input.url ? { url: input.url } : {}),
...(input.detail ? { detail: input.detail } : {}),
};
}
export async function listInjectableFrames(
tabId: number,
deps: RecordFrameSendDeps,
): Promise<{ injectable: WebNavigationFrameInfo[]; failures: FrameCaptureFailure[] }> {
const failures: FrameCaptureFailure[] = [];
let frames: WebNavigationFrameInfo[] = [];
try {
frames = await deps.getAllFrames(tabId);
} catch (err) {
failures.push(
makeFrameCaptureFailure("arm_failed", {
frameId: 0,
detail: err instanceof Error ? err.message : "webNavigation.getAllFrames failed",
}),
);
return { injectable: [], failures };
}
const injectable = frames.filter((frame) => {
if (!frame.documentId) {
failures.push(
makeFrameCaptureFailure("not_injectable", {
frameId: frame.frameId,
url: frame.url,
detail: "missing documentId",
}),
);
return false;
}
if (isFrameUrlRestricted(frame.url)) {
failures.push(
makeFrameCaptureFailure("not_injectable", {
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
}),
);
return false;
}
return true;
});
return { injectable, failures };
}
export async function sendRecordStartWithAckToDocument(
tabId: number,
documentId: string,
msg: RecordStartMessage,
deps: RecordFrameSendDeps,
): Promise<void> {
let lastError: unknown;
for (let attempt = 0; attempt < RECORD_START_RETRIES; attempt += 1) {
try {
const response = await deps.sendToDocument(tabId, documentId, msg);
if (isRecordStartAck(response)) return;
lastError = new Error("content script did not ack RECORD_START");
} catch (err) {
lastError = err;
}
if (attempt + 1 < RECORD_START_RETRIES) {
await sleep(RECORD_START_RETRY_DELAY_MS);
}
}
throw lastError ?? new Error("failed to start recording in content script");
}
export async function armDocumentCapture(
tabId: number,
frame: WebNavigationFrameInfo,
startMsg: RecordStartMessage,
deps: RecordFrameSendDeps,
): Promise<{ armed?: ArmedDocument; failure?: FrameCaptureFailure }> {
if (!frame.documentId) {
return {
failure: makeFrameCaptureFailure("not_injectable", {
frameId: frame.frameId,
url: frame.url,
detail: "missing documentId",
}),
};
}
if (isFrameUrlRestricted(frame.url)) {
return {
failure: makeFrameCaptureFailure("not_injectable", {
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
}),
};
}
try {
await sendRecordStartWithAckToDocument(tabId, frame.documentId, startMsg, deps);
return {
armed: {
tabId,
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
parentFrameId: frame.parentFrameId,
},
};
} catch (err) {
return {
failure: makeFrameCaptureFailure("arm_failed", {
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
detail: err instanceof Error ? err.message : "RECORD_START failed",
}),
};
}
}
export async function armAllDocumentsForTab(
tabId: number,
startMsg: RecordStartMessage,
deps: RecordFrameSendDeps,
): Promise<{ armed: ArmedDocument[]; failures: FrameCaptureFailure[] }> {
const { injectable, failures } = await listInjectableFrames(tabId, deps);
const armed: ArmedDocument[] = [];
for (const frame of injectable) {
const result = await armDocumentCapture(tabId, frame, startMsg, deps);
if (result.armed) armed.push(result.armed);
if (result.failure) failures.push(result.failure);
}
return { armed, failures };
}
export async function stopDocumentCapture(
armed: ArmedDocument,
stopMsg: RecordStopMessage,
deps: RecordFrameSendDeps,
): Promise<{ ok: true } | FrameCaptureFailure> {
try {
const response = await deps.sendToDocument(armed.tabId, armed.documentId, stopMsg);
if (isRecordStopAckOk(response)) return { ok: true };
return makeFrameCaptureFailure("flush_failed", {
frameId: armed.frameId,
documentId: armed.documentId,
url: armed.url,
detail: "content script did not confirm recorded steps",
});
} catch (err) {
return makeFrameCaptureFailure("flush_failed", {
frameId: armed.frameId,
documentId: armed.documentId,
url: armed.url,
detail: err instanceof Error ? err.message : "RECORD_STOP failed",
});
}
}
export async function cancelDocumentCapture(
armed: ArmedDocument,
cancelMsg: RecordCancelMessage,
deps: RecordFrameSendDeps,
): Promise<void> {
try {
await deps.sendToDocument(armed.tabId, armed.documentId, cancelMsg);
} catch {
// Best-effort cleanup.
}
}
@@ -0,0 +1,673 @@
import type { RenderedRef } from "@browser-skill/vom";
import { formatObservationFile, type ObservationAnnotation } from "@/lib/format-observation-file";
import { fallbackDescriptor, matchTarget } from "@/lib/match-target";
import { waitForPageSettled } from "@/lib/page-settled";
import {
CAPTURE_RETRY_DELAY_MS,
DEFAULT_MAX_PAGE_TOKENS,
OBSERVATION_MIN_INTERVAL_MS,
} from "@/lib/record-constants";
import { registerObservation, type StateRegistryEntry } from "@/lib/trace-reducer";
import { captureVomObservation } from "@/tools/capture-vom-observation";
import type { CdpRunner, ChromeTabsApi } from "@/tools/shared";
import type { CapturedNode } from "@/tools/vom/capture";
import type { DraftTraceStep, Step, StopReason, Trace, TraceState } from "@/transport/types";
import { reduceTraceSteps, resolveTraceStartUrl } from "./trace-reducer";
export interface LastSettledObservation {
stateId: string;
captured: CapturedNode[];
matchNodes: CapturedNode[];
documentTokenOwners: Map<string, number | null>;
refs: RenderedRef[];
url: string;
title?: string;
vomText: string;
}
/** A post-action observation that has been queued but not yet written down. */
interface PendingSettle {
draftIndex: number;
/** Set when a newer action has already decided where this step landed. */
cancelled: boolean;
}
/** Latest URL in an in-flight redirect chain awaiting coalesce + settle. */
export interface PendingRedirectLanding {
url: string;
/** Bumped on every hop so an in-flight `waitForPageSettled` can cancel. */
generation: number;
}
export interface RecordingObservationState {
stateRegistry: Map<string, StateRegistryEntry>;
lastSettled: LastSettledObservation | null;
/** Latest token→owner map from the most recent observation capture. */
documentTokenOwners: Map<string, number | null>;
maxPageTokens: number;
redactValues: boolean;
lastCaptureAtMs: number;
/**
* Tail of the settle chain. Observations run one at a time so that a slow
* capture can never land after a faster one and report the pages out of the
* order the user visited them.
*/
settleQueue: Promise<void>;
/** Queued and in-flight settles, by draft index, so newer actions can supersede them. */
settles: Map<number, PendingSettle>;
stepAnnotations: Map<number, ObservationAnnotation[]>;
/**
* OAuth / server redirects are coalesced here: intermediate hops only update
* `url`+`generation`, and one `navigate` is emitted after the page settles.
*/
pendingRedirect: PendingRedirectLanding | null;
redirectFlushQueue: Promise<void>;
}
export function createObservationState(options?: {
maxPageTokens?: number;
redactValues?: boolean;
}): RecordingObservationState {
return {
stateRegistry: new Map(),
lastSettled: null,
documentTokenOwners: new Map(),
maxPageTokens: options?.maxPageTokens ?? DEFAULT_MAX_PAGE_TOKENS,
redactValues: options?.redactValues ?? false,
lastCaptureAtMs: 0,
settleQueue: Promise.resolve(),
settles: new Map(),
stepAnnotations: new Map(),
pendingRedirect: null,
redirectFlushQueue: Promise.resolve(),
};
}
function sleep(ms: number): Promise<void> {
return new Promise((resolve) => setTimeout(resolve, ms));
}
async function readTabMeta(
tabsApi: ChromeTabsApi,
tabId: number,
): Promise<{ url: string; title?: string }> {
try {
const tab = await tabsApi.get(tabId);
return { url: tab.url ?? "about:blank", title: tab.title };
} catch {
return { url: "about:blank" };
}
}
export async function captureAndRegisterObservation(
obs: RecordingObservationState,
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
urlOverride?: string,
): Promise<LastSettledObservation> {
const now = Date.now();
const waitMs = Math.max(0, OBSERVATION_MIN_INTERVAL_MS - (now - obs.lastCaptureAtMs));
if (waitMs > 0) await sleep(waitMs);
const { url, title } = urlOverride
? { url: urlOverride, title: undefined }
: await readTabMeta(tabsApi, tabId);
const meta = urlOverride ? await readTabMeta(tabsApi, tabId) : { url, title };
const resolvedTitle = title ?? meta.title;
const rendered = await captureVomObservation(cdp, tabId, url, {
maxTokens: obs.maxPageTokens,
redactValues: obs.redactValues,
conditionalSurfaceProbe: false,
});
const stateId = registerObservation(obs.stateRegistry, {
url,
title: resolvedTitle,
rawVomText: rendered.text,
truncated: rendered.truncated,
});
obs.lastCaptureAtMs = Date.now();
obs.documentTokenOwners = rendered.documentTokenOwners;
const settled: LastSettledObservation = {
stateId,
captured: rendered.captured,
matchNodes: rendered.matchNodes,
documentTokenOwners: rendered.documentTokenOwners,
refs: rendered.refs,
url,
title: resolvedTitle,
vomText: rendered.text,
};
obs.lastSettled = settled;
return settled;
}
/**
* Bind a draft to the page it was performed on: origin state, geometric target
* match, and the inline annotation. All three read the *current* observation,
* which is only the page the user acted on while the draft is still fresh —
* once the action settles, `lastSettled` is the destination page instead.
*/
export function applyTargetMatching(
obs: RecordingObservationState,
draft: DraftTraceStep,
stepId: number,
): void {
if (!obs.lastSettled) {
// Initial observation can fail or race the first action. Keep whatever
// the content script captured so the exported step is still teachable.
if ("target" in draft && "captureTarget" in draft) {
draft.target = fallbackDescriptor(draft.captureTarget);
}
rememberStepAnnotation(obs, stepId, draft);
return;
}
draft.preStateId = obs.lastSettled.stateId;
if ("target" in draft) {
draft.target =
"geometry" in draft && draft.geometry
? matchTarget({
geometry: draft.geometry,
captured: obs.lastSettled.matchNodes,
refs: obs.lastSettled.refs,
fallback: draft.captureTarget,
})
: // No geometry to match on, but the capture still knew what the user
// touched — keep that instead of an anonymous unmatched target.
fallbackDescriptor(draft.captureTarget);
}
rememberStepAnnotation(obs, stepId, draft);
}
function fillDetailForDraft(
obs: RecordingObservationState,
draft: DraftTraceStep,
): string | undefined {
if (obs.redactValues) return undefined;
if (draft.op === "fill") return JSON.stringify(draft.value);
return undefined;
}
function refLineForDraft(
obs: RecordingObservationState,
draft: DraftTraceStep,
): number | undefined {
if (!("target" in draft) || !obs.lastSettled) return undefined;
const targetRef = draft.target?.ref;
if (!targetRef) return undefined;
return obs.lastSettled.refs.find((r) => r.ref === targetRef)?.line;
}
function rememberStepAnnotation(
obs: RecordingObservationState,
stepId: number,
draft: DraftTraceStep,
): void {
const line = refLineForDraft(obs, draft);
if (line === undefined || draft.op === "navigate" || draft.op === "scroll") return;
if (!draft.preStateId) return;
const ann: ObservationAnnotation = {
stepId,
op: draft.op,
line,
stateId: draft.preStateId,
detail: fillDetailForDraft(obs, draft),
};
const bucket = obs.stepAnnotations.get(line) ?? [];
bucket.push(ann);
obs.stepAnnotations.set(line, bucket);
}
/** Record the step under the page it was performed on, not the one it led to. */
export function rememberStepOnPage(
obs: RecordingObservationState,
draft: DraftTraceStep,
stepId: number,
): void {
const stateId = draft.preStateId;
if (!stateId) return;
const entry = obs.stateRegistry.get(stateId);
if (entry && !entry.stepsHere.includes(stepId)) {
entry.stepsHere.push(stepId);
}
}
/** Drop an in-flight redirect coalesce (e.g. a real navigate superseded it). */
export function clearPendingRedirectLanding(obs: RecordingObservationState): void {
obs.pendingRedirect = null;
}
function enqueueRedirectFlush(obs: RecordingObservationState, task: () => Promise<void>): void {
obs.redirectFlushQueue = obs.redirectFlushQueue.then(task, task).catch(() => {});
}
/**
* Remember a redirect hop and schedule a settle-then-emit of a single navigate
* to the final URL. Later hops only bump `generation` / replace `url`.
*/
export function scheduleRedirectLandingFlush(
obs: RecordingObservationState,
steps: DraftTraceStep[],
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
url: string,
options: { cancelled?: () => boolean } = {},
): void {
const generation = (obs.pendingRedirect?.generation ?? 0) + 1;
obs.pendingRedirect = { url, generation };
enqueueRedirectFlush(obs, () =>
coalesceRedirectLanding(obs, steps, cdp, tabId, tabsApi, options),
);
}
/**
* Drain any coalesced redirect so the next action matches against the real
* landing page. Safe to call when nothing is pending.
*/
export async function flushPendingRedirectLanding(
obs: RecordingObservationState,
steps: DraftTraceStep[],
cdp: CdpRunner | undefined,
tabId: number,
tabsApi: ChromeTabsApi,
options: { cancelled?: () => boolean } = {},
): Promise<void> {
if (obs.pendingRedirect && cdp) {
enqueueRedirectFlush(obs, () =>
coalesceRedirectLanding(obs, steps, cdp, tabId, tabsApi, options),
);
}
await obs.redirectFlushQueue;
}
/**
* Wait until the redirect chain stops changing the document, then emit one
* `navigate` (cause `browser`) to the tab's final URL and settle its observation.
*/
async function coalesceRedirectLanding(
obs: RecordingObservationState,
steps: DraftTraceStep[],
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
options: { cancelled?: () => boolean },
): Promise<void> {
while (obs.pendingRedirect) {
if (options.cancelled?.()) {
obs.pendingRedirect = null;
return;
}
const snap = obs.pendingRedirect;
const outcome = await waitForPageSettled(cdp, tabId, {
cancelled: () =>
!!options.cancelled?.() ||
obs.pendingRedirect === null ||
obs.pendingRedirect.generation !== snap.generation,
});
if (options.cancelled?.()) {
obs.pendingRedirect = null;
return;
}
// A newer hop cancelled this wait — loop and settle against the latest URL.
if (outcome === "cancelled") continue;
if (!obs.pendingRedirect || obs.pendingRedirect.generation !== snap.generation) {
continue;
}
const finalUrl = (await readTabMeta(tabsApi, tabId)).url || snap.url;
obs.pendingRedirect = null;
if (!finalUrl || finalUrl === "about:blank") return;
if (obs.lastSettled?.url === finalUrl) return;
const last = steps[steps.length - 1];
if (last?.op === "navigate" && last.url === finalUrl) {
if (!last.postStateId) {
const draftIndex = steps.length - 1;
scheduleDraftSettle(obs, draftIndex, draftIndex + 1, cdp, tabId, tabsApi, steps);
await obs.settleQueue;
}
return;
}
if (outcome === "timeout") {
console.debug(
`[bsk record] redirect landing still changing after settle budget; ` +
`recording navigate to ${finalUrl}`,
);
}
const draft: DraftTraceStep = {
op: "navigate",
url: finalUrl,
page_url: finalUrl,
cause: "browser",
preStateId: obs.lastSettled?.stateId,
};
steps.push(draft);
const draftIndex = steps.length - 1;
const stepId = draftIndex + 1;
rememberStepOnPage(obs, draft, stepId);
scheduleDraftSettle(obs, draftIndex, stepId, cdp, tabId, tabsApi, steps);
// Callers that flush before the next action need `lastSettled` to already
// be the landing page so target matching does not use the pre-redirect view.
await obs.settleQueue;
return;
}
}
/**
* A capture that lands mid-navigation fails on a destroyed execution context.
* That is transient, so give it one more chance before the step has to fall
* back to a stale state.
*/
async function captureWithRetry(
obs: RecordingObservationState,
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
urlOverride?: string,
): Promise<LastSettledObservation> {
try {
return await captureAndRegisterObservation(obs, cdp, tabId, tabsApi, urlOverride);
} catch (err) {
console.debug("[bsk record] observation failed, retrying once", err);
await sleep(CAPTURE_RETRY_DELAY_MS);
return captureAndRegisterObservation(obs, cdp, tabId, tabsApi, urlOverride);
}
}
export async function settleDraftObservation(
obs: RecordingObservationState,
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
steps: DraftTraceStep[],
draftIndex: number,
pending?: PendingSettle,
): Promise<void> {
const cancelled = () => pending?.cancelled === true;
const outcome = await waitForPageSettled(cdp, tabId, { cancelled });
if (outcome === "cancelled") return;
if (outcome === "timeout") {
console.debug(
`[bsk record] step ${draftIndex + 1} was still changing the page after the settle ` +
"budget; observing it as it stands",
);
}
const started = steps[draftIndex];
if (!started) return;
// The URL attached to the draft may only be an intermediate redirect hop.
// Once the page is settled, register the capture against live tab metadata.
const settled = await captureWithRetry(obs, cdp, tabId, tabsApi);
// The observation still counts as the latest view of the page even when this
// step no longer wants it — the next action will start from it.
if (cancelled()) return;
// Re-read the slot: a navigation observed while the capture ran may have
// rewritten this draft, and the observation belongs to whatever occupies the
// slot now — writing to the object we started with would strand it.
const draft = steps[draftIndex] ?? started;
draft.postStateId = settled.stateId;
}
/**
* Recover landing states from the shape of the recording itself: wherever the
* next action was performed is, by definition, where the previous one landed.
* Only the immediately following draft counts — a later one would claim a page
* that several unobserved actions away, which is worse than admitting we saw
* no change.
*/
export function inferMissingPostStates(steps: DraftTraceStep[]): void {
for (let i = 0; i < steps.length - 1; i += 1) {
const draft = steps[i];
const next = steps[i + 1];
if (!draft || !next || draft.postStateId || !next.preStateId) continue;
draft.postStateId = next.preStateId;
console.debug(
`[bsk record] step ${i + 1} (${draft.op}) had no post-action observation; ` +
`using where step ${i + 2} started (${next.preStateId})`,
);
}
}
/**
* Last chance for the closing actions of a recording. Stopping right after the
* final action is normal, and so is that action navigating away, so take one
* final look at the page. Only trailing drafts may claim it: an earlier step
* did not land on whatever the user happens to be looking at when they stop.
*/
export async function settleUnsettledDrafts(
obs: RecordingObservationState,
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
steps: DraftTraceStep[],
): Promise<void> {
const trailing: DraftTraceStep[] = [];
for (let i = steps.length - 1; i >= 0; i -= 1) {
const draft = steps[i];
if (!draft || draft.postStateId) break;
trailing.push(draft);
}
if (trailing.length === 0) return;
let settled: LastSettledObservation;
try {
settled = await captureWithRetry(obs, cdp, tabId, tabsApi);
} catch (err) {
console.warn("[bsk record] final observation at stop failed", err);
return;
}
for (const draft of trailing) draft.postStateId = settled.stateId;
console.debug(
`[bsk record] settled ${trailing.length} trailing step(s) against the page at stop ` +
`(${settled.stateId} ${settled.url})`,
);
}
/**
* Performing a new action ends the previous one: the page the user reached for
* is, by definition, where the earlier step left them. Taking the landing from
* the newer step also keeps the trace monotonic — a capture that finished late
* would otherwise credit an earlier step with a page that only exists because
* of the newer action.
*/
function supersedeEarlierSettles(
obs: RecordingObservationState,
draftIndex: number,
steps: DraftTraceStep[],
): void {
const landing = steps[draftIndex]?.preStateId;
for (const [index, pending] of obs.settles) {
if (index >= draftIndex) continue;
pending.cancelled = true;
obs.settles.delete(index);
const draft = steps[index];
if (!draft || draft.postStateId) continue;
if (!landing) continue;
draft.postStateId = landing;
console.debug(
`[bsk record] step ${index + 1} (${draft.op}) was still settling when step ` +
`${draftIndex + 1} started; landing it on ${landing}`,
);
}
}
export function scheduleDraftSettle(
obs: RecordingObservationState,
draftIndex: number,
stepId: number,
cdp: CdpRunner,
tabId: number,
tabsApi: ChromeTabsApi,
steps: DraftTraceStep[],
): void {
supersedeEarlierSettles(obs, draftIndex, steps);
// Rescheduling the same step (a navigation showed up after the action)
// replaces the pending observation rather than racing it.
const superseded = obs.settles.get(draftIndex);
if (superseded) superseded.cancelled = true;
const pending: PendingSettle = { draftIndex, cancelled: false };
obs.settles.set(draftIndex, pending);
enqueueSettle(obs, async () => {
if (pending.cancelled) return;
try {
await settleDraftObservation(obs, cdp, tabId, tabsApi, steps, draftIndex, pending);
} catch (err) {
// Recoverable: stop-time repair still gives the step a landing page.
const op = steps[draftIndex]?.op ?? "?";
console.warn(`[bsk record] post-action observation failed for step ${stepId} (${op})`, err);
} finally {
if (obs.settles.get(draftIndex) === pending) obs.settles.delete(draftIndex);
}
});
}
function enqueueSettle(obs: RecordingObservationState, task: () => Promise<void>): void {
obs.settleQueue = obs.settleQueue.then(task, task).catch(() => {});
}
export function cancelPendingSettles(obs: RecordingObservationState): void {
for (const pending of obs.settles.values()) pending.cancelled = true;
obs.settles.clear();
}
/** Bound the drain loop so a self-rescheduling settle cannot block stop. */
const MAX_SETTLE_FLUSH_ROUNDS = 10;
/**
* Drain settle work before exporting, including work queued while draining —
* a navigation observed during the last capture schedules another one, and
* the trace would drop it if stop did not wait.
*/
export async function flushPendingSettles(obs: RecordingObservationState): Promise<void> {
for (let round = 0; round < MAX_SETTLE_FLUSH_ROUNDS; round += 1) {
const drained = obs.settleQueue;
await drained;
if (obs.settleQueue === drained) return;
}
console.warn("[bsk record] settle queue kept growing at stop; exporting what has been observed");
}
function finalizeStateBodies(
entries: StateRegistryEntry[],
annotationsByState: Map<string, ObservationAnnotation[]>,
stepIdByDraftId: Map<number, number>,
idByOldId: Map<string, string>,
): TraceState[] {
return entries.map((entry) => {
const id = idByOldId.get(entry.id) ?? entry.id;
const body = formatObservationFile({
stateId: id,
url: entry.url,
title: entry.title,
stepsHere: remapStepIds(entry.stepsHere, stepIdByDraftId),
body: entry.rawVomText,
annotations: annotationsByState.get(entry.id) ?? [],
});
return {
id,
url: entry.url,
...(entry.title ? { title: entry.title } : {}),
body,
...(entry.truncated ? { truncated: true } : {}),
};
});
}
/** Draft ids only become step ids after collapsing and filtering. */
function remapStepIds(draftIds: number[], stepIdByDraftId: Map<number, number>): number[] {
const mapped = new Set<number>();
for (const draftId of draftIds) {
const stepId = stepIdByDraftId.get(draftId);
if (stepId !== undefined) mapped.add(stepId);
}
return [...mapped].sort((a, b) => a - b);
}
function collectAnnotationsByState(
obs: RecordingObservationState,
stepIdByDraftId: Map<number, number>,
): Map<string, ObservationAnnotation[]> {
const byState = new Map<string, ObservationAnnotation[]>();
for (const anns of obs.stepAnnotations.values()) {
for (const ann of anns) {
const stepId = stepIdByDraftId.get(ann.stepId);
if (stepId === undefined) continue;
const bucket = byState.get(ann.stateId) ?? [];
bucket.push({ ...ann, stepId });
byState.set(ann.stateId, bucket);
}
}
return byState;
}
/**
* Keep only the pages the published steps point at. Redirect hops and
* mid-load captures land in the registry too, and shipping them would invite
* a reader to treat a page the flow merely passed through as a real stop.
* With no steps at all the first observation is the whole artifact, so it
* stays.
*/
function selectPublishedStates(
registry: Map<string, StateRegistryEntry>,
steps: Step[],
): StateRegistryEntry[] {
const entries = [...registry.values()];
if (steps.length === 0) return entries.slice(0, 1);
const referenced = new Set<string>();
for (const step of steps) {
referenced.add(step.state);
referenced.add(step.result.state);
}
return entries.filter((entry) => referenced.has(entry.id));
}
export function buildTraceV3(input: {
obs: RecordingObservationState;
steps: DraftTraceStep[];
startedAt: string;
purpose?: string;
startUrl?: string;
stoppedBy: StopReason;
bskVersion: string;
frameCapture?: Trace["frame_capture"];
}): Trace {
const { steps, stepIdByDraftId } = reduceTraceSteps(input.steps, input.obs.stateRegistry);
const published = selectPublishedStates(input.obs.stateRegistry, steps);
// Renumber so the shipped dictionary reads s1..sN without holes where the
// dropped captures used to be.
const idByOldId = new Map(published.map((entry, index) => [entry.id, `s${index + 1}`]));
for (const step of steps) {
step.state = idByOldId.get(step.state) ?? step.state;
step.result.state = idByOldId.get(step.result.state) ?? step.result.state;
}
const annotationsByState = collectAnnotationsByState(input.obs, stepIdByDraftId);
const states = finalizeStateBodies(published, annotationsByState, stepIdByDraftId, idByOldId);
const startUrl = resolveTraceStartUrl(input.steps, input.startUrl, states);
return {
version: 3,
...(input.purpose ? { purpose: input.purpose } : {}),
recorded_at: new Date().toISOString(),
started_at: input.startedAt,
stopped_by: input.stoppedBy,
entry: { start_url: startUrl },
recorder: { bsk: input.bskVersion, vom: 1 },
...(input.frameCapture ? { frame_capture: input.frameCapture } : {}),
states,
steps,
};
}
+102 -26
View File
@@ -1,23 +1,29 @@
import type { DraftTraceStep } from "@/transport/types";
import type { DraftTraceStep, NavigationCause } from "@/transport/types";
import type { RecordStepPayload } from "./record-bridge";
import { mapNavigationCause } from "./record-constants";
export interface RecordingStepBuffer {
steps: DraftTraceStep[];
currentUrl?: string;
pendingNavigation: boolean;
pendingNavigationDeadline?: number;
lastTransitionType?: string;
lastTransitionQualifiers?: string[];
}
const NAVIGATION_TRIGGER_WINDOW_MS = 3_000;
function toDraftStep(payload: RecordStepPayload): DraftTraceStep | null {
const pageUrl = payload.page_url;
const geometry = payload.geometry;
switch (payload.op) {
case "click":
return payload.target
? {
op: "click",
target: payload.target,
target: { unmatched: true },
captureTarget: payload.target,
...(geometry ? { geometry } : {}),
...(pageUrl ? { page_url: pageUrl } : {}),
}
: null;
@@ -25,7 +31,9 @@ function toDraftStep(payload: RecordStepPayload): DraftTraceStep | null {
return payload.target
? {
op: "hover",
target: payload.target,
target: { unmatched: true },
captureTarget: payload.target,
...(geometry ? { geometry } : {}),
...(pageUrl ? { page_url: pageUrl } : {}),
}
: null;
@@ -33,8 +41,11 @@ function toDraftStep(payload: RecordStepPayload): DraftTraceStep | null {
return payload.target
? {
op: "fill",
target: payload.target,
target: { unmatched: true },
captureTarget: payload.target,
...(geometry ? { geometry } : {}),
value: payload.value ?? "",
...(payload.commit ? { commit: payload.commit } : {}),
...(payload.redacted ? { redacted: true } : {}),
...(pageUrl ? { page_url: pageUrl } : {}),
}
@@ -44,7 +55,10 @@ function toDraftStep(payload: RecordStepPayload): DraftTraceStep | null {
? {
op: "press",
key: payload.key,
...(payload.target ? { target: payload.target } : {}),
...(payload.target
? { target: { unmatched: true }, captureTarget: payload.target }
: {}),
...(geometry ? { geometry } : {}),
...(payload.modifiers?.length ? { modifiers: payload.modifiers } : {}),
...(pageUrl ? { page_url: pageUrl } : {}),
}
@@ -53,37 +67,69 @@ function toDraftStep(payload: RecordStepPayload): DraftTraceStep | null {
return payload.target && payload.values
? {
op: "select",
target: payload.target,
target: { unmatched: true },
captureTarget: payload.target,
...(geometry ? { geometry } : {}),
values: payload.values,
...(payload.labels?.length ? { labels: payload.labels } : {}),
...(pageUrl ? { page_url: pageUrl } : {}),
}
: null;
case "scroll":
return { op: "scroll", ...(pageUrl ? { page_url: pageUrl } : {}) };
case "navigate":
return null;
}
}
function annotateLastStepNavigation(buffer: RecordingStepBuffer, url: string): boolean {
/** Annotate the latest action with `navigated_to`. Returns its index, or -1. */
function annotateLastStepNavigation(buffer: RecordingStepBuffer, url: string): number {
for (let i = buffer.steps.length - 1; i >= 0; i -= 1) {
const step = buffer.steps[i];
if (!step) continue;
if (step.op === "click" || step.op === "press" || step.op === "select") {
buffer.steps[i] = { ...step, navigated_to: url };
return true;
if (step.op === "click" || step.op === "press" || step.op === "select" || step.op === "fill") {
// Mutate in place: an in-flight settle holds this object, and replacing
// it would strand that settle's observation on the discarded copy.
step.navigated_to = url;
return i;
}
break;
}
return false;
return -1;
}
function causeFromBuffer(buffer: RecordingStepBuffer): NavigationCause | null {
return mapNavigationCause({
transitionType: buffer.lastTransitionType,
transitionQualifiers: buffer.lastTransitionQualifiers,
});
}
/** Outcome of folding a main-frame URL change into the draft buffer. */
export type NavigationObserveResult =
| { kind: "noop" }
| { kind: "annotated"; index: number }
| { kind: "appended"; index: number }
/**
* Redirect hop: do not emit a step yet. The recorder coalesces the chain and
* emits one `navigate` after `waitForPageSettled` on the final URL.
*/
| { kind: "coalesce_redirect"; url: string };
export function observeRecordedNavigation(
buffer: RecordingStepBuffer,
url: string,
causedByAction?: boolean,
): void {
if (!url || url === buffer.currentUrl) return;
transitionType?: string,
transitionQualifiers?: string[],
): NavigationObserveResult {
if (!url || url === buffer.currentUrl) return { kind: "noop" };
buffer.currentUrl = url;
if (transitionType !== undefined) buffer.lastTransitionType = transitionType;
if (transitionQualifiers !== undefined) {
buffer.lastTransitionQualifiers = transitionQualifiers;
}
const pendingIsCurrent =
buffer.pendingNavigation &&
(buffer.pendingNavigationDeadline === undefined ||
@@ -92,40 +138,69 @@ export function observeRecordedNavigation(
if (causedByAction === true || (causedByAction === undefined && pendingIsCurrent)) {
buffer.pendingNavigation = false;
buffer.pendingNavigationDeadline = undefined;
if (!annotateLastStepNavigation(buffer, url)) {
buffer.steps.push({
op: "navigate",
url,
page_url: url,
});
const annotatedIndex = annotateLastStepNavigation(buffer, url);
if (annotatedIndex >= 0) {
return { kind: "annotated", index: annotatedIndex };
}
return;
const cause = causeFromBuffer(buffer);
if (cause === null) {
return { kind: "coalesce_redirect", url };
}
buffer.steps.push({
op: "navigate",
url,
page_url: url,
cause,
transitionType: buffer.lastTransitionType,
transitionQualifiers: buffer.lastTransitionQualifiers,
});
return { kind: "appended", index: buffer.steps.length - 1 };
}
buffer.pendingNavigation = false;
buffer.pendingNavigationDeadline = undefined;
const cause = causeFromBuffer(buffer);
if (cause === null) {
return { kind: "coalesce_redirect", url };
}
buffer.steps.push({
op: "navigate",
url,
page_url: url,
cause,
transitionType: buffer.lastTransitionType,
transitionQualifiers: buffer.lastTransitionQualifiers,
});
return { kind: "appended", index: buffer.steps.length - 1 };
}
/**
* Buffer one captured payload. Returns the index of the draft it appended, or
* `null` when the payload produced none (an unnamed click target, a navigation
* that only annotated the previous draft, …). Callers must not fall back to
* "the last draft" on `null`: that would re-match an already recorded step
* against a newer observation and overwrite its origin state.
*/
export function appendRecordedPayload(
buffer: RecordingStepBuffer,
payload: RecordStepPayload,
): void {
): number | null {
if (payload.op === "navigate") {
if (payload.url) {
observeRecordedNavigation(buffer, payload.url, payload.navigation_caused_by_action);
}
return;
if (!payload.url) return null;
const result = observeRecordedNavigation(
buffer,
payload.url,
payload.navigation_caused_by_action,
payload.transitionType,
payload.transitionQualifiers,
);
return result.kind === "appended" ? result.index : null;
}
const step = toDraftStep({
...payload,
page_url: payload.page_url ?? buffer.currentUrl,
});
if (!step) return;
if (!step) return null;
buffer.steps.push(step);
if (step.op === "click" || step.op === "press" || step.op === "select" || step.op === "fill") {
buffer.pendingNavigation = payload.expects_navigation === true;
@@ -133,4 +208,5 @@ export function appendRecordedPayload(
? Date.now() + NAVIGATION_TRIGGER_WINDOW_MS
: undefined;
}
return buffer.steps.length - 1;
}
+151 -139
View File
@@ -1,4 +1,17 @@
import type { DraftTraceStep, PageRef, SelectedOption, Step } from "@/transport/types";
import type {
DraftTraceStep,
FillCommit,
NavigationCause,
Step,
TraceState,
} from "@/transport/types";
import {
DEFAULT_FILL_COMMIT,
fnv1a64,
mapNavigationCause,
nextStateId,
resetStateIdCounterForTests,
} from "./record-constants";
const CLIPBOARD_KEYS = new Set(["a", "c", "v", "x", "A", "C", "V", "X"]);
const MODIFIER_ONLY_KEYS = new Set(["Meta", "Control", "Alt", "Shift", "OS", "Hyper", "Super"]);
@@ -12,7 +25,6 @@ export function shouldRecordPress(
const hasCtrlOrMeta = mods.includes("ctrl") || mods.includes("meta");
if (hasCtrlOrMeta && CLIPBOARD_KEYS.has(key)) return false;
if (key === "Enter" || key === "Escape") return true;
// Drop bare character typing — FillSession already records the value.
if (key.length === 1 && !hasCtrlOrMeta && !mods.includes("alt")) return false;
return false;
}
@@ -22,195 +34,191 @@ function shouldIncludeDraft(step: DraftTraceStep): boolean {
return true;
}
/** Collapse consecutive navigations to the last hop. */
function collapseNavigations(steps: DraftTraceStep[]): DraftTraceStep[] {
const out: DraftTraceStep[] = [];
for (const step of steps) {
/** A reduced draft plus the capture-order ids of every draft folded into it. */
interface CollapsedDraft {
draft: DraftTraceStep;
draftIds: number[];
}
/**
* Collapse a redirect chain into a single hop.
*
* The surviving step keeps the *first* hop's origin state and cause — that is
* where the user was and why they left — and the *last* hop's destination, so
* the intermediate URL the browser only passed through never becomes the
* step's `state` or `to`.
*/
function collapseNavigations(steps: DraftTraceStep[]): CollapsedDraft[] {
const out: CollapsedDraft[] = [];
steps.forEach((step, index) => {
const draftId = index + 1;
const prev = out[out.length - 1];
if (step.op === "navigate" && prev?.op === "navigate") {
out[out.length - 1] = step;
continue;
if (step.op === "navigate" && prev?.draft.op === "navigate") {
prev.draft = {
...prev.draft,
url: step.url,
postStateId: step.postStateId ?? prev.draft.postStateId,
};
prev.draftIds.push(draftId);
return;
}
out.push(step);
}
out.push({ draft: step, draftIds: [draftId] });
});
return out;
}
function collectUrls(steps: DraftTraceStep[], startUrl?: string): string[] {
const urls: string[] = [];
if (startUrl) urls.push(startUrl);
for (const step of steps) {
if (step.op === "navigate") {
urls.push(step.url);
continue;
}
if ("page_url" in step && step.page_url) urls.push(step.page_url);
if ("navigated_to" in step && step.navigated_to) urls.push(step.navigated_to);
}
const seen = new Set<string>();
const unique: string[] = [];
for (const url of urls) {
if (!url || seen.has(url)) continue;
seen.add(url);
unique.push(url);
}
return unique;
export interface StateRegistryEntry extends TraceState {
contentHash: string;
rawVomText: string;
stepsHere: number[];
}
function buildPageRegistry(
steps: DraftTraceStep[],
startUrl?: string,
): { pages: PageRef[]; urlToId: Map<string, string> } {
const urls = collectUrls(steps, startUrl);
const urlToId = new Map<string, string>();
const pages = urls.map((url, index) => {
const id = `p${index + 1}`;
urlToId.set(url, id);
return { id, url };
});
return { pages, urlToId };
export function stateDedupKey(body: string, url: string): string {
return `${fnv1a64(body)}:${url}`;
}
function pageIdFor(
url: string | undefined,
urlToId: Map<string, string>,
fallbackUrl?: string,
/** Register or reuse a page observation in the states dictionary. */
export function registerObservation(
registry: Map<string, StateRegistryEntry>,
input: {
url: string;
title?: string;
rawVomText: string;
truncated?: boolean;
},
): string {
if (url && urlToId.has(url)) return urlToId.get(url)!;
if (fallbackUrl && urlToId.has(fallbackUrl)) return urlToId.get(fallbackUrl)!;
return urlToId.values().next().value ?? "p1";
const key = stateDedupKey(input.rawVomText, input.url);
for (const entry of registry.values()) {
if (entry.contentHash === key) return entry.id;
}
const id = nextStateId();
registry.set(id, {
id,
url: input.url,
...(input.title ? { title: input.title } : {}),
truncated: input.truncated ?? false,
contentHash: key,
rawVomText: input.rawVomText,
stepsHere: [],
});
return id;
}
function pageUrlForDraft(step: DraftTraceStep, fallbackUrl?: string): string | undefined {
if (step.op === "navigate") return step.page_url ?? step.url;
if ("page_url" in step && step.page_url) return step.page_url;
return fallbackUrl;
}
function effectForNavigation(
navigatedTo: string | undefined,
urlToId: Map<string, string>,
): Step["effect"] {
if (!navigatedTo) return undefined;
const pageId = urlToId.get(navigatedTo);
if (!pageId) return undefined;
return { navigated_to: pageId };
}
function withEffect(step: Step, effect: Step["effect"]): Step {
if (!effect) return step;
return { ...step, effect };
}
function toSelection(values: string[], labels?: string[]): SelectedOption[] {
function toSelection(
values: string[],
labels?: string[],
): Array<{ value: string; label?: string }> {
return values.map((value, index) => ({
value,
...(labels?.[index] ? { label: labels[index] } : {}),
}));
}
function toV2Step(
step: DraftTraceStep,
id: number,
urlToId: Map<string, string>,
fallbackUrl?: string,
): Step | null {
if (!shouldIncludeDraft(step)) return null;
function navigateCauseForDraft(
draft: Extract<DraftTraceStep, { op: "navigate" }>,
): NavigationCause {
if (draft.cause) return draft.cause;
const mapped = mapNavigationCause({
transitionType: draft.transitionType,
transitionQualifiers: draft.transitionQualifiers,
});
return mapped ?? "browser";
}
const pageUrl = pageUrlForDraft(step, fallbackUrl);
const page = pageIdFor(pageUrl, urlToId, fallbackUrl);
function toV3Step(draft: DraftTraceStep, id: number): Step | null {
if (!shouldIncludeDraft(draft)) return null;
// An action the user really performed must not disappear because one of its
// observations is missing — a capture racing a document swap is routine, and
// the last step of a recording is the most exposed to it. Fall back to the
// endpoint we do know; only a step with no observation at all has nothing to
// point at.
const preState = draft.preStateId ?? draft.postStateId;
const postState = draft.postStateId ?? draft.preStateId;
if (!preState || !postState) return null;
switch (step.op) {
const common = { id, state: preState, result: { state: postState } };
switch (draft.op) {
case "navigate":
return {
op: "navigate",
id,
page: pageIdFor(step.url, urlToId, fallbackUrl),
to: step.url,
...common,
to: draft.url,
cause: navigateCauseForDraft(draft),
};
case "click":
return withEffect(
{
op: "click",
id,
page,
target: step.target,
},
effectForNavigation(step.navigated_to, urlToId),
);
return { op: "click", ...common, target: draft.target };
case "hover":
return {
op: "hover",
id,
page,
target: step.target,
};
return { op: "hover", ...common, target: draft.target };
case "fill":
return {
op: "fill",
id,
page,
target: step.target,
value: step.value,
...(step.redacted ? { redacted: true } : {}),
...common,
target: draft.target,
value: draft.value,
commit: draft.commit ?? DEFAULT_FILL_COMMIT,
...(draft.redacted ? { redacted: true } : {}),
};
case "press":
return withEffect(
{
op: "press",
id,
page,
key: step.key,
...(step.target ? { target: step.target } : {}),
...(step.modifiers?.length ? { modifiers: step.modifiers } : {}),
},
effectForNavigation(step.navigated_to, urlToId),
);
return {
op: "press",
...common,
key: draft.key,
...(draft.target ? { target: draft.target } : {}),
...(draft.modifiers?.length ? { modifiers: draft.modifiers } : {}),
};
case "select":
return withEffect(
{
op: "select",
id,
page,
target: step.target,
selection: toSelection(step.values, step.labels),
},
effectForNavigation(step.navigated_to, urlToId),
);
return {
op: "select",
...common,
target: draft.target,
selection: toSelection(draft.values, draft.labels),
};
case "scroll":
return { op: "scroll", ...common };
}
}
export interface ReducedTrace {
pages: PageRef[];
states: TraceState[];
steps: Step[];
/**
* Capture-order draft id (`draftIndex + 1`) → published step id. Collapsed
* and dropped drafts make the two diverge, so anything the recorder tagged
* with a draft id (`steps_here`, inline annotations) must be remapped.
*/
stepIdByDraftId: Map<number, number>;
}
/**
* Compile capture drafts into record-only trace v2 steps.
* Variable inputs are NOT classified here — executing agents infer that at run time.
* Compile capture drafts into record-only trace v3 steps.
* Drafts must already carry `preStateId` / `postStateId` from the recorder.
*/
export function reduceTraceSteps(steps: DraftTraceStep[], startUrl?: string): ReducedTrace {
export function reduceTraceSteps(
steps: DraftTraceStep[],
registry: Map<string, StateRegistryEntry>,
): ReducedTrace {
const collapsed = collapseNavigations(steps);
const { pages, urlToId } = buildPageRegistry(collapsed, startUrl);
const out: Step[] = [];
const stepIdByDraftId = new Map<number, number>();
let id = 1;
let lastUrl = startUrl;
for (const draft of collapsed) {
if (draft.op === "navigate") lastUrl = draft.url;
else if ("navigated_to" in draft && draft.navigated_to) lastUrl = draft.navigated_to;
else if ("page_url" in draft && draft.page_url) lastUrl = draft.page_url;
const step = toV2Step(draft, id, urlToId, lastUrl);
for (const { draft, draftIds } of collapsed) {
const step = toV3Step(draft, id);
if (!step) continue;
out.push(step);
for (const draftId of draftIds) stepIdByDraftId.set(draftId, id);
id += 1;
}
return { pages, steps: out };
const states = [...registry.values()].map(
({ contentHash: _hash, rawVomText: _body, stepsHere: _steps, ...state }) => state,
);
return { states, steps: out, stepIdByDraftId };
}
export function resolveTraceStartUrl(
drafts: DraftTraceStep[],
startUrl?: string,
pages?: PageRef[],
states?: TraceState[],
): string {
if (startUrl) return startUrl;
const navigate = drafts.find((step): step is Extract<DraftTraceStep, { op: "navigate" }> => {
@@ -220,5 +228,9 @@ export function resolveTraceStartUrl(
for (const draft of drafts) {
if ("page_url" in draft && draft.page_url) return draft.page_url;
}
return pages?.[0]?.url ?? "about:blank";
return states?.[0]?.url ?? "about:blank";
}
export type { FillCommit, NavigationCause };
/** Test seam */
export { resetStateIdCounterForTests };
@@ -75,52 +75,6 @@ describe("SessionManager", () => {
await expect(sm.start("aa11")).rejects.toThrow(/already exists/);
});
it("removes a newly created Agent Window when startup is aborted", async () => {
const aw = fakeAgentWindow();
let resolveCreate: (windowId: number) => void = () => {};
aw.createMock.mockImplementationOnce(
() =>
new Promise<number>((resolve) => {
resolveCreate = resolve;
}),
);
const sm = new SessionManager({ agentWindow: aw });
const controller = new AbortController();
const pending = sm.start("aa11", { signal: controller.signal });
controller.abort();
resolveCreate(777);
await expect(pending).rejects.toMatchObject({ name: "AbortError" });
expect(aw.removeMock).toHaveBeenCalledWith(777);
expect(sm.has("aa11")).toBe(false);
});
it("removes an incomplete Agent Window when active-tab setup fails", async () => {
const aw = fakeAgentWindow();
aw.ensureActiveTabMock.mockRejectedValueOnce(new Error("tab setup failed"));
const sm = new SessionManager({ agentWindow: aw });
await expect(sm.start("aa11")).rejects.toThrow("tab setup failed");
expect(aw.removeMock).toHaveBeenCalledWith(100);
expect(sm.has("aa11")).toBe(false);
});
it("surfaces the orphan Agent Window id when startup cleanup fails", async () => {
const aw = fakeAgentWindow();
aw.ensureActiveTabMock.mockRejectedValueOnce(new Error("tab setup failed"));
aw.removeMock.mockRejectedValueOnce(new Error("window removal denied"));
const sm = new SessionManager({ agentWindow: aw });
await expect(sm.start("aa11")).rejects.toMatchObject({
name: "SessionStartCleanupError",
windowId: 100,
message: expect.stringMatching(/cleanup of Agent Window 100 failed.*window removal denied/),
});
expect(sm.has("aa11")).toBe(false);
});
it("stop() closes the Agent Window and forgets the session", async () => {
const aw = fakeAgentWindow();
const sm = new SessionManager({ agentWindow: aw });
@@ -33,19 +33,4 @@ describe("RefStore", () => {
expect(s.resolve("@e11")).toBe(11);
expect(s.size()).toBe(2);
});
it("preserves the child CDP session as part of ref identity", () => {
const s = new RefStore();
s.set("e1", 42, {
tabId: 7,
frameId: "child-frame",
cdpSessionId: "child-session",
});
expect(s.resolveEntry("e1")).toMatchObject({
backendNodeId: 42,
tabId: 7,
frameId: "child-frame",
cdpSessionId: "child-session",
});
});
});
@@ -43,11 +43,12 @@ export const chromeAgentWindowApi: AgentWindowApi = {
return win.id;
},
async remove(windowId: number): Promise<void> {
// Callers decide whether a missing/failed removal is benign. In
// particular, transactional session-start cleanup must be able to
// surface a window it could not remove instead of reporting a false
// cancellation success while the Agent Window remains open.
await chrome.windows.remove(windowId);
try {
await chrome.windows.remove(windowId);
} catch (err) {
// Window may have been closed by the user already; ignore.
console.debug("[bh] chrome.windows.remove failed", err);
}
},
async ensureActiveTab(windowId: number, url: string): Promise<void> {
const tabs = await chrome.tabs.query({ windowId });
+12 -62
View File
@@ -31,38 +31,6 @@ export interface SessionStartOptions {
size?: { width: number; height: number };
/** Defaults to true so existing clients keep visible Agent Windows. */
focused?: boolean;
/** Cancellation for the transactional Agent Window startup sequence. */
signal?: AbortSignal;
}
export class SessionStartCleanupError extends Error {
readonly windowId: number;
readonly startupError: unknown;
readonly cleanupError: unknown;
constructor(windowId: number, startupError: unknown, cleanupError: unknown) {
const startupMessage =
startupError instanceof Error ? startupError.message : String(startupError);
const cleanupMessage =
cleanupError instanceof Error ? cleanupError.message : String(cleanupError);
super(
`session_start failed (${startupMessage}) and cleanup of Agent Window ${windowId} failed: ${cleanupMessage}`,
);
this.name = "SessionStartCleanupError";
this.windowId = windowId;
this.startupError = startupError;
this.cleanupError = cleanupError;
}
}
function sessionStartAbortError(): Error {
const error = new Error("session_start aborted");
error.name = "AbortError";
return error;
}
function throwIfSessionStartAborted(signal: AbortSignal | undefined): void {
if (signal?.aborted) throw sessionStartAbortError();
}
/**
@@ -172,36 +140,18 @@ export class SessionManager {
if (this.sessions.has(sessionId)) {
throw new Error(`[bh] session ${sessionId} already exists`);
}
throwIfSessionStartAborted(opts.signal);
let windowId: number | null = null;
try {
const { signal: _signal, ...createOptions } = opts;
windowId = await this.agentWindow.create(AGENT_WINDOW_HOME, createOptions);
throwIfSessionStartAborted(opts.signal);
await this.agentWindow.ensureActiveTab(windowId, AGENT_WINDOW_HOME);
throwIfSessionStartAborted(opts.signal);
const ctx: SessionContext = {
sessionId,
agentWindowId: windowId,
refStore: new RefStore(),
borrowedTabs: new Map(),
createdAtMs: this.now(),
};
this.sessions.set(sessionId, ctx);
this.windowIndex.set(windowId, sessionId);
return ctx;
} catch (startupError) {
if (windowId !== null) {
try {
await this.agentWindow.remove(windowId);
} catch (cleanupError) {
throw new SessionStartCleanupError(windowId, startupError, cleanupError);
}
}
throw startupError;
}
const windowId = await this.agentWindow.create(AGENT_WINDOW_HOME, opts);
await this.agentWindow.ensureActiveTab(windowId, AGENT_WINDOW_HOME);
const ctx: SessionContext = {
sessionId,
agentWindowId: windowId,
refStore: new RefStore(),
borrowedTabs: new Map(),
createdAtMs: this.now(),
};
this.sessions.set(sessionId, ctx);
this.windowIndex.set(windowId, sessionId);
return ctx;
}
/**
@@ -1,10 +1,9 @@
/**
* Per-session map from `@e<N>` snapshot refs to a CDP node address.
* Per-session map from `@e<N>` snapshot refs to CDP `backendNodeId`.
*
* Each fresh `tool.snapshot` resets the store: M6 will call
* `replace(...)` with the new ref → node address pairs. A node address
* includes the owning flat CDP session/frame when the element lives in
* an OOPIF; tools resolve live geometry from that identity at call time.
* `replace(...)` with the new ref → backendNodeId pairs. Tools like
* `tool.click` consume the store via `resolve("@e1")`.
*
* Refs are session-scoped (§7): looking up a ref in the wrong session
* returns `null`, never silently leaks. Storing values in different
@@ -16,19 +15,10 @@ export type BackendNodeId = number;
export interface RefEntry {
backendNodeId: BackendNodeId;
tabId: number | null;
frameId?: string;
cdpSessionId?: string;
generation: number;
}
export type RefInput =
| BackendNodeId
| {
backendNodeId: BackendNodeId;
tabId: number;
frameId?: string;
cdpSessionId?: string;
};
type RefInput = BackendNodeId | { backendNodeId: BackendNodeId; tabId: number };
export class RefStore {
private readonly map = new Map<string, RefEntry>();
@@ -54,7 +44,7 @@ export class RefStore {
}
/**
* Replace the entire store with a new ref → CDP node identity mapping.
* Replace the entire store with a new ref → backendNodeId mapping.
* Used after every fresh `tool.snapshot`.
*/
replace(entries: Iterable<readonly [string, RefInput]>): void {
@@ -63,20 +53,10 @@ export class RefStore {
for (const [ref, input] of entries) this.map.set(normaliseRef(ref), this.entry(input));
}
set(
ref: string,
id: BackendNodeId,
opts: {
tabId?: number;
frameId?: string;
cdpSessionId?: string;
} = {},
): void {
set(ref: string, id: BackendNodeId, opts: { tabId?: number } = {}): void {
this.map.set(normaliseRef(ref), {
backendNodeId: id,
tabId: opts.tabId ?? null,
...(opts.frameId ? { frameId: opts.frameId } : {}),
...(opts.cdpSessionId ? { cdpSessionId: opts.cdpSessionId } : {}),
generation: this.generation,
});
}
@@ -100,8 +80,6 @@ export class RefStore {
return {
backendNodeId: input.backendNodeId,
tabId: input.tabId,
...(input.frameId ? { frameId: input.frameId } : {}),
...(input.cdpSessionId ? { cdpSessionId: input.cdpSessionId } : {}),
generation: this.generation,
};
}
@@ -612,14 +612,4 @@ describe("isInjectableContentScriptUrl", () => {
).toBe(false);
expect(isInjectableContentScriptUrl("https://chromewebstore.google.com/")).toBe(false);
});
it("rejects the Edge Add-ons store", () => {
expect(
isInjectableContentScriptUrl(
"https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg",
),
).toBe(false);
// Non-store paths on the same host stay injectable.
expect(isInjectableContentScriptUrl("https://microsoftedge.microsoft.com/")).toBe(true);
});
});
@@ -0,0 +1,86 @@
import { describe, expect, it } from "vitest";
import { RECORD_DOCUMENT_TOKEN_ATTR } from "@/lib/record-document-token";
import type { CapturedNode } from "@/tools/vom/capture";
import { collectDocumentTokenOwners } from "../capture-vom-observation";
function htmlNode(
backendNodeId: number,
token: string,
ownerFrameBackendNodeId: number | null = null,
): CapturedNode {
return {
backendNodeId,
parentBackendNodeId: null,
tag: "html",
attrs: { [RECORD_DOCUMENT_TOKEN_ATTR]: token },
rect: { x: 0, y: 0, w: 1280, h: 720 },
paintOrder: 0,
position: "static",
pointerEvents: "auto",
...(ownerFrameBackendNodeId === null ? {} : { ownerFrameBackendNodeId }),
};
}
describe("collectDocumentTokenOwners", () => {
it("maps document tokens to their immediate iframe owner backend node", () => {
const topToken = "top-doc-token";
const childToken = "child-doc-token";
const owners = collectDocumentTokenOwners({
viewport: { width: 1280, height: 720 },
nodes: [htmlNode(1, topToken, null)],
iframeNodes: new Map([
[
13,
[
htmlNode(101, childToken, 13),
{
backendNodeId: 102,
parentBackendNodeId: 101,
tag: "input",
attrs: {},
rect: { x: 10, y: 20, w: 200, h: 30 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
ownerFrameBackendNodeId: 13,
},
],
],
]),
excludedBackendNodeIds: new Set(),
});
expect(owners.get(topToken)).toBeNull();
expect(owners.get(childToken)).toBe(13);
});
it("maps nested iframe documents to their direct owner, not the outer iframe", () => {
const nestedToken = "nested-doc-token";
const owners = collectDocumentTokenOwners({
viewport: { width: 1280, height: 720 },
nodes: [htmlNode(1, "top-token", null)],
iframeNodes: new Map([
[
13,
[
{
backendNodeId: 23,
parentBackendNodeId: 101,
tag: "iframe",
attrs: {},
rect: { x: 0, y: 0, w: 400, h: 300 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
ownerFrameBackendNodeId: 13,
},
],
],
[23, [htmlNode(201, nestedToken, 23)]],
]),
excludedBackendNodeIds: new Set(),
});
expect(owners.get(nestedToken)).toBe(23);
});
});
@@ -203,55 +203,6 @@ describe("ToolDispatcher", () => {
expect(sent[0]).toEqual({ id: "r-1", result: {} });
});
it("forwards daemon cancellation into an in-flight session stop", async () => {
const { transport, sent, deliver } = fakeTransport();
const remove = vi.fn(async () => {});
const sessions = new SessionManager({
agentWindow: {
create: vi.fn(async () => 4242),
remove,
ensureActiveTab: vi.fn(async () => {}),
},
});
await sessions.start("aa11");
let finishDetach: () => void = () => {};
const detach = new Promise<void>((resolve) => {
finishDetach = resolve;
});
const cdp = {
send: vi.fn(),
detachSession: vi.fn(() => detach),
ensureNetworkCapture: vi.fn(async () => {}),
networkEntriesSince: vi.fn(() => ({
tab_id: 7,
entries: [],
next_since: 0,
truncated: false,
})),
setDeviceMetricsOverride: vi.fn(async () => {}),
clearDeviceMetricsOverride: vi.fn(async () => {}),
setUserAgentOverride: vi.fn(async () => {}),
setTouchEmulationEnabled: vi.fn(async () => {}),
};
const dispatcher = new ToolDispatcher({ transport, sessions, cdp: cdp as TestDispatcherCdp });
dispatcher.start();
deliver(makeRequest("tool.session_stop", { session_id: "aa11" }));
await vi.waitFor(() => expect(cdp.detachSession).toHaveBeenCalledWith("aa11"));
deliver({ id: "cancel-stop", method: "cancel", params: { rpc_id: "r-1" } });
await flushMicrotasks();
finishDetach();
await flushMicrotasks();
expect(sent).toContainEqual({ id: "cancel-stop", result: { cancelled: true } });
expect(sent).toContainEqual({
id: "r-1",
error: expect.objectContaining({ code: "cancelled" }),
});
expect(sessions.has("aa11")).toBe(true);
expect(remove).not.toHaveBeenCalled();
});
it("returns not_found when stopping an unknown session", async () => {
const { transport, sent, deliver } = fakeTransport();
const sessions = new SessionManager({
@@ -726,31 +677,23 @@ describe("ToolDispatcher", () => {
await flushMicrotasks();
expect(ac?.signal.aborted).toBe(true);
// Cancel ack arrives synchronously, but the original RPC must not reply
// until the in-progress window creation has completed and been rolled back.
// Cancel ack arrived synchronously; the slow tool replies with
// `cancelled` once the dispatcher's race observes the abort.
const ack = sent.find(
(m) =>
typeof (m as { id?: string }).id === "string" && (m as { id: string }).id === "cancel-1",
);
expect(ack).toEqual({ id: "cancel-1", result: { cancelled: true } });
expect(
sent.find(
(m) => typeof (m as { id?: string }).id === "string" && (m as { id: string }).id === "r-1",
),
).toBeUndefined();
expect(dispatcher.inflightAbortControllers.has("r-1")).toBe(true);
resolveCreate(9999);
await flushMicrotasks();
const slow = sent.find(
(m) => typeof (m as { id?: string }).id === "string" && (m as { id: string }).id === "r-1",
);
expect(slow).toMatchObject({ id: "r-1", error: { code: "cancelled" } });
expect(dispatcher.inflightAbortControllers.has("r-1")).toBe(false);
expect(sessions.has("aa44")).toBe(false);
expect(sessions.list()).toEqual([]);
// Drain the dangling create promise so vitest does not warn.
resolveCreate(9999);
await flushMicrotasks();
});
it("cancel for an unknown rpc_id replies with cancelled=false", async () => {
@@ -1,54 +1,24 @@
import { describe, expect, it } from "vitest";
import { nodeContentRegion } from "../element-geometry";
import { nodeBoundingRect, quadBoundingRect } from "../element-geometry";
import type { CdpRunner } from "../shared";
describe("nodeContentRegion", () => {
it("rejects invalid content quads before using the box model fallback", async () => {
const send = async (_tabId: number, method: string) => {
if (method === "DOM.getContentQuads") return { quads: [[0, 0, 0, 0, 0, 0, 0, 0]] };
if (method === "DOM.getBoxModel") {
return { model: { content: [5, 10, 25, 10, 25, 30, 5, 30] } };
}
throw new Error(`unexpected CDP call ${method}`);
};
await expect(nodeContentRegion({ send: send as CdpRunner["send"] }, 7, 555)).resolves.toEqual([
[
{ x: 5, y: 10 },
{ x: 25, y: 10 },
{ x: 25, y: 30 },
{ x: 5, y: 30 },
],
]);
});
it("preserves every visible content fragment", async () => {
const send = async () => ({
quads: [
[10, 20, 60, 20, 60, 40, 10, 40],
[15, 50, 90, 50, 90, 80, 15, 80],
],
describe("quadBoundingRect", () => {
it("computes axis-aligned bounds from an 8-double quad", () => {
expect(quadBoundingRect([10, 20, 110, 20, 110, 60, 10, 60])).toEqual({
x: 10,
y: 20,
width: 100,
height: 40,
});
const cdp: CdpRunner = {
send: send as CdpRunner["send"],
};
await expect(nodeContentRegion(cdp, 7, 555)).resolves.toEqual([
[
{ x: 10, y: 20 },
{ x: 60, y: 20 },
{ x: 60, y: 40 },
{ x: 10, y: 40 },
],
[
{ x: 15, y: 50 },
{ x: 90, y: 50 },
{ x: 90, y: 80 },
{ x: 15, y: 80 },
],
]);
});
it("returns null for degenerate quads", () => {
expect(quadBoundingRect([0, 0, 0, 0, 0, 0, 0, 0])).toBeNull();
expect(quadBoundingRect([1, 2, 3])).toBeNull();
});
});
describe("nodeBoundingRect", () => {
it("falls back to visible descendant bounds for zero-size containers", async () => {
const calls: Array<{ method: string; params?: object }> = [];
const send = async (_tabId: number, method: string, params?: object) => {
@@ -68,14 +38,12 @@ describe("nodeContentRegion", () => {
};
const cdp = { send: send as CdpRunner["send"] };
await expect(nodeContentRegion(cdp, 7, 555)).resolves.toEqual([
[
{ x: 242, y: 468 },
{ x: 1011, y: 468 },
{ x: 1011, y: 648 },
{ x: 242, y: 648 },
],
]);
await expect(nodeBoundingRect(cdp, 7, 555)).resolves.toEqual({
x: 242,
y: 468,
width: 769,
height: 180,
});
expect(calls.map((c) => c.method)).toEqual([
"DOM.getContentQuads",
"DOM.getBoxModel",
@@ -1,286 +0,0 @@
import { describe, expect, it, vi } from "vitest";
import { resolveFrameProjection, resolveNodeGeometry } from "../frame-geometry";
import {
clipPolygon,
polygonArea,
polygonBounds,
projectAndClipRegion,
projectUnitPoint,
type Quad,
rectPolygon,
regionBounds,
} from "../geometry";
import type { CdpRunner } from "../shared";
describe("frame geometry projection", () => {
it("keeps region bounds separate from polygon area", () => {
const region = [
rectPolygon({ x: 0, y: 0, w: 10, h: 10 }),
rectPolygon({ x: 20, y: 5, w: 5, h: 5 }),
];
expect(regionBounds(region)).toEqual({ x: 0, y: 0, width: 25, height: 10 });
expect(polygonArea(region[0])).toBe(100);
});
it("maps through the iframe content box instead of its border box", () => {
const contentQuad: Quad = [
{ x: 204, y: 306 },
{ x: 604, y: 306 },
{ x: 604, y: 506 },
{ x: 204, y: 506 },
];
const projected = projectAndClipRegion(
[rectPolygon({ x: 10, y: 20, w: 100, h: 40 })],
[{ sourceViewport: { width: 200, height: 100 }, destinationQuad: contentQuad }],
{ width: 1000, height: 800 },
);
expect(polygonBounds(projected.flat())).toEqual({ x: 224, y: 346, width: 200, height: 80 });
});
it("uses a projective mapping for perspective-transformed iframe quads", () => {
const quad: Quad = [
{ x: 0, y: 0 },
{ x: 200, y: 20 },
{ x: 180, y: 140 },
{ x: 20, y: 100 },
];
expect(projectUnitPoint({ x: 0, y: 0 }, quad)).toEqual(quad[0]);
expect(projectUnitPoint({ x: 1, y: 0 }, quad)).toEqual(quad[1]);
expect(projectUnitPoint({ x: 1, y: 1 }, quad)).toEqual(quad[2]);
expect(projectUnitPoint({ x: 0, y: 1 }, quad)).toEqual(quad[3]);
});
it("clips against the iframe polygon rather than its bounding box", () => {
const frameQuad: Quad = [
{ x: 50, y: 0 },
{ x: 100, y: 50 },
{ x: 50, y: 100 },
{ x: 0, y: 50 },
];
expect(clipPolygon(rectPolygon({ x: 0, y: 0, w: 10, h: 10 }), frameQuad)).toEqual([]);
});
it("does not reapply same-target frame transforms around an OOPIF", async () => {
const send = vi.fn(async (_tabId, method, params) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 800, clientHeight: 600 } };
}
if (method === "DOM.getBoxModel") {
const backendNodeId = (params as { backendNodeId?: number }).backendNodeId;
return backendNodeId === 20
? { model: { content: [110, 70, 310, 70, 310, 170, 110, 170] } }
: { model: { content: [100, 50, 400, 50, 400, 250, 100, 250] } };
}
throw new Error(`unexpected root command ${method}`);
});
const cdp: CdpRunner = {
send: send as CdpRunner["send"],
sendToTarget: vi.fn(async (_target, method) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
}
throw new Error(`unexpected child command ${method}`);
}) as CdpRunner["sendToTarget"],
};
const projection = await resolveFrameProjection(
cdp,
{
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "same-process-parent",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 4 },
},
{
frameId: "oopif",
parentFrameId: "same-process-parent",
ownerBackendNodeId: 20,
target: { tabId: 4, sessionId: "oopif-session" },
},
],
},
"oopif",
);
expect(projection?.edges).toEqual([
{
sourceViewport: { width: 200, height: 100 },
destinationQuad: [
{ x: 110, y: 70 },
{ x: 310, y: 70 },
{ x: 310, y: 170 },
{ x: 110, y: 170 },
],
destinationClips: [
[
{ x: 100, y: 50 },
{ x: 400, y: 50 },
{ x: 400, y: 250 },
{ x: 100, y: 250 },
],
],
},
]);
});
it("clips live OOPIF geometry at every frame boundary", async () => {
const cdp: CdpRunner = {
send: vi.fn(async (_tabId, method) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 800, clientHeight: 600 } };
}
if (method === "DOM.getBoxModel") {
return { model: { content: [100, 100, 300, 100, 300, 200, 100, 200] } };
}
throw new Error(`unexpected root command ${method}`);
}) as CdpRunner["send"],
sendToTarget: vi.fn(async (_target, method) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
}
if (method === "DOM.getContentQuads") {
return { quads: [[-50, -20, 250, -20, 250, 120, -50, 120]] };
}
throw new Error(`unexpected child command ${method}`);
}) as CdpRunner["sendToTarget"],
getFrameGraph: vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 4, sessionId: "child-session" },
},
],
})),
};
const geometry = await resolveNodeGeometry(cdp, 4, {
target: { tabId: 4, sessionId: "child-session" },
frameId: "child",
backendNodeId: 101,
});
expect(geometry).toMatchObject({
topBounds: { x: 100, y: 100, width: 200, height: 100 },
actionPoint: { x: 200, y: 150 },
});
});
it("reuses one frame graph while scrolling and projecting live geometry", async () => {
const getFrameGraph = vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 4, sessionId: "child-session" },
},
],
}));
const cdp: CdpRunner = {
send: vi.fn(async (_tabId, method) => {
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 800, clientHeight: 600 } };
}
if (method === "DOM.getBoxModel") {
return { model: { content: [100, 100, 300, 100, 300, 200, 100, 200] } };
}
throw new Error(`unexpected root command ${method}`);
}) as CdpRunner["send"],
sendToTarget: vi.fn(async (_target, method) => {
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
}
if (method === "DOM.getContentQuads") {
return { quads: [[0, 0, 20, 0, 20, 20, 0, 20]] };
}
throw new Error(`unexpected child command ${method}`);
}) as CdpRunner["sendToTarget"],
getFrameGraph,
};
await expect(
resolveNodeGeometry(
cdp,
4,
{
target: { tabId: 4, sessionId: "child-session" },
frameId: "child",
backendNodeId: 101,
},
{ scrollIntoView: true },
),
).resolves.not.toMatchObject({ code: expect.any(String) });
expect(getFrameGraph).toHaveBeenCalledOnce();
});
it("clips live geometry to same-target iframe ancestors", async () => {
const cdp: CdpRunner = {
send: vi.fn(async (_tabId, method) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 800, clientHeight: 600 } };
}
if (method === "DOM.getBoxModel") {
return { model: { content: [100, 100, 300, 100, 300, 200, 100, 200] } };
}
if (method === "DOM.getContentQuads") {
return { quads: [[50, 50, 350, 50, 350, 250, 50, 250]] };
}
throw new Error(`unexpected ${method}`);
}) as CdpRunner["send"],
getFrameGraph: vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 4 },
},
],
})),
};
const geometry = await resolveNodeGeometry(cdp, 4, {
target: { tabId: 4 },
frameId: "child",
backendNodeId: 101,
});
expect(geometry).toMatchObject({
topBounds: { x: 100, y: 100, width: 200, height: 100 },
actionPoint: { x: 200, y: 150 },
});
});
it("fails closed when an OOPIF frame graph is unavailable", async () => {
const cdp: CdpRunner = {
send: vi.fn(),
sendToTarget: vi.fn(async (_target, method) => {
if (method === "DOM.getContentQuads") {
return { quads: [[0, 0, 20, 0, 20, 20, 0, 20]] };
}
throw new Error(`unexpected ${method}`);
}) as CdpRunner["sendToTarget"],
};
await expect(
resolveNodeGeometry(cdp, 4, {
target: { tabId: 4, sessionId: "child-session" },
frameId: "child",
backendNodeId: 101,
}),
).resolves.toMatchObject({ code: "cdp_failed" });
});
});
@@ -1,6 +1,5 @@
import { afterEach, describe, expect, it, vi } from "vitest";
import { SessionManager } from "@/session-manager/manager";
import { RefStore } from "@/session-manager/ref-store";
import type { RequestHelpParams } from "@/transport/types";
import { handleRequestHelp, type RequestHelpDeps, resetHelpLifecycleForTests } from "../human-loop";
@@ -51,10 +50,11 @@ function installHelpLifecycleChrome() {
}
function fakeManager(sessionId: string, agentWindowId: number, tabId: number) {
const refStore = new RefStore();
const mgr = {
get: (id: string) =>
id === sessionId ? { sessionId, agentWindowId, refStore, borrowedTabs: new Map() } : null,
id === sessionId
? { sessionId, agentWindowId, refStore: { resolve: () => null }, borrowedTabs: new Map() }
: null,
findByWindowId: (wid: number) => (wid === agentWindowId ? { sessionId } : null),
} as unknown as SessionManager;
return mgr;
@@ -339,23 +339,26 @@ describe("handleRequestHelp", () => {
ac.abort();
await expect(pending).resolves.toMatchObject({ code: "cancelled" });
expect(cdpSend).not.toHaveBeenCalledWith(
expect.anything(),
"DOM.removeAttribute",
expect.anything(),
expect(cdpSend).toHaveBeenCalledWith(
5,
"DOM.querySelectorAll",
expect.objectContaining({ selector: "[data-bsk-help]" }),
);
expect(cdpSend).toHaveBeenCalledWith(
6,
"DOM.querySelectorAll",
expect.objectContaining({ selector: "[data-bsk-help]" }),
);
});
it("resolves ref targets to explicit viewport rectangles without mutating the page", async () => {
const refStore = new RefStore();
refStore.set("e1", 42, { tabId: 5 });
it("tags ref targets via CDP and reports them matched", async () => {
const mgr = {
get: (id: string) =>
id === "abcd"
? {
sessionId: "abcd",
agentWindowId: 99,
refStore,
refStore: { resolve: () => 42 },
borrowedTabs: new Map(),
}
: null,
@@ -363,17 +366,7 @@ describe("handleRequestHelp", () => {
} as unknown as SessionManager;
const deps = baseDeps({
cdp: {
send: vi.fn(async (_tabId, method) => {
if (method === "DOM.getDocument") return { root: { nodeId: 1 } };
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "DOM.getContentQuads") {
return { quads: [[10, 20, 110, 20, 110, 60, 10, 60]] };
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1280, clientHeight: 720 } };
}
throw new Error(`unexpected ${method}`);
}),
send: vi.fn(async () => ({ object: { objectId: "obj-1" } })),
} as unknown as RequestHelpDeps["cdp"],
});
const res = await handleRequestHelp(
@@ -382,97 +375,14 @@ describe("handleRequestHelp", () => {
deps,
);
const sentMsg = (deps.sendToTab as ReturnType<typeof vi.fn>).mock.calls[0][1];
expect(sentMsg.selectors).toEqual([]);
expect(sentMsg.rects).toEqual([{ top: 20, left: 10, width: 100, height: 40 }]);
expect(sentMsg.selectors).toContain('[data-bsk-help="0"]');
expect(res).toMatchObject({
outcome: "continued",
tab_id: 5,
resolved_targets: [
{
matched: true,
ref: "@e1",
},
],
resolved_targets: [{ matched: true, ref: "@e1" }],
});
});
it("projects an OOPIF ref through its live content quad for the help overlay", async () => {
const mgr = {
get: (id: string) =>
id === "abcd"
? {
sessionId: "abcd",
agentWindowId: 99,
refStore: {
resolveEntry: () => ({
backendNodeId: 42,
tabId: 5,
frameId: "child",
cdpSessionId: "child-session",
generation: 1,
}),
},
borrowedTabs: new Map(),
}
: null,
findByWindowId: (wid: number) => (wid === 99 ? { sessionId: "abcd" } : null),
} as unknown as SessionManager;
const send = vi.fn(async (_tabId, method) => {
if (method === "DOM.getDocument") return { root: { nodeId: 1 } };
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "DOM.getBoxModel") {
return { model: { content: [204, 306, 604, 306, 604, 506, 204, 506] } };
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1280, clientHeight: 720 } };
}
throw new Error(`unexpected root ${method}`);
});
const sendToTarget = vi.fn(async (_target, method) => {
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "DOM.getContentQuads") {
return { quads: [[10, 20, 110, 20, 110, 60, 10, 60]] };
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
}
throw new Error(`unexpected child ${method}`);
});
const deps = baseDeps({
cdp: {
send,
sendToTarget,
getFrameGraph: vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 5 } },
{
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 99,
target: { tabId: 5, sessionId: "child-session" },
},
],
})),
} as unknown as NonNullable<RequestHelpDeps["cdp"]>,
});
const res = await handleRequestHelp(
mgr,
baseParams({ tab_id: 5, targets: [{ ref: "@e1" }] }),
deps,
);
const sentMsg = (deps.sendToTab as ReturnType<typeof vi.fn>).mock.calls[0][1];
expect(sentMsg.rects).toEqual([{ top: 346, left: 224, width: 200, height: 80 }]);
expect(res).toMatchObject({ resolved_targets: [{ matched: true, ref: "@e1" }] });
expect(sendToTarget).toHaveBeenCalledWith(
{ tabId: 5, sessionId: "child-session" },
"DOM.getContentQuads",
{ backendNodeId: 42 },
);
});
it("reports ref target unmatched when ref does not resolve", async () => {
const deps = baseDeps();
const res = await handleRequestHelp(
@@ -488,15 +398,16 @@ describe("handleRequestHelp", () => {
});
it("reports ref target unmatched when ref is for another tab", async () => {
const refStore = new RefStore();
refStore.set("e1", 42, { tabId: 4 });
const mgr = {
get: (id: string) =>
id === "abcd"
? {
sessionId: "abcd",
agentWindowId: 99,
refStore,
refStore: {
resolve: (ref: string, opts: { tabId?: number }) =>
ref === "e1" && opts.tabId === 4 ? 42 : null,
},
borrowedTabs: new Map(),
}
: null,
@@ -30,9 +30,6 @@ function makeFakeCdp(handlers: Record<string, (params: unknown) => unknown>) {
const sendImpl = async (tabId: number, method: string, params?: object) => {
sent.push({ tabId, method, params });
const h = handlers[method];
if (!h && method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1280, clientHeight: 720 } };
}
if (!h) throw new Error(`unexpected CDP call ${method}`);
return h(params);
};
@@ -167,63 +164,6 @@ describe("handleClick", () => {
});
});
it("resolves frame refs in their CDP session and dispatches input in top coordinates", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.refStore.set("e3", 1234, {
tabId: 4,
frameId: "child-frame",
cdpSessionId: "child-session",
});
const fake = makeFakeCdp({
"Input.dispatchMouseEvent": () => ({}),
"DOM.scrollIntoViewIfNeeded": () => ({}),
"DOM.getBoxModel": () => ({
model: { content: [204, 306, 604, 306, 604, 506, 204, 506] },
}),
});
fake.cdp.getFrameGraph = vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "child-frame",
parentFrameId: "main",
ownerBackendNodeId: 99,
target: { tabId: 4, sessionId: "child-session" },
},
],
}));
const targetCalls: Array<{ sessionId?: string; method: string }> = [];
fake.cdp.sendToTarget = vi.fn(async (target, method) => {
targetCalls.push({ sessionId: target.sessionId, method });
if (method === "DOM.scrollIntoViewIfNeeded") return {};
if (method === "DOM.getContentQuads") {
return { quads: [[10, 20, 110, 20, 110, 60, 10, 60]] };
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
}
throw new Error(`unexpected child CDP call ${method}`);
}) as CdpRunner["sendToTarget"];
const res = await handleClick(
sm,
{ session_id: "aa11", ref: "@e3" },
{ cdp: fake.cdp, tabsApi: fake.tabsApi },
);
if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
// The iframe content box starts at (204, 306) after its border and is scaled 2x.
expect(res).toMatchObject({ x: 324, y: 386 });
expect(targetCalls).toEqual([
{ sessionId: "child-session", method: "DOM.scrollIntoViewIfNeeded" },
{ sessionId: "child-session", method: "DOM.getContentQuads" },
{ sessionId: "child-session", method: "Page.getLayoutMetrics" },
]);
expect(fake.sent.filter((call) => call.method === "Input.dispatchMouseEvent")).toHaveLength(3);
});
it("enables overlay bypass before mouse events when overlay blocks the click point", async () => {
const order: string[] = [];
const bypassOverlay = vi.fn(async (_tabId: number, enabled: boolean) => {
@@ -5,10 +5,8 @@ import { OVERLAY_HOST_MARKER_ATTR, OVERLAY_HOST_NAME } from "@/lib/overlay-bridg
import { SessionManager } from "@/session-manager/manager";
import type { CdpRunner } from "@/tools/shared";
import {
buildFrameVomScene,
buildVomScene,
type CdpAxNode,
captureVomObservation,
handleGetHtml,
handleObserve,
handleScreenshot,
@@ -64,9 +62,6 @@ function makeFakeCdp(handlers: Record<string, (params?: object) => unknown>) {
const send = vi.fn(async (_tabId: number, method: string, params?: object) => {
sent.push({ method, params });
const handler = handlers[method];
if (!handler && method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1280, clientHeight: 720 } };
}
if (!handler) throw new Error(`unexpected CDP call ${method}`);
return handler(params);
});
@@ -513,292 +508,6 @@ describe("handleScreenshot overlay suppression", () => {
// ---------------------------------------------------------------------------
describe("buildVomScene", () => {
it("merges multiple sibling and nested frames without mixing their targets", () => {
const node = (
backendNodeId: number,
parentBackendNodeId: number | null,
frameId: string,
tag: string,
attrs: Record<string, string> = {},
): CapturedNode => ({
backendNodeId,
parentBackendNodeId,
frameId,
tag,
attrs,
rect: { x: 0, y: 0, w: 100, h: 30 },
localRect: { x: 0, y: 0, w: 100, h: 30 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
});
const mainNodes = [
node(1, null, "main", "body"),
node(10, 1, "main", "iframe", { title: "First" }),
node(20, 1, "main", "iframe", { title: "Second" }),
];
const firstNodes = [
node(100, null, "first", "body"),
node(101, 100, "first", "button", { "aria-label": "First action" }),
];
const secondNodes = [
node(200, null, "second", "body"),
node(201, 200, "second", "h2"),
node(210, 200, "second", "iframe", { title: "Nested" }),
];
const nestedNodes = [
node(300, null, "nested", "body"),
node(301, 300, "nested", "input", { placeholder: "Nested field" }),
];
const captured: CapturedViewModel = {
viewport: { width: 1200, height: 800 },
nodes: mainNodes,
iframeNodes: new Map([
[10, firstNodes],
[20, secondNodes],
[210, nestedNodes],
]),
frameNodes: new Map([
["main", mainNodes],
["first", firstNodes],
["second", secondNodes],
["nested", nestedNodes],
]),
frameOwnerBackendNodeIds: new Map([
["first", 10],
["second", 20],
["nested", 210],
]),
rootFrameId: "main",
excludedBackendNodeIds: new Set(),
};
const axNode = (
nodeId: string,
backendDOMNodeId: number,
role: string,
parentId?: string,
name?: string,
): CdpAxNode => ({
nodeId,
backendDOMNodeId,
role: { type: "role", value: role },
...(parentId ? { parentId } : {}),
...(name ? { name: { type: "computedString", value: name } } : {}),
});
const result = buildFrameVomScene(
[
{
frameId: "main",
contextScopeId: "main",
target: { tabId: 7 },
domNodes: mainNodes,
axNodes: [
axNode("root", 1, "RootWebArea"),
axNode("frame-1", 10, "Iframe", "root", "First"),
axNode("frame-2", 20, "Iframe", "root", "Second"),
],
},
{
frameId: "first",
contextScopeId: "first",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 7, sessionId: "session-first" },
domNodes: firstNodes,
axNodes: [
axNode("root", 100, "RootWebArea"),
axNode("button", 101, "button", "root", "First action"),
],
},
{
frameId: "second",
contextScopeId: "second",
parentFrameId: "main",
ownerBackendNodeId: 20,
target: { tabId: 7, sessionId: "session-second" },
domNodes: secondNodes,
axNodes: [
axNode("root", 200, "RootWebArea"),
axNode("heading", 201, "heading", "root", "Second heading"),
axNode("nested-frame", 210, "Iframe", "root", "Nested"),
],
},
{
frameId: "nested",
contextScopeId: "nested",
parentFrameId: "second",
ownerBackendNodeId: 210,
target: { tabId: 7, sessionId: "session-nested" },
domNodes: nestedNodes,
axNodes: [
axNode("root", 300, "RootWebArea"),
axNode("input", 301, "textbox", "root", "Nested field"),
],
},
],
captured,
);
const rendered = renderVom(result);
expect(rendered.text).toContain('Iframe "First"');
expect(rendered.text).toContain('@e1 button "First action"');
expect(rendered.text).toContain('heading "Second heading"');
expect(rendered.text).toContain('Iframe "Nested"');
expect(rendered.text).toContain('textbox "Nested field"');
expect(rendered.refs.find((ref) => ref.backendNodeId === 101)?.frameId).toBe("first");
expect(rendered.refs.find((ref) => ref.backendNodeId === 301)?.frameId).toBe("nested");
const firstAction = result.nodes.find((item) => item.backendNodeId === 101);
const nestedInput = result.nodes.find((item) => item.backendNodeId === 301);
expect(firstAction).toMatchObject({ frameId: "first", contextScopeId: "first" });
expect(nestedInput).toMatchObject({ frameId: "nested", contextScopeId: "nested" });
expect(nestedInput?.parentId).toBe(result.nodes.find((item) => item.backendNodeId === 210)?.id);
});
it("does not place a child document at the page root when its frame boundary is unresolved", () => {
const childNode: CapturedNode = {
backendNodeId: 101,
parentBackendNodeId: null,
frameId: "child",
tag: "button",
attrs: {},
rect: { x: 0, y: 0, w: 100, h: 30 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
};
const captured: CapturedViewModel = {
viewport: { width: 800, height: 600 },
nodes: [],
iframeNodes: new Map(),
frameNodes: new Map([["child", [childNode]]]),
rootFrameId: "main",
excludedBackendNodeIds: new Set(),
};
const result = buildFrameVomScene(
[
{
frameId: "main",
contextScopeId: "main",
target: { tabId: 7 },
domNodes: [],
axNodes: [],
},
{
frameId: "child",
contextScopeId: "child",
parentFrameId: "main",
target: { tabId: 7 },
domNodes: [childNode],
axNodes: [
{
nodeId: "button",
frameId: "child",
backendDOMNodeId: 101,
role: { type: "role", value: "button" },
name: { type: "computedString", value: "Child action" },
},
],
},
],
captured,
);
expect(result.nodes).toEqual([]);
});
it("keeps AX-only frame content attached to its iframe boundary", () => {
const mainNodes: CapturedNode[] = [
{
backendNodeId: 1,
parentBackendNodeId: null,
frameId: "main",
tag: "body",
attrs: {},
rect: { x: 0, y: 0, w: 800, h: 600 },
paintOrder: 0,
position: "static",
pointerEvents: "auto",
},
{
backendNodeId: 10,
parentBackendNodeId: 1,
frameId: "main",
tag: "iframe",
attrs: { title: "Remote" },
rect: { x: 100, y: 100, w: 400, h: 300 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
},
];
const captured: CapturedViewModel = {
viewport: { width: 800, height: 600 },
nodes: mainNodes,
iframeNodes: new Map(),
frameNodes: new Map([
["main", mainNodes],
["child", []],
]),
rootFrameId: "main",
excludedBackendNodeIds: new Set(),
};
const result = buildFrameVomScene(
[
{
frameId: "main",
contextScopeId: "main",
target: { tabId: 7 },
domNodes: mainNodes,
axNodes: [
{
nodeId: "main-root",
backendDOMNodeId: 1,
role: { type: "role", value: "RootWebArea" },
},
{
nodeId: "frame-owner",
parentId: "main-root",
backendDOMNodeId: 10,
role: { type: "role", value: "Iframe" },
name: { type: "computedString", value: "Remote" },
},
],
},
{
frameId: "child",
contextScopeId: "child",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 7, sessionId: "child-session" },
domNodes: [],
axNodes: [
{
nodeId: "child-root",
backendDOMNodeId: 100,
role: { type: "role", value: "RootWebArea" },
},
{
nodeId: "child-button",
parentId: "child-root",
backendDOMNodeId: 101,
role: { type: "role", value: "button" },
name: { type: "computedString", value: "AX fallback action" },
},
],
},
],
captured,
);
const childButton = result.nodes.find((node) => node.backendNodeId === 101);
expect(childButton).toMatchObject({ frameId: "child", contextScopeId: "child" });
expect(childButton?.parentId).toBe(result.nodes.find((node) => node.backendNodeId === 10)?.id);
expect(renderVom(result).text).toContain('@e1 button "AX fallback action"');
});
it("joins AX semantics with captured geometry by backendDOMNodeId", () => {
const axNodes: CdpAxNode[] = [
{
@@ -1107,7 +816,6 @@ describe("buildVomScene", () => {
parentId: "2",
role: { type: "role", value: "textbox" },
name: { type: "x", value: "输入密码" },
value: { value: "iframe-secret" },
backendDOMNodeId: 202,
properties: [{ name: "inputType", value: { value: "password" } }],
},
@@ -1162,10 +870,6 @@ describe("buildVomScene", () => {
expect(passwordNodes).toHaveLength(1);
expect(passwordNodes[0]).toEqual(expect.objectContaining({ id: 202, sensitive: true }));
expect(passwordNodes[0].value).toBeUndefined();
const rendered = renderVom(buildVomScene(axNodes, captured)).text;
expect(rendered).toContain('textbox "输入密码" [filled] ="•••"');
expect(rendered).not.toContain("iframe-secret");
});
it("keeps unnamed iframe controls but skips unnamed iframe links", () => {
@@ -2757,64 +2461,6 @@ describe("handleSnapshot", () => {
expect(ctx.refStore.resolve("e1")).toBeNull();
});
it("keeps the previous RefStore when cancellation lands during DOM capture", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.refStore.set("e1", 999, { tabId: 4 });
const controller = new AbortController();
let resolveCapture: (value: unknown) => void = () => {};
const send = vi.fn(async (_tabId: number, method: string) => {
if (method === "Accessibility.enable") return {};
if (method === "Accessibility.getFullAXTree") {
return {
nodes: [
{
nodeId: "new",
role: { type: "role", value: "button" },
name: { type: "computedString", value: "New" },
backendDOMNodeId: 123,
},
],
};
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1000, clientHeight: 800 } };
}
if (method === "DOMSnapshot.enable") return {};
if (method === "DOMSnapshot.captureSnapshot") {
return new Promise((resolve) => {
resolveCapture = resolve;
});
}
throw new Error(`unexpected CDP method ${method}`);
});
const deps = {
cdp: {
send: send as unknown as <T = unknown>(
tabId: number,
method: string,
params?: object,
) => Promise<T>,
trackSessionTab: vi.fn(),
},
tabsApi: {
get: vi.fn(),
query: vi.fn(async () => [{ id: 4, windowId: 100, active: true } as chrome.tabs.Tab]),
},
};
const pending = handleSnapshot(sm, { session_id: "aa11" }, deps, controller.signal);
await vi.waitFor(() =>
expect(send).toHaveBeenCalledWith(4, "DOMSnapshot.captureSnapshot", expect.any(Object)),
);
controller.abort();
resolveCapture({ strings: [], documents: [] });
await expect(pending).resolves.toMatchObject({ code: "cancelled" });
expect(ctx.refStore.resolve("e1", { tabId: 4 })).toBe(999);
expect(ctx.refStore.resolve("e2")).toBeNull();
});
it("surfaces CDP failures as cdp_failed", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
await sm.start("aa11");
@@ -2909,360 +2555,10 @@ describe("handleSnapshot", () => {
],
};
}
function secretsSnapshotReply(secrets: {
email: string;
password: string;
otp: string;
card: string;
}) {
const S = [
"html",
"body",
"input",
"position",
"static",
"pointer-events",
"auto",
"type",
"text",
"password",
"autocomplete",
"one-time-code",
"cc-number",
"value",
secrets.email,
secrets.password,
secrets.otp,
secrets.card,
];
const i = (s: string) => S.indexOf(s);
const style = [i("static"), i("auto"), i("auto")];
return {
strings: S,
documents: [
{
nodes: {
parentIndex: [-1, 0, 1, 1, 1, 1],
nodeName: [i("html"), i("body"), i("input"), i("input"), i("input"), i("input")],
backendNodeId: [10, 11, 12, 13, 14, 15],
attributes: [
[],
[],
[i("type"), i("text"), i("value"), i(secrets.email)],
[i("type"), i("password"), i("value"), i(secrets.password)],
[
i("type"),
i("text"),
i("autocomplete"),
i("one-time-code"),
i("value"),
i(secrets.otp),
],
[
i("type"),
i("text"),
i("autocomplete"),
i("cc-number"),
i("value"),
i(secrets.card),
],
],
inputValue: {
index: [2, 3, 4, 5],
value: [i(secrets.email), i(secrets.password), i(secrets.otp), i(secrets.card)],
},
},
layout: {
nodeIndex: [1, 2, 3, 4, 5],
styles: [style, style, style, style, style],
bounds: [
[0, 0, 1000, 800],
[10, 10, 200, 32],
[10, 50, 200, 32],
[10, 90, 200, 32],
[10, 130, 200, 32],
],
paintOrders: [0, 1, 2, 3, 4],
},
},
],
};
}
const VP_METRICS = {
cssLayoutViewport: { clientWidth: 1000, clientHeight: 800, pageX: 0, pageY: 0 },
};
function makeFrameAwareDeps() {
const strings = ["html", "body", "iframe", "button", "title", "Remote", "static", "auto"];
const i = (value: string) => strings.indexOf(value);
const styles = [i("static"), i("auto"), i("auto")];
const snapshot = {
strings,
documents: [
{
frameId: "main",
nodes: {
parentIndex: [-1, 0, 1],
nodeName: [i("html"), i("body"), i("iframe")],
backendNodeId: [10, 11, 12],
attributes: [[], [], [i("title"), i("Remote")]],
contentDocumentIndex: { index: [2], value: [1] },
},
layout: {
nodeIndex: [0, 1, 2],
styles: [styles, styles, styles],
bounds: [
[0, 0, 1000, 800],
[0, 0, 1000, 800],
[100, 100, 400, 300],
],
paintOrders: [0, 0, 1],
},
},
{
frameId: "child",
nodes: {
parentIndex: [-1, 0, 1],
nodeName: [i("html"), i("body"), i("button")],
backendNodeId: [20, 21, 22],
attributes: [[], [], []],
},
layout: {
nodeIndex: [0, 1, 2],
styles: [styles, styles, styles],
bounds: [
[0, 0, 400, 300],
[0, 0, 400, 300],
[20, 30, 120, 40],
],
paintOrders: [0, 0, 1],
},
},
],
};
const mainAx: CdpAxNode[] = [
{
nodeId: "main-root",
role: { type: "role", value: "RootWebArea" },
backendDOMNodeId: 11,
childIds: ["frame-owner"],
},
{
nodeId: "frame-owner",
parentId: "main-root",
role: { type: "role", value: "Iframe" },
name: { type: "computedString", value: "Remote" },
backendDOMNodeId: 12,
},
];
const childAx: CdpAxNode[] = [
{
nodeId: "child-root",
role: { type: "role", value: "RootWebArea" },
backendDOMNodeId: 21,
childIds: ["child-button"],
},
{
nodeId: "child-button",
parentId: "child-root",
role: { type: "role", value: "button" },
name: { type: "computedString", value: "Frame action" },
backendDOMNodeId: 22,
},
];
const send = vi.fn(async (_tabId: number, method: string) => {
if (method === "Page.getLayoutMetrics") return VP_METRICS;
if (method === "DOMSnapshot.enable" || method === "Accessibility.enable") return {};
if (method === "DOMSnapshot.captureSnapshot") return snapshot;
if (method === "Accessibility.getFullAXTree") return { nodes: mainAx };
throw new Error(`unexpected root CDP method ${method}`);
});
const sendToTarget = vi.fn(async (_target, method: string) => {
if (method === "Accessibility.enable") return {};
if (method === "Accessibility.getFullAXTree") return { nodes: childAx };
throw new Error(`unexpected child CDP method ${method}`);
});
const cdp = {
send: send as unknown as CdpRunner["send"],
sendToTarget: sendToTarget as unknown as NonNullable<CdpRunner["sendToTarget"]>,
getFrameGraph: vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 12,
target: { tabId: 4, sessionId: "child-session" },
},
],
})),
trackSessionTab: vi.fn(),
} satisfies CdpRunner;
return {
cdp,
tabsApi: {
get: vi.fn(
async (tabId: number) => ({ id: tabId, windowId: 100, active: true }) as chrome.tabs.Tab,
),
query: vi.fn(async () => [{ id: 4, windowId: 100, active: true } as chrome.tabs.Tab]),
},
};
}
it("exposes the frame-aware observation result for recording", async () => {
const ax: CdpAxNode[] = [
{
nodeId: "root",
role: { type: "role", value: "RootWebArea" },
backendDOMNodeId: 11,
childIds: ["password"],
},
{
nodeId: "password",
parentId: "root",
role: { type: "role", value: "textbox" },
name: { type: "computedString", value: "Password" },
value: { value: "hunter2" },
backendDOMNodeId: 13,
},
];
const deps = makeOverlayDeps(ax, loginSnapshotReply(), VP_METRICS);
const result = await captureVomObservation(deps.cdp, 4, "https://example.com", {
redactValues: true,
});
expect(result.text).not.toContain("hunter2");
expect(result.text).toContain("•••");
expect(result.refs[0]).toMatchObject({
backendNodeId: 13,
role: "textbox",
name: "Password",
line: expect.any(Number),
});
expect(result.rootFrameId).toBe("root");
expect(result).not.toHaveProperty("frameDocuments");
expect(result.frames).toEqual([{ frameId: "root", target: { tabId: 4 } }]);
expect(result.matchNodes.find((node) => node.backendNodeId === 13)).toMatchObject({
frameId: "root",
backendNodeId: 13,
tag: "input",
rect: { x: 400, y: 300, w: 200, h: 40 },
localRect: { x: 400, y: 300, w: 200, h: 40 },
});
});
it("does not leak form secrets through the record-safe observation payload", async () => {
const secrets = {
email: "user@example.com",
password: "hunter2",
otp: "847291",
card: "4111111111111111",
};
const ax: CdpAxNode[] = [
{
nodeId: "root",
role: { type: "role", value: "RootWebArea" },
backendDOMNodeId: 11,
childIds: ["email", "password", "otp", "card"],
},
{
nodeId: "email",
parentId: "root",
role: { type: "role", value: "textbox" },
name: { type: "computedString", value: "Email" },
value: { value: secrets.email },
backendDOMNodeId: 12,
},
{
nodeId: "password",
parentId: "root",
role: { type: "role", value: "textbox" },
name: { type: "computedString", value: "Password" },
value: { value: secrets.password },
backendDOMNodeId: 13,
},
{
nodeId: "otp",
parentId: "root",
role: { type: "role", value: "textbox" },
name: { type: "computedString", value: "Code" },
value: { value: secrets.otp },
backendDOMNodeId: 14,
},
{
nodeId: "card",
parentId: "root",
role: { type: "role", value: "textbox" },
name: { type: "computedString", value: "Card" },
value: { value: secrets.card },
backendDOMNodeId: 15,
},
];
const deps = makeOverlayDeps(ax, secretsSnapshotReply(secrets), VP_METRICS);
const result = await captureVomObservation(deps.cdp, 4, "https://example.com", {
redactValues: true,
});
const dumped = JSON.stringify(result);
expect(result).not.toHaveProperty("frameDocuments");
expect(dumped).not.toContain(secrets.email);
expect(dumped).not.toContain(secrets.password);
expect(dumped).not.toContain(secrets.otp);
expect(dumped).not.toContain(secrets.card);
expect(dumped).not.toContain("formValue");
expect(dumped).not.toContain("formDefaultValue");
expect(dumped).not.toContain("axNodes");
expect(dumped).not.toContain("domNodes");
expect(dumped).not.toContain("attrs");
expect(result.text).toContain("•••");
});
it("keeps frames and rendered refs on the same frame identity", async () => {
const deps = makeFrameAwareDeps();
const result = await captureVomObservation(deps.cdp, 4, "https://example.com");
expect(result.rootFrameId).toBe("main");
expect(result.frames).toHaveLength(2);
expect(result.frames.find((frame) => frame.frameId === "child")).toEqual({
frameId: "child",
parentFrameId: "main",
ownerBackendNodeId: 12,
target: { tabId: 4, sessionId: "child-session" },
});
expect(result.matchNodes.find((node) => node.backendNodeId === 22)).toMatchObject({
frameId: "child",
tag: "button",
localRect: { x: 20, y: 30, w: 120, h: 40 },
});
expect(result.refs.find((ref) => ref.backendNodeId === 22)).toMatchObject({
frameId: "child",
name: "Frame action",
});
});
it("stores iframe refs with their owning CDP session", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
const deps = makeFrameAwareDeps();
const result = await handleSnapshot(sm, { session_id: "aa11" }, deps);
if ("code" in result) throw new Error(`unexpected error: ${JSON.stringify(result)}`);
expect(result.text).toContain('@e1 button "Frame action"');
const frameRef = [...ctx.refStore.entries()].find(([, entry]) => entry.backendNodeId === 22);
expect(frameRef?.[1]).toMatchObject({
tabId: 4,
frameId: "child",
cdpSessionId: "child-session",
});
});
it("renders a blocking login overlay as the focused top layer", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
await sm.start("aa11");
@@ -3773,32 +3069,6 @@ describe("handleGetHtml", () => {
expect(deps.send).toHaveBeenCalledTimes(1);
});
it("reads a frame ref through its child CDP session", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.refStore.set("e7", 4242, {
tabId: 4,
frameId: "child-frame",
cdpSessionId: "child-session",
});
const deps = makeDeps({});
const sendToTarget = vi.fn(async (target, method, params) => {
expect(target).toEqual({ tabId: 4, sessionId: "child-session" });
expect(method).toBe("DOM.getOuterHTML");
expect(params).toEqual({ backendNodeId: 4242 });
return { outerHTML: "<button>frame</button>" };
});
(deps.cdp as CdpRunner).sendToTarget = sendToTarget as unknown as NonNullable<
CdpRunner["sendToTarget"]
>;
const res = await handleGetHtml(sm, { session_id: "aa11", ref: "@e7" }, deps);
if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
expect(res.html).toBe("<button>frame</button>");
expect(deps.send).not.toHaveBeenCalled();
});
it("returns not_found when a ref belongs to another tab", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
File diff suppressed because it is too large Load Diff
@@ -60,58 +60,11 @@ function makeApis(
}),
getLastFocused: vi.fn(async () => ({ id: 500 }) as chrome.windows.Window),
create: vi.fn(async () => ({ id: 999 }) as chrome.windows.Window),
remove: vi.fn(async () => {}),
};
return { tabs, windows };
}
describe("handleSessionStop with auto-return", () => {
it("leaves the session untouched when cancellation arrived before teardown", async () => {
const aw = fakeAgentWindow([100]);
const sm = new SessionManager({ agentWindow: aw });
await sm.start("aa11");
const cdp = { detachSession: vi.fn(async () => {}) };
const controller = new AbortController();
controller.abort();
const res = await handleSessionStop(
sm,
{ session_id: "aa11" },
{ cdp, signal: controller.signal },
);
expect(res).toMatchObject({ code: "cancelled" });
expect(sm.has("aa11")).toBe(true);
expect(cdp.detachSession).not.toHaveBeenCalled();
expect(aw.remove).not.toHaveBeenCalled();
});
it("does not close the Agent Window when cancellation lands during CDP detach", async () => {
const aw = fakeAgentWindow([100]);
const sm = new SessionManager({ agentWindow: aw });
await sm.start("aa11");
const controller = new AbortController();
let finishDetach: () => void = () => {};
const detach = new Promise<void>((resolve) => {
finishDetach = resolve;
});
const cdp = { detachSession: vi.fn(() => detach) };
const stopping = handleSessionStop(
sm,
{ session_id: "aa11" },
{ cdp, signal: controller.signal },
);
await vi.waitFor(() => expect(cdp.detachSession).toHaveBeenCalledWith("aa11"));
controller.abort();
finishDetach();
const res = await stopping;
expect(res).toMatchObject({ code: "cancelled" });
expect(sm.has("aa11")).toBe(true);
expect(aw.remove).not.toHaveBeenCalled();
});
it("returns every borrowed tab and closes the Agent Window in the right order", async () => {
const aw = fakeAgentWindow([100]);
const sm = new SessionManager({ agentWindow: aw });
@@ -46,25 +46,6 @@ describe("lookupSnapshotRef", () => {
});
describe("resolveSnapshotRef", () => {
it("preserves the frame and child session needed to route iframe refs", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.refStore.set("e3", 1234, {
tabId: 4,
frameId: "child-frame",
cdpSessionId: "child-session",
});
const expected = {
backendNodeId: 1234,
refKey: "e3",
frameId: "child-frame",
cdpSessionId: "child-session",
};
expect(lookupSnapshotRef(ctx, "@e3", 4)).toEqual(expected);
expect(resolveSnapshotRef(ctx, "@e3", 4)).toEqual(expected);
});
it("returns not_found for unknown ref", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
@@ -179,7 +179,6 @@ function makeWindowsApi(
get: ReturnType<typeof vi.fn>;
lastFocused: ReturnType<typeof vi.fn>;
create: ReturnType<typeof vi.fn>;
remove: ReturnType<typeof vi.fn>;
};
} {
const get = vi.fn(async (windowId: number) => {
@@ -198,13 +197,7 @@ function makeWindowsApi(
const id = opts?.createWindowId ?? 999;
return { id } as chrome.windows.Window;
});
const remove = vi.fn(async (windowId: number) => {
state.windowsClosed.add(windowId);
});
return {
api: { get, getLastFocused: lastFocused, create, remove },
spies: { get, lastFocused, create, remove },
};
return { api: { get, getLastFocused: lastFocused, create }, spies: { get, lastFocused, create } };
}
describe("handleTabCreate", () => {
@@ -608,86 +601,6 @@ describe("handleTabReturn", () => {
expect(winSpies.create).toHaveBeenCalledOnce();
});
it("closes a newly created fallback window when cancellation wins before the move", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.borrowedTabs.set(7, { tabId: 7, originalWindowId: 200, originalIndex: 4 });
const state: FakeTabState = {
tabs: new Map([[7, { id: 7, windowId: 100 } as chrome.tabs.Tab]]),
nextTabId: 50,
windowsClosed: new Set([200]),
};
const { api, spies } = makeTabMutationApi(state);
const { api: windowsApi, spies: winSpies } = makeWindowsApi(state, {
lastFocused: 100,
createWindowId: 777,
});
let resolveCreate: (window: chrome.windows.Window) => void = () => {};
winSpies.create.mockImplementationOnce(
() =>
new Promise((resolve) => {
resolveCreate = resolve;
}),
);
const controller = new AbortController();
const pending = handleTabReturn(
sm,
{ session_id: "aa11", tab_id: 7 },
{ tabs: api, windows: windowsApi, signal: controller.signal },
);
await vi.waitFor(() => expect(winSpies.create).toHaveBeenCalledOnce());
controller.abort();
resolveCreate({ id: 777 } as chrome.windows.Window);
await expect(pending).resolves.toMatchObject({ code: "cancelled" });
expect(winSpies.remove).toHaveBeenCalledWith(777);
expect(spies.move).not.toHaveBeenCalled();
expect(ctx.borrowedTabs.has(7)).toBe(true);
});
it("surfaces the fallback window id when cancellation cleanup fails", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
ctx.borrowedTabs.set(7, { tabId: 7, originalWindowId: 200, originalIndex: 4 });
const state: FakeTabState = {
tabs: new Map([[7, { id: 7, windowId: 100 } as chrome.tabs.Tab]]),
nextTabId: 50,
windowsClosed: new Set([200]),
};
const { api, spies } = makeTabMutationApi(state);
const { api: windowsApi, spies: winSpies } = makeWindowsApi(state, {
lastFocused: 100,
createWindowId: 777,
});
let resolveCreate: (window: chrome.windows.Window) => void = () => {};
winSpies.create.mockImplementationOnce(
() =>
new Promise((resolve) => {
resolveCreate = resolve;
}),
);
winSpies.remove.mockRejectedValueOnce(new Error("window removal denied"));
const controller = new AbortController();
const pending = handleTabReturn(
sm,
{ session_id: "aa11", tab_id: 7 },
{ tabs: api, windows: windowsApi, signal: controller.signal },
);
await vi.waitFor(() => expect(winSpies.create).toHaveBeenCalledOnce());
controller.abort();
resolveCreate({ id: 777 } as chrome.windows.Window);
await expect(pending).resolves.toMatchObject({
code: "protocol_error",
data: { reason: "cleanup_failed", resource_type: "window", resource_id: 777 },
message: expect.stringMatching(/fallback window 777.*window removal denied/),
});
expect(spies.move).not.toHaveBeenCalled();
expect(ctx.borrowedTabs.has(7)).toBe(true);
});
it("falls back to a new window when getLastFocused only returns the Agent Window", async () => {
const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
const ctx = await sm.start("aa11");
@@ -250,10 +250,9 @@ const DEFAULT_NOTIFICATION_COPY: BorrowNotificationCopy = {
// URL injection-eligibility test
// ---------------------------------------------------------------------------
const EXTENSION_STORE_RES = [
const CHROME_WEB_STORE_RES = [
/^https:\/\/chrome\.google\.com\/webstore/i,
/^https:\/\/chromewebstore\.google\.com/i,
/^https:\/\/microsoftedge\.microsoft\.com\/addons/i,
];
/**
@@ -265,15 +264,14 @@ const EXTENSION_STORE_RES = [
* - `ftp://` no longer hosts content scripts reliably in modern Chrome.
* - `about:` / `chrome:` / `chrome-extension:` / `edge:` / `devtools:` /
* `view-source:` / `data:` / `blob:` are blocked by the platform.
* - Extension storefronts are blocked by their own browser: the Chrome Web
* Store (legacy and new domains) under Chrome, and Edge Add-ons under Edge.
* - The Chrome Web Store (both legacy and new domains) is also blocked.
*
* Mirrors the implicit scheme list documented in
* https://developer.chrome.com/docs/extensions/develop/concepts/match-patterns.
*/
export function isInjectableContentScriptUrl(url: string | undefined): boolean {
if (!url) return false;
for (const re of EXTENSION_STORE_RES) {
for (const re of CHROME_WEB_STORE_RES) {
if (re.test(url)) return false;
}
return /^https?:\/\//i.test(url);
@@ -1,9 +1,124 @@
export {
type CaptureVomObservationOptions,
captureVomObservation,
} from "./observation";
export {
type CaptureVomFrame,
type CaptureVomMatchNode,
type CaptureVomObservationResult,
} from "./vom/record-safe-observation";
import { type RenderedRef, renderVom, type VomOptions } from "@browser-skill/vom";
import { readRecordDocumentTokenFromAttrs } from "@/lib/record-document-token";
import type { CdpAxNode } from "./observation";
import { buildVomScene } from "./observation";
import type { CdpRunner } from "./shared";
import {
type CapturedNode,
type CapturedSurfaceProbe,
type CapturedViewModel,
captureViewModel,
collectOverlayExcludedBackendIds,
} from "./vom/capture";
export interface CaptureVomObservationOptions extends VomOptions {
conditionalSurfaceProbe?: boolean;
hoverProbeBypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>;
}
export interface CaptureVomObservationResult {
text: string;
refs: RenderedRef[];
truncated: boolean;
captured: CapturedNode[];
/** All nodes including iframe sub-documents, for geometric target matching. */
matchNodes: CapturedNode[];
/** Maps ephemeral record document tokens to immediate iframe owner backend ids. */
documentTokenOwners: Map<string, number | null>;
surfaceProbes?: CapturedSurfaceProbe[];
}
export function flattenCapturedNodes(vm: CapturedViewModel): CapturedNode[] {
return [...vm.nodes, ...[...vm.iframeNodes.values()].flat()];
}
export function collectDocumentTokenOwners(vm: CapturedViewModel): Map<string, number | null> {
const owners = new Map<string, number | null>();
for (const node of flattenCapturedNodes(vm)) {
if (node.tag !== "html") continue;
const token = readRecordDocumentTokenFromAttrs(node.attrs);
if (!token) continue;
owners.set(token, node.ownerFrameBackendNodeId ?? null);
}
return owners;
}
function emptyCapturedViewModel(viewport = { width: 0, height: 0 }): CapturedViewModel {
return { viewport, nodes: [], iframeNodes: new Map(), excludedBackendNodeIds: new Set() };
}
interface LayoutMetricsViewportReply {
cssLayoutViewport?: { clientWidth?: number; clientHeight?: number };
layoutViewport?: { clientWidth?: number; clientHeight?: number };
}
async function fallbackCapturedViewModel(
cdp: CdpRunner,
tabId: number,
): Promise<CapturedViewModel> {
let viewport = { width: 0, height: 0 };
try {
const metrics = await cdp.send<LayoutMetricsViewportReply>(tabId, "Page.getLayoutMetrics", {});
const vpSrc = metrics.cssLayoutViewport ?? metrics.layoutViewport ?? {};
viewport = {
width: vpSrc.clientWidth ?? 0,
height: vpSrc.clientHeight ?? 0,
};
} catch {
// viewport stays zero-sized
}
const excludedBackendNodeIds = await collectOverlayExcludedBackendIds(cdp, tabId);
return { ...emptyCapturedViewModel(viewport), excludedBackendNodeIds };
}
async function captureForVom(
cdp: CdpRunner,
tabId: number,
conditionalSurfaceProbe: boolean,
hoverProbeBypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>,
): Promise<CapturedViewModel> {
try {
return await captureViewModel(cdp, tabId, {
conditionalSurfaceProbe,
hoverProbeBypassOverlay,
});
} catch {
return fallbackCapturedViewModel(cdp, tabId);
}
}
/** Pure VOM capture pipeline shared by snapshot, observe, and recording. */
export async function captureVomObservation(
cdp: CdpRunner,
tabId: number,
url: string | undefined,
opts: CaptureVomObservationOptions = {},
): Promise<CaptureVomObservationResult> {
await cdp.ensureAttachedToUrl?.(tabId, url);
await cdp.send<unknown>(tabId, "Accessibility.enable", {});
const result = await cdp.send<{ nodes: CdpAxNode[] }>(tabId, "Accessibility.getFullAXTree", {});
const axNodes = result.nodes ?? [];
const captured = await captureForVom(
cdp,
tabId,
opts.conditionalSurfaceProbe ?? false,
opts.hoverProbeBypassOverlay,
);
const scene = buildVomScene(axNodes, captured, { pageUrl: url });
const rendered = renderVom(scene, {
maxDepth: opts.maxDepth,
maxTokens: opts.maxTokens,
redactValues: opts.redactValues,
activeRegionPolicy: opts.activeRegionPolicy ?? true,
});
const matchNodes = flattenCapturedNodes(captured);
return {
text: rendered.text,
refs: rendered.refs,
truncated: rendered.truncated,
captured: captured.nodes,
matchNodes,
documentTokenOwners: collectDocumentTokenOwners(captured),
surfaceProbes: captured.surfaceProbes,
};
}
-4
View File
@@ -27,23 +27,19 @@ export async function handleConsole(
manager: SessionManager,
params: ConsoleParams,
deps: ConsoleDeps = defaultConsoleDeps(),
signal?: AbortSignal,
): Promise<ConsoleResult | RpcError> {
if (signal?.aborted) return { code: "cancelled", message: "console aborted" };
const ctxOrErr = lookupSession(manager, params, "console");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const bounds = parseBufferedReadBounds(params);
if (isRpcError(bounds)) return bounds;
const target = await resolveTargetTab(manager, ctxOrErr, params.tab_id, deps.tabsApi);
if (isRpcError(target)) return target;
if (signal?.aborted) return { code: "cancelled", message: "console aborted" };
if (!deps.cdp.ensureConsoleCapture || !deps.cdp.consoleEntriesSince) {
return { code: "cdp_failed", message: "console capture requires CDP console support" };
}
try {
await deps.cdp.ensureConsoleCapture(target.tabId);
if (signal?.aborted) return { code: "cancelled", message: "console aborted" };
deps.cdp.trackSessionTab?.(ctxOrErr.sessionId, target.tabId);
return deps.cdp.consoleEntriesSince(
target.tabId,
+108 -133
View File
@@ -125,10 +125,11 @@ export interface DispatcherDeps {
* `AbortController` keyed by its wire `id` in
* [`inflightAbortControllers`]. When the daemon pushes a `cancel`
* request the dispatcher trips the matching controller; tool
* handlers observe that signal between awaited operations. The
* original RPC remains pending until its handler has stopped or
* completed compensation; only the separate cancel acknowledgement
* takes the fast path.
* handlers that already accept a `signal` (waits, navigation,
* interaction, evaluate, tabs) react in line, and the dispatcher
* additionally races the in-flight invocation against the abort
* promise so handlers without explicit signal plumbing still respond
* promptly with `cancelled`.
*/
export class ToolDispatcher {
private readonly transport: Transport;
@@ -220,7 +221,7 @@ export class ToolDispatcher {
try {
const sessionId = sessionIdForBrowserControlMethod(req);
if (sessionId) this.onBrowserControlResumed?.(sessionId);
const result = await this.invoke(req, ac.signal);
const result = await Promise.race([this.invoke(req, ac.signal), abortPromise(ac.signal)]);
if (isRpcError(result)) {
body = { id: req.id, error: result };
} else {
@@ -285,45 +286,37 @@ export class ToolDispatcher {
private async invoke(req: RequestFrame, signal: AbortSignal): Promise<unknown | RpcError> {
switch (req.method) {
case "tool.session_start":
return handleSessionStart(this.sessions, req.params as SessionStartParams, { signal });
return handleSessionStart(this.sessions, req.params as SessionStartParams);
case "tool.session_stop": {
await this.releaseHoverLatch((req.params as SessionStopParams).session_id);
return handleSessionStop(this.sessions, req.params as SessionStopParams, {
cdp: this.cdp,
signal,
});
}
case "tool.tab_list":
return handleTabList(this.sessions, req.params as TabListParams, chromeTabsApi, signal);
return handleTabList(this.sessions, req.params as TabListParams);
case "tool.tab_create":
return handleTabCreate(this.sessions, req.params as TabCreateParams, { signal });
return handleTabCreate(this.sessions, req.params as TabCreateParams);
case "tool.tab_close":
return this.withHoverReleaseForRequest(
req.params as TabCloseParams,
() => handleTabClose(this.sessions, req.params as TabCloseParams, { signal }),
signal,
return this.withHoverReleaseForRequest(req.params as TabCloseParams, () =>
handleTabClose(this.sessions, req.params as TabCloseParams),
);
case "tool.tab_select":
return handleTabSelect(this.sessions, req.params as TabSelectParams, { signal });
return handleTabSelect(this.sessions, req.params as TabSelectParams);
case "tool.tab_borrow":
return handleTabBorrow(this.sessions, req.params as TabBorrowParams, {
signal,
approveBorrow: this.approveBorrow,
});
case "tool.tab_return":
return handleTabReturn(this.sessions, req.params as TabReturnParams, { signal });
return handleTabReturn(this.sessions, req.params as TabReturnParams);
case "tool.window_resize":
return handleWindowResize(
this.sessions,
req.params as WindowResizeParams,
undefined,
signal,
);
return handleWindowResize(this.sessions, req.params as WindowResizeParams);
case "tool.emulate":
return handleEmulate(
this.sessions,
req.params as EmulateParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi } : undefined,
);
case "tool.screenshot":
return handleScreenshot(
@@ -332,57 +325,43 @@ export class ToolDispatcher {
this.cdp
? { cdp: this.cdp, tabsApi: chromeTabsCaptureApi, captureApi: chromeTabsCaptureApi }
: undefined,
signal,
);
case "tool.console":
return handleConsole(
this.sessions,
req.params as ConsoleParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi } : undefined,
signal,
);
case "tool.network":
return handleNetwork(
this.sessions,
req.params as NetworkParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi } : undefined,
signal,
);
case "tool.snapshot":
return this.withHoverReassert(
req.params as SnapshotParams,
() =>
handleSnapshot(
this.sessions,
req.params as SnapshotParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsCaptureApi } : undefined,
signal,
),
{},
signal,
return this.withHoverReassert(req.params as SnapshotParams, () =>
handleSnapshot(
this.sessions,
req.params as SnapshotParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsCaptureApi } : undefined,
),
);
case "tool.observe": {
const params = req.params as ObserveParams;
const hoverScope = await this.resolveHoverLatchScope(params);
throwIfDispatchAborted(signal);
return this.withHoverReassert(
params,
() =>
handleObserve(
this.sessions,
params,
this.cdp
? {
cdp: this.cdp,
tabsApi: chromeTabsCaptureApi,
conditionalSurfaceProbe: !this.hasHoverLatchForScope(hoverScope),
hoverProbeBypassOverlay: bypassOverlay,
}
: undefined,
signal,
),
{},
signal,
return this.withHoverReassert(params, () =>
handleObserve(
this.sessions,
params,
this.cdp
? {
cdp: this.cdp,
tabsApi: chromeTabsCaptureApi,
conditionalSurfaceProbe: !this.hasHoverLatchForScope(hoverScope),
hoverProbeBypassOverlay: bypassOverlay,
}
: undefined,
),
);
}
case "tool.get_html":
@@ -390,51 +369,38 @@ export class ToolDispatcher {
this.sessions,
req.params as GetHtmlParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsCaptureApi } : undefined,
signal,
);
case "tool.navigate":
return this.withHoverReleaseForRequest(
req.params as NavigateParams,
() =>
handleNavigate(
this.sessions,
req.params as NavigateParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as NavigateParams, () =>
handleNavigate(
this.sessions,
req.params as NavigateParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.navigate_back":
return this.withHoverReleaseForRequest(
req.params as NavigateBackParams,
() =>
handleNavigateBack(
this.sessions,
req.params as NavigateBackParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as NavigateBackParams, () =>
handleNavigateBack(
this.sessions,
req.params as NavigateBackParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.navigate_forward":
return this.withHoverReleaseForRequest(
req.params as NavigateForwardParams,
() =>
handleNavigateForward(
this.sessions,
req.params as NavigateForwardParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as NavigateForwardParams, () =>
handleNavigateForward(
this.sessions,
req.params as NavigateForwardParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.reload":
return this.withHoverReleaseForRequest(
req.params as ReloadParams,
() =>
handleReload(
this.sessions,
req.params as ReloadParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as ReloadParams, () =>
handleReload(
this.sessions,
req.params as ReloadParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.click":
return this.withHoverReassert(
@@ -453,7 +419,6 @@ export class ToolDispatcher {
: undefined,
),
{ releaseAfter: true },
signal,
);
case "tool.hover": {
const result = await handleHover(
@@ -473,37 +438,28 @@ export class ToolDispatcher {
return this.rememberHover((req.params as HoverParams).session_id, result);
}
case "tool.fill":
return this.withHoverReleaseForRequest(
req.params as FillParams,
() =>
handleFill(
this.sessions,
req.params as FillParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as FillParams, () =>
handleFill(
this.sessions,
req.params as FillParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.press":
return this.withHoverReleaseForRequest(
req.params as PressParams,
() =>
handlePress(
this.sessions,
req.params as PressParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as PressParams, () =>
handlePress(
this.sessions,
req.params as PressParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.select":
return this.withHoverReleaseForRequest(
req.params as SelectParams,
() =>
handleSelect(
this.sessions,
req.params as SelectParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
signal,
return this.withHoverReleaseForRequest(req.params as SelectParams, () =>
handleSelect(
this.sessions,
req.params as SelectParams,
this.cdp ? { cdp: this.cdp, tabsApi: chromeTabsApi, signal } : undefined,
),
);
case "tool.evaluate":
return handleEvaluate(
@@ -610,13 +566,9 @@ export class ToolDispatcher {
params: { session_id: string; tab_id?: number },
work: () => Promise<T>,
options: { releaseAfter?: boolean } = {},
signal?: AbortSignal,
): Promise<T> {
throwIfDispatchAborted(signal);
const scope = await this.resolveHoverLatchScope(params);
throwIfDispatchAborted(signal);
await this.reassertHover(scope);
throwIfDispatchAborted(signal);
try {
return await work();
} finally {
@@ -629,13 +581,9 @@ export class ToolDispatcher {
private async withHoverReleaseForRequest<T>(
params: { session_id: string; tab_id?: number },
work: () => Promise<T>,
signal?: AbortSignal,
): Promise<T> {
throwIfDispatchAborted(signal);
const scope = await this.resolveHoverLatchScope(params);
throwIfDispatchAborted(signal);
await this.releaseHoverLatch(scope.session_id, scope.tab_id);
throwIfDispatchAborted(signal);
return work();
}
@@ -745,14 +693,41 @@ async function bypassOverlay(tabId: number, enabled: boolean): Promise<void> {
}
}
function throwIfDispatchAborted(signal: AbortSignal | undefined): void {
if (!signal?.aborted) return;
const error = new Error("rpc aborted by daemon cancel");
error.name = "AbortError";
throw error;
/**
* Resolves never; rejects with `AbortLikeError` as soon as the signal
* fires (or immediately if it is already aborted). Used by the
* dispatcher to race the tool invocation so handlers without explicit
* signal plumbing still surface a `cancelled` reply promptly.
*/
function abortPromise(signal: AbortSignal): Promise<never> {
return new Promise<never>((_, reject) => {
if (signal.aborted) {
reject(new AbortLikeError());
return;
}
signal.addEventListener(
"abort",
() => {
reject(new AbortLikeError());
},
{ once: true },
);
});
}
/**
* Sentinel error class so [`isAbortLikeError`] can recognise our own
* race-rejection without confusing it with a real CDP failure.
*/
class AbortLikeError extends Error {
constructor() {
super("rpc aborted by daemon cancel");
this.name = "BhAbortError";
}
}
function isAbortLikeError(err: unknown): boolean {
if (err instanceof AbortLikeError) return true;
if (err instanceof DOMException && err.name === "AbortError") return true;
if (typeof err === "object" && err !== null && (err as { name?: string }).name === "AbortError") {
return true;
+96 -41
View File
@@ -1,15 +1,9 @@
// Resolve and scroll nodes inside one CDP target. Cross-frame projection
// belongs to frame-geometry.ts.
// Shared CDP helpers for scrolling nodes into view and computing
// viewport-space bounding rectangles. Used by interaction tools (click
// centre) and observation tools (element screenshot clip).
import type { RpcError } from "@/transport/types";
import { rpcError } from "./errors";
import {
parseCdpQuad,
polygonBounds,
type Region,
rectPolygon,
type ViewportRect,
} from "./geometry";
import { type CdpRunner, isRpcError } from "./shared";
const ELEMENT_NOT_VISIBLE_MESSAGE =
@@ -19,6 +13,45 @@ function elementNotVisibleError(): RpcError {
return rpcError("permission_denied", "element_not_visible", ELEMENT_NOT_VISIBLE_MESSAGE);
}
/** Viewport-space axis-aligned bounding box for CDP `Page.captureScreenshot` clip. */
export interface ViewportRect {
x: number;
y: number;
width: number;
height: number;
}
/**
* Compute the axis-aligned bounding box of an 8-double CDP content
* quad / box-model polygon. Exported for unit tests.
*/
export function quadBoundingRect(quad: number[]): ViewportRect | null {
if (quad.length !== 8) return null;
const xs = [quad[0], quad[2], quad[4], quad[6]];
const ys = [quad[1], quad[3], quad[5], quad[7]];
const minX = Math.min(...xs);
const maxX = Math.max(...xs);
const minY = Math.min(...ys);
const maxY = Math.max(...ys);
const width = maxX - minX;
const height = maxY - minY;
if (width <= 0 || height <= 0) return null;
return { x: minX, y: minY, width, height };
}
function rectToQuad(rect: ViewportRect): number[] {
return [
rect.x,
rect.y,
rect.x + rect.width,
rect.y,
rect.x + rect.width,
rect.y + rect.height,
rect.x,
rect.y + rect.height,
];
}
/**
* Resolve `backendNodeId` → CDP `objectId` so we can invoke
* `Runtime.callFunctionOn` against the live JS object.
@@ -79,24 +112,48 @@ export async function scrollNodeIntoView(
}
}
function validCdpRegion(quads: number[][] | undefined): Region {
return (quads ?? []).flatMap((raw) => {
const polygon = parseCdpQuad(raw);
return polygon && polygonBounds(polygon) ? [polygon] : [];
});
}
async function resolveVisibleContentRegion(
/**
* Viewport-space bounding box for `backendNodeId`. Prefers
* `DOM.getContentQuads`, then `DOM.getBoxModel`, then the union of
* visible descendant client rects for zero-size overflow containers.
*/
export async function nodeBoundingRect(
cdp: CdpRunner,
tabId: number,
backendNodeId: number,
): Promise<Region | RpcError> {
): Promise<ViewportRect | RpcError> {
const polygon = await visibleContentPolygon(cdp, tabId, backendNodeId);
if (isRpcError(polygon)) return polygon;
// visibleContentPolygon only returns quads with positive area.
return quadBoundingRect(polygon)!;
}
/**
* Compute the centroid of a CDP content quad. Quads are reported as
* 8 doubles in clockwise order: (x1, y1, x2, y2, x3, y3, x4, y4).
*/
export function quadCentre(quad: number[]): { x: number; y: number } {
const x = (quad[0] + quad[4]) / 2;
const y = (quad[1] + quad[5]) / 2;
return { x, y };
}
/** Centre of a `DOM.getBoxModel` `content` polygon (also 8 doubles). */
export function boxCentre(box: number[]): { x: number; y: number } {
return quadCentre(box);
}
async function visibleContentPolygon(
cdp: CdpRunner,
tabId: number,
backendNodeId: number,
): Promise<number[] | RpcError> {
try {
const quads = await cdp.send<{ quads?: number[][] }>(tabId, "DOM.getContentQuads", {
backendNodeId,
});
const visible = validCdpRegion(quads.quads);
if (visible.length > 0) return visible;
const quad = quads.quads?.find((q) => q.length === 8 && quadBoundingRect(q) !== null);
if (quad) return quad;
} catch (err) {
console.debug("[bsk element-geometry] getContentQuads failed", err);
}
@@ -105,34 +162,15 @@ async function resolveVisibleContentRegion(
backendNodeId,
});
const content = box.model?.content;
const fallback = validCdpRegion(content ? [content] : undefined);
if (fallback.length > 0) return fallback;
if (content && content.length === 8 && quadBoundingRect(content)) return content;
} catch (err) {
console.debug("[bsk element-geometry] getBoxModel failed", err);
}
const descendantRect = await descendantBoundingRect(cdp, tabId, backendNodeId);
if (descendantRect) {
return [
rectPolygon({
x: descendantRect.x,
y: descendantRect.y,
w: descendantRect.width,
h: descendantRect.height,
}),
];
}
if (descendantRect) return rectToQuad(descendantRect);
return elementNotVisibleError();
}
/** Resolve every visible content region in the node's target-local viewport. */
export async function nodeContentRegion(
cdp: CdpRunner,
tabId: number,
backendNodeId: number,
): Promise<Region | RpcError> {
return resolveVisibleContentRegion(cdp, tabId, backendNodeId);
}
async function descendantBoundingRect(
cdp: CdpRunner,
tabId: number,
@@ -203,3 +241,20 @@ function parseViewportRect(value: unknown): ViewportRect | null {
if (![x, y, width, height].every(Number.isFinite) || width <= 0 || height <= 0) return null;
return { x, y, width, height };
}
/**
* Get the click point for `backendNodeId`. Prefers
* `DOM.getContentQuads` (richer for rotated / transformed elements);
* falls back to `DOM.getBoxModel`, then visible descendant bounds.
* Returns `permission_denied` with "element not visible" when every
* geometry path comes back empty / fails.
*/
export async function nodeCentre(
cdp: CdpRunner,
tabId: number,
backendNodeId: number,
): Promise<{ x: number; y: number } | RpcError> {
const polygon = await visibleContentPolygon(cdp, tabId, backendNodeId);
if (isRpcError(polygon)) return polygon;
return quadCentre(polygon);
}
-17
View File
@@ -44,7 +44,6 @@ export interface EmulateCdpRunner {
export interface EmulateDeps {
cdp: EmulateCdpRunner;
tabsApi: ChromeTabsApi;
signal?: AbortSignal;
}
function defaultEmulateDeps(): EmulateDeps {
@@ -211,13 +210,11 @@ export async function handleEmulate(
params: EmulateParams,
deps: EmulateDeps = defaultEmulateDeps(),
): Promise<EmulateResult | RpcError> {
if (deps.signal?.aborted) return { code: "cancelled", message: "emulate aborted" };
const ctxOrErr = lookupSession(manager, params, "emulate");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const ctx = ctxOrErr;
const target = await resolveTargetTab(manager, ctx, params?.tab_id, deps.tabsApi);
if (isRpcError(target)) return target;
if (deps.signal?.aborted) return { code: "cancelled", message: "emulate aborted" };
const denied = enforceAgentWindow(ctx, target, "emulate");
if (denied) return denied;
@@ -226,16 +223,9 @@ export async function handleEmulate(
return invalidParams("emulate off cannot be combined with overrides");
}
try {
if (deps.signal?.aborted) return { code: "cancelled", message: "emulate aborted" };
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
await deps.cdp.clearDeviceMetricsOverride(target.tabId);
if (deps.signal?.aborted) {
return { code: "cancelled", message: "emulate aborted after clearing device metrics" };
}
await deps.cdp.setTouchEmulationEnabled(target.tabId, false);
if (deps.signal?.aborted) {
return { code: "cancelled", message: "emulate aborted after clearing touch emulation" };
}
// An empty userAgent string clears the override (CDP convention).
await deps.cdp.setUserAgentOverride(target.tabId, { userAgent: "" });
} catch (err) {
@@ -256,7 +246,6 @@ export async function handleEmulate(
// values; the merged state is what gets applied (and echoed back).
const merged = mergeEmulateOverrides(tabEmulationStates.get(target.tabId), overrides);
try {
if (deps.signal?.aborted) return { code: "cancelled", message: "emulate aborted" };
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
if (merged.width !== undefined && merged.height !== undefined) {
await deps.cdp.setDeviceMetricsOverride(target.tabId, {
@@ -265,9 +254,6 @@ export async function handleEmulate(
deviceScaleFactor: merged.device_scale_factor ?? 0,
mobile: merged.mobile ?? false,
});
if (deps.signal?.aborted) {
return { code: "cancelled", message: "emulate aborted after applying device metrics" };
}
}
if (merged.user_agent !== undefined) {
await deps.cdp.setUserAgentOverride(target.tabId, {
@@ -277,9 +263,6 @@ export async function handleEmulate(
? { userAgentMetadata: toCdpUserAgentMetadata(merged.user_agent_metadata) }
: {}),
});
if (deps.signal?.aborted) {
return { code: "cancelled", message: "emulate aborted after applying user agent" };
}
}
if (merged.touch !== undefined || merged.max_touch_points !== undefined) {
await deps.cdp.setTouchEmulationEnabled(
-290
View File
@@ -1,290 +0,0 @@
import { type CdpFrame, type CdpFrameGraph, type CdpTarget } from "@/browser-driver/frame-graph";
import type { RpcError } from "@/transport/types";
import { nodeContentRegion, scrollNodeIntoView } from "./element-geometry";
import {
clipPolygon,
type GeometryProjection,
type Point,
type Polygon,
type ProjectiveEdge,
parseCdpQuad,
polygonArea,
polygonCentroid,
projectRegionToViewport,
type Quad,
type Region,
rectPolygon,
regionBounds,
type Size,
type ViewportRect,
} from "./geometry";
import { type CdpRunner, cdpRunnerForTarget, isRpcError, sendToCdpTarget } from "./shared";
export interface NodeAddress {
target: CdpTarget;
backendNodeId: number;
frameId?: string;
}
export interface ResolvedNodeGeometry {
topVisibleRegions: Region;
topBounds: ViewportRect;
actionPoint: Point;
}
function geometryError(message: string): RpcError {
return { code: "cdp_failed", message };
}
function frameMap(graph: CdpFrameGraph): Map<string, CdpFrame> {
return new Map(graph.frames.map((frame) => [frame.frameId, frame]));
}
function targetRootFrame(byId: Map<string, CdpFrame>, frame: CdpFrame): CdpFrame | null {
const seen = new Set<string>();
let current = frame;
while (current.parentFrameId) {
if (seen.has(current.frameId)) return null;
seen.add(current.frameId);
const parent = byId.get(current.parentFrameId);
if (!parent || parent.target.sessionId !== current.target.sessionId) break;
current = parent;
}
return current;
}
function frameAncestry(graph: CdpFrameGraph, frameId: string): CdpFrame[] | null {
const byId = frameMap(graph);
const path: CdpFrame[] = [];
const seen = new Set<string>();
let current = byId.get(frameId);
if (!current) return null;
while (current.parentFrameId) {
if (seen.has(current.frameId)) return null;
seen.add(current.frameId);
path.push(current);
const parent = byId.get(current.parentFrameId);
if (!parent) return null;
current = parent;
}
return path;
}
async function targetViewport(cdp: CdpRunner, target: CdpTarget): Promise<Size | null> {
const metrics = await sendToCdpTarget<{
cssLayoutViewport?: { clientWidth?: number; clientHeight?: number };
layoutViewport?: { clientWidth?: number; clientHeight?: number };
}>(cdp, target, "Page.getLayoutMetrics", {});
const viewport = metrics.cssLayoutViewport ?? metrics.layoutViewport ?? {};
const width = viewport.clientWidth ?? 0;
const height = viewport.clientHeight ?? 0;
return width > 0 && height > 0 ? { width, height } : null;
}
async function ownerContentQuad(
cdp: CdpRunner,
parent: CdpFrame,
ownerBackendNodeId: number,
): Promise<Quad | null> {
const result = await sendToCdpTarget<{ model?: { content?: number[] } }>(
cdp,
parent.target,
"DOM.getBoxModel",
{ backendNodeId: ownerBackendNodeId },
);
return parseCdpQuad(result.model?.content);
}
async function sameTargetFrameClips(
cdp: CdpRunner,
byId: Map<string, CdpFrame>,
frame: CdpFrame,
targetRoot: CdpFrame,
): Promise<Polygon[] | null> {
const clips: Polygon[] = [];
let current = frame;
const seen = new Set<string>();
while (current.frameId !== targetRoot.frameId) {
if (seen.has(current.frameId) || !current.parentFrameId) return null;
seen.add(current.frameId);
const parent = byId.get(current.parentFrameId);
if (!parent || current.ownerBackendNodeId === undefined) return null;
const clip = await ownerContentQuad(cdp, parent, current.ownerBackendNodeId);
if (!clip) return null;
clips.push(clip);
current = parent;
}
return clips;
}
export async function resolveFrameProjection(
cdp: CdpRunner,
graph: CdpFrameGraph,
frameId: string,
): Promise<GeometryProjection | null> {
const byId = frameMap(graph);
const frame = byId.get(frameId);
if (!frame) return null;
let targetRoot = targetRootFrame(byId, frame);
if (!targetRoot) return null;
const sourceViewport = await targetViewport(cdp, targetRoot.target);
if (!sourceViewport) return null;
const sourceClips = await sameTargetFrameClips(cdp, byId, frame, targetRoot);
if (!sourceClips) return null;
const edges: ProjectiveEdge[] = [];
while (targetRoot.parentFrameId) {
const parent = byId.get(targetRoot.parentFrameId);
if (!parent || targetRoot.ownerBackendNodeId === undefined) return null;
const destinationQuad = await ownerContentQuad(cdp, parent, targetRoot.ownerBackendNodeId);
if (!destinationQuad) return null;
const source =
edges.length === 0 ? sourceViewport : await targetViewport(cdp, targetRoot.target);
if (!source) return null;
const parentTargetRoot = targetRootFrame(byId, parent);
if (!parentTargetRoot) return null;
const destinationClips = await sameTargetFrameClips(cdp, byId, parent, parentTargetRoot);
if (!destinationClips) return null;
edges.push({ sourceViewport: source, destinationQuad, destinationClips });
targetRoot = parentTargetRoot;
}
const topViewport =
edges.length === 0 ? sourceViewport : await targetViewport(cdp, targetRoot.target);
return topViewport ? { sourceClips, edges, topViewport } : null;
}
async function loadFrameGraph(cdp: CdpRunner, tabId: number): Promise<CdpFrameGraph | null> {
if (!cdp.getFrameGraph) return null;
try {
return await cdp.getFrameGraph(tabId);
} catch (error) {
console.debug("[bsk frame-geometry] frame graph resolution failed", error);
return null;
}
}
async function scrollFrameOwners(
cdp: CdpRunner,
tabId: number,
graph: CdpFrameGraph,
frameId: string,
): Promise<RpcError | null> {
const byId = frameMap(graph);
const ancestry = frameAncestry(graph, frameId);
if (!ancestry) return geometryError(`could not resolve frame ancestry for ${frameId}`);
for (const child of [...ancestry].reverse()) {
const parent = child.parentFrameId ? byId.get(child.parentFrameId) : undefined;
if (!parent || child.ownerBackendNodeId === undefined) {
return geometryError(`could not resolve frame owner for ${child.frameId}`);
}
const error = await scrollNodeIntoView(
cdpRunnerForTarget(cdp, parent.target),
tabId,
child.ownerBackendNodeId,
);
if (error) return error;
}
return null;
}
async function scrollElementWithFrameGraph(
cdp: CdpRunner,
tabId: number,
target: CdpTarget,
backendNodeId: number,
frameId: string | undefined,
graph: CdpFrameGraph | null,
): Promise<RpcError | null> {
if (frameId) {
if (!graph) return geometryError(`could not resolve frame graph for ${frameId}`);
const error = await scrollFrameOwners(cdp, tabId, graph, frameId);
if (error) return error;
} else if (target.sessionId) {
return geometryError("an OOPIF node address requires frameId");
}
return scrollNodeIntoView(cdpRunnerForTarget(cdp, target), tabId, backendNodeId);
}
export async function scrollElementAndFramesIntoView(
cdp: CdpRunner,
tabId: number,
target: CdpTarget,
backendNodeId: number,
frameId?: string,
): Promise<RpcError | null> {
const graph = frameId ? await loadFrameGraph(cdp, tabId) : null;
return scrollElementWithFrameGraph(cdp, tabId, target, backendNodeId, frameId, graph);
}
function largestRegion(regions: Region): Polygon | null {
let largest: { polygon: Polygon; area: number } | null = null;
for (const polygon of regions) {
const area = polygonArea(polygon);
if (area <= 0) continue;
if (!largest || area > largest.area) largest = { polygon, area };
}
return largest?.polygon ?? null;
}
export async function resolveNodeGeometry(
cdp: CdpRunner,
tabId: number,
address: NodeAddress,
options: { scrollIntoView?: boolean } = {},
): Promise<ResolvedNodeGeometry | RpcError> {
try {
if (address.target.sessionId && !address.frameId) {
return geometryError("an OOPIF node address requires frameId");
}
const graph = address.frameId ? await loadFrameGraph(cdp, tabId) : null;
if (address.frameId && !graph) {
return geometryError(`could not resolve frame graph for ${address.frameId}`);
}
if (options.scrollIntoView) {
const scrollError = await scrollElementWithFrameGraph(
cdp,
tabId,
address.target,
address.backendNodeId,
address.frameId,
graph,
);
if (scrollError) return scrollError;
}
const localRegion = await nodeContentRegion(
cdpRunnerForTarget(cdp, address.target),
tabId,
address.backendNodeId,
);
if (isRpcError(localRegion)) return localRegion;
let topVisibleRegions: Region;
if (address.frameId && graph) {
const projection = await resolveFrameProjection(cdp, graph, address.frameId);
if (!projection)
return geometryError(`could not resolve frame geometry for ${address.frameId}`);
topVisibleRegions = projectRegionToViewport(localRegion, projection);
} else {
const viewport = await targetViewport(cdp, address.target);
if (!viewport) return geometryError("could not resolve top viewport geometry");
topVisibleRegions = localRegion
.map((polygon) =>
clipPolygon(polygon, rectPolygon({ x: 0, y: 0, w: viewport.width, h: viewport.height })),
)
.filter((polygon) => polygon.length >= 3);
}
const topBounds = regionBounds(topVisibleRegions);
const actionRegion = largestRegion(topVisibleRegions);
const actionPoint = actionRegion ? polygonCentroid(actionRegion) : null;
if (!topBounds || !actionPoint) {
return { code: "permission_denied", message: "element not visible" };
}
return { topVisibleRegions, topBounds, actionPoint };
} catch (error) {
return geometryError(error instanceof Error ? error.message : String(error));
}
}
-266
View File
@@ -1,266 +0,0 @@
export interface Point {
x: number;
y: number;
}
export type Polygon = Point[];
export type Region = Polygon[];
export type Quad = [Point, Point, Point, Point];
export interface Size {
width: number;
height: number;
}
export interface ViewportRect {
x: number;
y: number;
width: number;
height: number;
}
export interface ProjectiveEdge {
sourceViewport: Size;
destinationQuad: Quad;
destinationClips?: Polygon[];
}
export interface GeometryProjection {
sourceClips: Polygon[];
edges: ProjectiveEdge[];
topViewport: Size;
}
export function parseCdpQuad(raw: number[] | undefined): Quad | null {
if (!raw || raw.length !== 8 || !raw.every(Number.isFinite)) return null;
return [
{ x: raw[0], y: raw[1] },
{ x: raw[2], y: raw[3] },
{ x: raw[4], y: raw[5] },
{ x: raw[6], y: raw[7] },
];
}
export function rectPolygon(rect: { x: number; y: number; w: number; h: number }): Quad {
return [
{ x: rect.x, y: rect.y },
{ x: rect.x + rect.w, y: rect.y },
{ x: rect.x + rect.w, y: rect.y + rect.h },
{ x: rect.x, y: rect.y + rect.h },
];
}
export function viewportPolygon(viewport: Size): Quad {
return rectPolygon({ x: 0, y: 0, w: viewport.width, h: viewport.height });
}
/** Project a normalized unit-square point into an arbitrary content quad. */
export function projectUnitPoint(point: Point, quad: Quad): Point {
const [p0, p1, p2, p3] = quad;
const dx1 = p1.x - p2.x;
const dx2 = p3.x - p2.x;
const dx3 = p0.x - p1.x + p2.x - p3.x;
const dy1 = p1.y - p2.y;
const dy2 = p3.y - p2.y;
const dy3 = p0.y - p1.y + p2.y - p3.y;
let a: number;
let b: number;
let c: number;
let d: number;
let e: number;
let f: number;
let g = 0;
let h = 0;
if (Math.abs(dx3) < 1e-9 && Math.abs(dy3) < 1e-9) {
a = p1.x - p0.x;
b = p3.x - p0.x;
c = p0.x;
d = p1.y - p0.y;
e = p3.y - p0.y;
f = p0.y;
} else {
const denominator = dx1 * dy2 - dx2 * dy1;
if (Math.abs(denominator) < 1e-9) {
return {
x: p0.x + (p1.x - p0.x) * point.x + (p3.x - p0.x) * point.y,
y: p0.y + (p1.y - p0.y) * point.x + (p3.y - p0.y) * point.y,
};
}
g = (dx3 * dy2 - dx2 * dy3) / denominator;
h = (dx1 * dy3 - dx3 * dy1) / denominator;
a = p1.x - p0.x + g * p1.x;
b = p3.x - p0.x + h * p3.x;
c = p0.x;
d = p1.y - p0.y + g * p1.y;
e = p3.y - p0.y + h * p3.y;
f = p0.y;
}
const scale = g * point.x + h * point.y + 1;
return {
x: (a * point.x + b * point.y + c) / scale,
y: (d * point.x + e * point.y + f) / scale,
};
}
export function projectPolygon(polygon: Polygon, edge: ProjectiveEdge): Polygon {
const { width, height } = edge.sourceViewport;
if (width <= 0 || height <= 0) return [];
return polygon.map((point) =>
projectUnitPoint({ x: point.x / width, y: point.y / height }, edge.destinationQuad),
);
}
function polygonSignedArea(points: Polygon): number {
let area = 0;
for (let index = 0; index < points.length; index += 1) {
const current = points[index];
const next = points[(index + 1) % points.length];
area += current.x * next.y - next.x * current.y;
}
return area / 2;
}
function edgeIntersection(start: Point, end: Point, clipStart: Point, clipEnd: Point): Point {
const subjectX = end.x - start.x;
const subjectY = end.y - start.y;
const clipX = clipEnd.x - clipStart.x;
const clipY = clipEnd.y - clipStart.y;
const denominator = subjectX * clipY - subjectY * clipX;
if (Math.abs(denominator) < 1e-9) return end;
const offsetX = clipStart.x - start.x;
const offsetY = clipStart.y - start.y;
const scale = (offsetX * clipY - offsetY * clipX) / denominator;
return { x: start.x + scale * subjectX, y: start.y + scale * subjectY };
}
/** Clip a polygon to a convex polygon, preserving every resulting vertex. */
export function clipPolygon(subject: Polygon, clip: Polygon): Polygon {
if (subject.length < 3 || clip.length < 3) return [];
const orientation = polygonSignedArea(clip) >= 0 ? 1 : -1;
let output = [...subject];
for (let edge = 0; edge < clip.length && output.length > 0; edge += 1) {
const clipStart = clip[edge];
const clipEnd = clip[(edge + 1) % clip.length];
const input = output;
output = [];
const inside = (point: Point) =>
orientation *
((clipEnd.x - clipStart.x) * (point.y - clipStart.y) -
(clipEnd.y - clipStart.y) * (point.x - clipStart.x)) >=
-1e-9;
let start = input[input.length - 1];
for (const end of input) {
const startInside = inside(start);
const endInside = inside(end);
if (endInside) {
if (!startInside) output.push(edgeIntersection(start, end, clipStart, clipEnd));
output.push(end);
} else if (startInside) {
output.push(edgeIntersection(start, end, clipStart, clipEnd));
}
start = end;
}
}
return output;
}
export function polygonBounds(points: Polygon): ViewportRect | null {
if (points.length < 3) return null;
const xs = points.map((point) => point.x);
const ys = points.map((point) => point.y);
const x = Math.min(...xs);
const y = Math.min(...ys);
const right = Math.max(...xs);
const bottom = Math.max(...ys);
if (right <= x || bottom <= y) return null;
return { x, y, width: right - x, height: bottom - y };
}
export function regionBounds(region: Region): ViewportRect | null {
return polygonBounds(region.flat());
}
export function polygonArea(points: Polygon): number {
return Math.abs(polygonSignedArea(points));
}
export function polygonCentroid(points: Polygon): Point | null {
if (points.length < 3) return null;
let twiceArea = 0;
let x = 0;
let y = 0;
for (let index = 0; index < points.length; index += 1) {
const current = points[index];
const next = points[(index + 1) % points.length];
const cross = current.x * next.y - next.x * current.y;
twiceArea += cross;
x += (current.x + next.x) * cross;
y += (current.y + next.y) * cross;
}
if (Math.abs(twiceArea) < 1e-9) return null;
return { x: x / (3 * twiceArea), y: y / (3 * twiceArea) };
}
export function projectAndClipRegion(
region: Region,
edges: ProjectiveEdge[],
topViewport: Size,
): Region {
let projected = region;
for (const edge of edges) {
const sourceClip = viewportPolygon(edge.sourceViewport);
projected = projected
.map((polygon) => clipPolygon(polygon, sourceClip))
.filter((polygon) => polygon.length >= 3)
.map((polygon) =>
[edge.destinationQuad, ...(edge.destinationClips ?? [])].reduce(
(current, clip) => clipPolygon(current, clip),
projectPolygon(polygon, edge),
),
)
.filter((polygon) => polygon.length >= 3);
}
const topClip = viewportPolygon(topViewport);
return projected
.map((polygon) => clipPolygon(polygon, topClip))
.filter((polygon) => polygon.length >= 3);
}
export function projectRegionToViewport(region: Region, projection: GeometryProjection): Region {
const clipped = region
.map((polygon) =>
projection.sourceClips.reduce((current, clip) => clipPolygon(current, clip), polygon),
)
.filter((polygon) => polygon.length >= 3);
return projectAndClipRegion(clipped, projection.edges, projection.topViewport);
}
export function projectRectToViewport(
rect: { x: number; y: number; w: number; h: number } | null,
projection: GeometryProjection,
): ViewportRect | null {
if (!rect) return null;
return regionBounds(projectRegionToViewport([rectPolygon(rect)], projection));
}
export function childFrameProjection(
parent: GeometryProjection,
ownerRectInParent: { x: number; y: number; w: number; h: number },
): GeometryProjection {
const destinationQuad = rectPolygon(ownerRectInParent);
return {
sourceClips: [],
edges: [
{
sourceViewport: { width: ownerRectInParent.w, height: ownerRectInParent.h },
destinationQuad,
destinationClips: parent.sourceClips,
},
...parent.edges,
],
topViewport: parent.topViewport,
};
}
+68 -62
View File
@@ -8,7 +8,6 @@ import {
HELP_REQUEST,
type HelpCancelMessage,
type HelpFinishMessage,
type HelpHighlightRect,
type HelpQueryResponse,
type HelpRequestMessage,
isHelpAckMessage,
@@ -26,7 +25,6 @@ import type {
ResolvedTarget,
RpcError,
} from "@/transport/types";
import { resolveNodeGeometry } from "./frame-geometry";
import {
type CdpRunner,
type ChromeTabsApi,
@@ -39,6 +37,7 @@ import {
import { lookupSnapshotRef } from "./snapshot-ref";
const DEFAULT_HELP_TIMEOUT_MS = 300_000;
const HELP_ATTR = "data-bsk-help";
const HELP_SEND_RETRIES = 3;
const HELP_SEND_RETRY_DELAY_MS = 350;
const HELP_REARM_DEBOUNCE_MS = 150;
@@ -75,7 +74,6 @@ interface ActiveHelpRequest {
title?: string;
targets: HelpTarget[];
selectors: string[];
rects: HelpHighlightRect[];
timeoutMs: number;
notificationId: string;
resolvedTargets?: ResolvedTarget[];
@@ -136,7 +134,6 @@ function helpRequestMessage(help: ActiveHelpRequest, tabId: number): HelpRequest
...(help.title ? { title: help.title } : {}),
displayMode: isPrimaryTab ? "full" : "compact",
selectors: isPrimaryTab ? help.selectors : [],
rects: isPrimaryTab ? help.rects : [],
timeoutMs: help.timeoutMs,
};
}
@@ -168,6 +165,33 @@ async function sendHelpRequestWithAck(
throw lastError ?? new Error("failed to show help overlay");
}
async function tagRefTarget(
cdp: CdpRunner,
tabId: number,
backendNodeId: number,
index: number,
): Promise<string | null> {
let objectId: string | undefined;
try {
const resolved = await cdp.send<{ object?: { objectId?: string } }>(tabId, "DOM.resolveNode", {
backendNodeId,
});
objectId = resolved.object?.objectId;
if (!objectId) return null;
await cdp.send(tabId, "Runtime.callFunctionOn", {
objectId,
functionDeclaration: `function(){ this.setAttribute("${HELP_ATTR}", "${index}"); }`,
});
return `[${HELP_ATTR}="${index}"]`;
} catch {
return null;
} finally {
if (objectId) {
await cdp.send(tabId, "Runtime.releaseObject", { objectId }).catch(() => {});
}
}
}
async function selectorExists(
cdp: CdpRunner,
tabId: number,
@@ -185,22 +209,36 @@ async function selectorExists(
}
}
async function clearRefTags(cdp: CdpRunner | undefined, tabId: number): Promise<void> {
if (!cdp) return;
try {
const doc = await cdp.send<{ root?: { nodeId?: number } }>(tabId, "DOM.getDocument", {
depth: 0,
});
const rootId = doc.root?.nodeId;
if (rootId === undefined) return;
const { nodeIds } = await cdp.send<{ nodeIds: number[] }>(tabId, "DOM.querySelectorAll", {
nodeId: rootId,
selector: `[${HELP_ATTR}]`,
});
for (const nodeId of nodeIds ?? []) {
await cdp.send(tabId, "DOM.removeAttribute", { nodeId, name: HELP_ATTR }).catch(() => {});
}
} catch {
// Best-effort cleanup.
}
}
async function resolveHelpTargets(
ctx: ReturnType<SessionManager["get"]>,
tabId: number,
targets: HelpTarget[],
deps: RequestHelpDeps,
options: { scrollIntoView: boolean },
): Promise<{
selectors: string[];
rects: HelpHighlightRect[];
resolvedTargets?: ResolvedTarget[];
}> {
): Promise<{ selectors: string[]; resolvedTargets?: ResolvedTarget[] }> {
const selectors: string[] = [];
const rects: HelpHighlightRect[] = [];
const resolved: ResolvedTarget[] = [];
let rootNodeId: number | undefined;
if (deps.cdp && targets.some((target) => Boolean(target.selector))) {
if (deps.cdp && targets.length > 0) {
try {
const doc = await deps.cdp.send<{ root?: { nodeId?: number } }>(tabId, "DOM.getDocument", {
depth: 0,
@@ -224,55 +262,28 @@ async function resolveHelpTargets(
resolved.push({ matched, selector: tgt.selector });
} else if (tgt.ref) {
const looked = ctx ? lookupSnapshotRef(ctx, tgt.ref, tabId) : null;
let rect: HelpHighlightRect | null = null;
if (looked && deps.cdp) {
const target = {
tabId,
...(looked.cdpSessionId ? { sessionId: looked.cdpSessionId } : {}),
};
const geometry = await resolveNodeGeometry(
deps.cdp,
tabId,
{
target,
backendNodeId: looked.backendNodeId,
...(looked.frameId ? { frameId: looked.frameId } : {}),
},
{ scrollIntoView: options.scrollIntoView },
);
if (!isRpcError(geometry)) {
rect = {
top: geometry.topBounds.y,
left: geometry.topBounds.x,
width: geometry.topBounds.width,
height: geometry.topBounds.height,
};
}
const backendNodeId = looked?.backendNodeId ?? null;
let sel: string | null = null;
if (backendNodeId !== null && deps.cdp) {
sel = await tagRefTarget(deps.cdp, tabId, backendNodeId, i);
}
if (rect) rects.push(rect);
resolved.push({
matched: rect !== null,
ref: tgt.ref,
});
if (sel) selectors.push(sel);
resolved.push({ matched: sel !== null, ref: tgt.ref });
}
}
return { selectors, rects, resolvedTargets: resolved.length > 0 ? resolved : undefined };
return { selectors, resolvedTargets: resolved.length > 0 ? resolved : undefined };
}
async function refreshHelpTargets(
help: ActiveHelpRequest,
options: { scrollIntoView: boolean } = { scrollIntoView: true },
): Promise<void> {
const { selectors, rects, resolvedTargets } = await resolveHelpTargets(
async function refreshHelpTargets(help: ActiveHelpRequest): Promise<void> {
await clearRefTags(help.deps.cdp, help.primaryTabId);
const { selectors, resolvedTargets } = await resolveHelpTargets(
help.ctx,
help.primaryTabId,
help.targets,
help.deps,
options,
);
help.selectors = selectors;
help.rects = rects;
help.resolvedTargets = resolvedTargets;
}
@@ -283,7 +294,12 @@ async function cleanupHelp(help: ActiveHelpRequest): Promise<void> {
const tabsToCancel = new Set([help.primaryTabId, ...help.overlayTabIds]);
await Promise.all(
[...tabsToCancel].map((tabId) =>
help.deps.sendToTab(tabId, { type: HELP_CANCEL, requestId: help.requestId }).catch(() => {}),
Promise.all([
clearRefTags(help.deps.cdp, tabId),
help.deps
.sendToTab(tabId, { type: HELP_CANCEL, requestId: help.requestId })
.catch(() => {}),
]),
),
);
}
@@ -460,9 +476,7 @@ function attachHelpRuntimeListener(deps: RequestHelpDeps): () => void {
sendResponse({ active: false });
return;
}
if (tabId === help.primaryTabId) {
await refreshHelpTargets(help, { scrollIntoView: false });
}
if (tabId === help.primaryTabId) await refreshHelpTargets(help);
help.overlayTabIds.add(tabId);
sendResponse({
active: true,
@@ -472,7 +486,6 @@ function attachHelpRuntimeListener(deps: RequestHelpDeps): () => void {
...(help.title ? { title: help.title } : {}),
displayMode: tabId === help.primaryTabId ? "full" : "compact",
selectors: tabId === help.primaryTabId ? help.selectors : [],
rects: tabId === help.primaryTabId ? help.rects : [],
timeoutMs: help.timeoutMs,
},
});
@@ -673,13 +686,7 @@ export async function handleRequestHelp(
if (deps.autoAttachLifecycle !== false) ensureHelpLifecycleListeners(deps);
const tabId = target.tabId;
const initialTargets = params.targets ?? [];
const { selectors, rects, resolvedTargets } = await resolveHelpTargets(
ctx,
tabId,
initialTargets,
deps,
{ scrollIntoView: true },
);
const { selectors, resolvedTargets } = await resolveHelpTargets(ctx, tabId, initialTargets, deps);
await deps.windows.update(target.windowId, { focused: true }).catch(() => {});
await deps.activateTab(tabId).catch(() => {});
@@ -722,7 +729,6 @@ export async function handleRequestHelp(
...(params.title ? { title: params.title } : {}),
targets: initialTargets,
selectors,
rects,
timeoutMs,
notificationId,
resolvedTargets,
+56 -88
View File
@@ -3,15 +3,14 @@
//
// All interaction tools:
// 1. Resolve target tab (sandbox: must be inside Agent Window).
// 2. Resolve target element by `ref` (compound frame/session identity)
// or `selector` (DOM.querySelector + describeNode).
// 3. Scroll the node and its frame owners into view, project its live
// content quad into the top viewport, then dispatch `Input.*` events.
// 2. Resolve target element by `ref` (RefStore.resolve with tabId
// binding) or `selector` (DOM.querySelector + describeNode).
// 3. Scroll the node into view, then dispatch the appropriate
// `Input.*` CDP events.
// 4. Honour `AbortSignal` so canceled calls don't issue follow-up CDP
// commands.
import { ChromiumCdp } from "@/browser-driver/chromium-cdp";
import type { CdpTarget } from "@/browser-driver/frame-graph";
import type { SessionContext, SessionManager } from "@/session-manager/manager";
import type {
ClickParams,
@@ -29,13 +28,17 @@ import type {
SelectResult,
} from "@/transport/types";
import { attachDialogs, markDialogCursor } from "./dialogs";
import { backendNodeToObject } from "./element-geometry";
import {
backendNodeToObject,
boxCentre,
nodeCentre,
quadCentre,
scrollNodeIntoView,
} from "./element-geometry";
import { rpcError } from "./errors";
import { resolveNodeGeometry, scrollElementAndFramesIntoView } from "./frame-geometry";
import {
type CdpRunner,
type ChromeTabsApi,
cdpRunnerForTarget,
chromeTabsApi,
enforceAgentWindow,
isRpcError,
@@ -134,16 +137,7 @@ async function resolveBackendNode(
target: { tabId: number },
params: { ref?: string; selector?: string },
toolName: string,
): Promise<
| {
backendNodeId: number;
cdpTarget: CdpTarget;
frameId?: string;
usedRef?: string;
usedSelector?: string;
}
| RpcError
> {
): Promise<{ backendNodeId: number; usedRef?: string; usedSelector?: string } | RpcError> {
const hasRef = typeof params.ref === "string" && params.ref.length > 0;
const hasSelector = typeof params.selector === "string" && params.selector.length > 0;
if (hasRef && hasSelector) {
@@ -161,15 +155,7 @@ async function resolveBackendNode(
if (hasRef) {
const resolved = resolveSnapshotRef(ctx, params.ref as string, target.tabId);
if (isRpcError(resolved)) return resolved;
return {
backendNodeId: resolved.backendNodeId,
cdpTarget: {
tabId: target.tabId,
...(resolved.cdpSessionId ? { sessionId: resolved.cdpSessionId } : {}),
},
...(resolved.frameId ? { frameId: resolved.frameId } : {}),
usedRef: resolved.refKey,
};
return { backendNodeId: resolved.backendNodeId, usedRef: resolved.refKey };
}
// selector path
try {
@@ -206,7 +192,7 @@ async function resolveBackendNode(
message: "DOM.describeNode returned no backendNodeId",
};
}
return { backendNodeId, cdpTarget: { tabId: target.tabId }, usedSelector: params.selector };
return { backendNodeId, usedSelector: params.selector };
} catch (err) {
return {
code: "cdp_failed",
@@ -242,19 +228,19 @@ export async function handleClick(
return { code: "cancelled", message: "click aborted" };
}
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const geometry = await resolveNodeGeometry(
deps.cdp,
target.tabId,
{
target: node.cdpTarget,
backendNodeId: node.backendNodeId,
...(node.frameId ? { frameId: node.frameId } : {}),
},
{ scrollIntoView: true },
);
if (isRpcError(geometry)) return geometry;
const centre = geometry.actionPoint;
try {
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const scrollErr = await scrollNodeIntoView(deps.cdp, target.tabId, node.backendNodeId);
if (scrollErr) return scrollErr;
} catch (err) {
return {
code: "cdp_failed",
message: err instanceof Error ? err.message : String(err),
};
}
const centre = await nodeCentre(deps.cdp, target.tabId, node.backendNodeId);
if (isRpcError(centre)) return centre;
if (throwIfAborted(deps.signal)) {
return { code: "cancelled", message: "click aborted" };
@@ -367,19 +353,19 @@ export async function handleHover(
const node = await resolveBackendNode(deps.cdp, ctx, target, params, "hover");
if (isRpcError(node)) return node;
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const geometry = await resolveNodeGeometry(
deps.cdp,
target.tabId,
{
target: node.cdpTarget,
backendNodeId: node.backendNodeId,
...(node.frameId ? { frameId: node.frameId } : {}),
},
{ scrollIntoView: true },
);
if (isRpcError(geometry)) return geometry;
const centre = geometry.actionPoint;
try {
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const scrollErr = await scrollNodeIntoView(deps.cdp, target.tabId, node.backendNodeId);
if (scrollErr) return scrollErr;
} catch (err) {
return {
code: "cdp_failed",
message: err instanceof Error ? err.message : String(err),
};
}
const centre = await nodeCentre(deps.cdp, target.tabId, node.backendNodeId);
if (isRpcError(centre)) return centre;
if (throwIfAborted(deps.signal)) {
return { code: "cancelled", message: "hover aborted" };
@@ -599,11 +585,10 @@ export async function handleFill(
const node = await resolveBackendNode(deps.cdp, ctx, target, params, "fill");
if (isRpcError(node)) return node;
const nodeCdp = cdpRunnerForTarget(deps.cdp, node.cdpTarget);
try {
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const described = await nodeCdp.send<{ node?: DescribedNode }>(
const described = await deps.cdp.send<{ node?: DescribedNode }>(
target.tabId,
"DOM.describeNode",
{
@@ -617,18 +602,12 @@ export async function handleFill(
`element ${described.node?.nodeName ?? "?"} not fillable (need input/textarea/contenteditable)`,
);
}
const scrollErr = await scrollElementAndFramesIntoView(
deps.cdp,
target.tabId,
node.cdpTarget,
node.backendNodeId,
node.frameId,
);
const scrollErr = await scrollNodeIntoView(deps.cdp, target.tabId, node.backendNodeId);
if (scrollErr) return scrollErr;
if (throwIfAborted(deps.signal)) {
return { code: "cancelled", message: "fill aborted" };
}
await nodeCdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
await deps.cdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
} catch (err) {
return {
code: "cdp_failed",
@@ -640,7 +619,7 @@ export async function handleFill(
return { code: "cancelled", message: "fill aborted" };
}
const objectIdOrErr = await backendNodeToObject(nodeCdp, target.tabId, node.backendNodeId);
const objectIdOrErr = await backendNodeToObject(deps.cdp, target.tabId, node.backendNodeId);
if (isRpcError(objectIdOrErr)) return objectIdOrErr;
const objectId = objectIdOrErr;
const clearBefore = params.clear_before ?? true;
@@ -649,7 +628,7 @@ export async function handleFill(
if (clearBefore) {
// Clear input/textarea value or wipe contenteditable innerText,
// then fire `input` so frameworks observe the empty state.
await nodeCdp.send(target.tabId, "Runtime.callFunctionOn", {
await deps.cdp.send(target.tabId, "Runtime.callFunctionOn", {
objectId,
functionDeclaration: `function() {
if (this.isContentEditable) { this.textContent = ''; }
@@ -676,7 +655,7 @@ export async function handleFill(
return { code: "cancelled", message: "fill aborted" };
}
// Fire `input` + `change` so React / Vue controlled inputs commit.
await nodeCdp.send(target.tabId, "Runtime.callFunctionOn", {
await deps.cdp.send(target.tabId, "Runtime.callFunctionOn", {
objectId,
functionDeclaration: `function() {
this.dispatchEvent(new Event('input', { bubbles: true }));
@@ -874,21 +853,14 @@ export async function handlePress(
if (params.ref || params.selector) {
const node = await resolveBackendNode(deps.cdp, ctx, target, params, "press");
if (isRpcError(node)) return node;
const nodeCdp = cdpRunnerForTarget(deps.cdp, node.cdpTarget);
try {
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const scrollErr = await scrollElementAndFramesIntoView(
deps.cdp,
target.tabId,
node.cdpTarget,
node.backendNodeId,
node.frameId,
);
const scrollErr = await scrollNodeIntoView(deps.cdp, target.tabId, node.backendNodeId);
if (scrollErr) return scrollErr;
if (throwIfAborted(deps.signal)) {
return { code: "cancelled", message: "press aborted" };
}
await nodeCdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
await deps.cdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
} catch (err) {
return {
code: "cdp_failed",
@@ -993,11 +965,10 @@ export async function handleSelect(
const node = await resolveBackendNode(deps.cdp, ctx, target, params, "select");
if (isRpcError(node)) return node;
const nodeCdp = cdpRunnerForTarget(deps.cdp, node.cdpTarget);
try {
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const described = await nodeCdp.send<{ node?: DescribedNode }>(
const described = await deps.cdp.send<{ node?: DescribedNode }>(
target.tabId,
"DOM.describeNode",
{ backendNodeId: node.backendNodeId },
@@ -1020,18 +991,12 @@ export async function handleSelect(
"single-select <select> requires exactly one value",
);
}
const scrollErr = await scrollElementAndFramesIntoView(
deps.cdp,
target.tabId,
node.cdpTarget,
node.backendNodeId,
node.frameId,
);
const scrollErr = await scrollNodeIntoView(deps.cdp, target.tabId, node.backendNodeId);
if (scrollErr) return scrollErr;
if (throwIfAborted(deps.signal)) {
return { code: "cancelled", message: "select aborted" };
}
await nodeCdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
await deps.cdp.send(target.tabId, "DOM.focus", { backendNodeId: node.backendNodeId });
} catch (err) {
return {
code: "cdp_failed",
@@ -1043,12 +1008,12 @@ export async function handleSelect(
return { code: "cancelled", message: "select aborted" };
}
const objectIdOrErr = await backendNodeToObject(nodeCdp, target.tabId, node.backendNodeId);
const objectIdOrErr = await backendNodeToObject(deps.cdp, target.tabId, node.backendNodeId);
if (isRpcError(objectIdOrErr)) return objectIdOrErr;
const objectId = objectIdOrErr;
try {
const evaluated = await nodeCdp.send<{
const evaluated = await deps.cdp.send<{
result?: { value?: SelectMutationResult | null };
}>(target.tabId, "Runtime.callFunctionOn", {
objectId,
@@ -1107,6 +1072,9 @@ export async function handleSelect(
export const __testing__ = {
DEFAULT_TIMEOUT_MS,
quadCentre,
boxCentre,
resolveBackendNode,
nodeCentre,
isFillable,
};
-5
View File
@@ -37,22 +37,17 @@ export async function handleNetwork(
manager: SessionManager,
params: NetworkParams,
deps: NetworkDeps = defaultNetworkDeps(),
signal?: AbortSignal,
): Promise<NetworkResult | RpcError> {
if (signal?.aborted) return { code: "cancelled", message: "network aborted" };
const ctxOrErr = lookupSession(manager, params, "network");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const bounds = parseBufferedReadBounds(params);
if (isRpcError(bounds)) return bounds;
const target = await resolveTargetTab(manager, ctxOrErr, params.tab_id, deps.tabsApi);
if (isRpcError(target)) return target;
if (signal?.aborted) return { code: "cancelled", message: "network aborted" };
try {
if (signal?.aborted) return { code: "cancelled", message: "network aborted" };
deps.cdp.trackSessionTab?.(ctxOrErr.sessionId, target.tabId);
await deps.cdp.ensureNetworkCapture(target.tabId);
if (signal?.aborted) return { code: "cancelled", message: "network aborted" };
return deps.cdp.networkEntriesSince(
target.tabId,
bounds.since,
+42 -361
View File
@@ -8,13 +8,10 @@ import {
applyVomInteractionRecovery,
type CondSurface,
isVomReferenceNode,
renderVom,
type VomNode,
type VomOptions,
type VomScene,
} from "@browser-skill/vom";
import { ChromiumCdp } from "@/browser-driver/chromium-cdp";
import type { CdpTarget } from "@/browser-driver/frame-graph";
import {
type CaptureSuppressSendToTab,
withOverlaysHiddenForCapture,
@@ -31,9 +28,10 @@ import type {
SnapshotParams,
SnapshotResult,
} from "@/transport/types";
import { captureVomObservation } from "./capture-vom-observation";
import { attachDialogs, markDialogCursor } from "./dialogs";
import { nodeBoundingRect, scrollNodeIntoView } from "./element-geometry";
import { rpcError } from "./errors";
import { resolveNodeGeometry } from "./frame-geometry";
import {
type ChromeTabsApi,
enforceToolTargetScope,
@@ -42,22 +40,11 @@ import {
type ResolvedTargetTab,
resolveCdpAccessibleTargetTab,
type CdpRunner as SharedCdpRunner,
sendToCdpTarget,
normaliseRef as sharedNormaliseRef,
type ToolEffect,
} from "./shared";
import { resolveSnapshotRef } from "./snapshot-ref";
import {
type CapturedNode,
type CapturedViewModel,
captureViewModel,
collectOverlayExcludedBackendIds,
} from "./vom/capture";
import { type CapturedFrameDocument, captureFrameData } from "./vom/frame-capture";
import {
type CaptureVomObservationResult,
projectRecordSafeObservation,
} from "./vom/record-safe-observation";
import { type CapturedNode, type CapturedViewModel } from "./vom/capture";
// ---------------------------------------------------------------------------
// Shared helpers (legacy aliases — observation.ts kept exporting these
@@ -145,66 +132,35 @@ function defaultScreenshotDeps(): ScreenshotDeps {
};
}
function cancelled(tool: string): RpcError {
return { code: "cancelled", message: `${tool} aborted` };
}
function abortError(tool: string): Error {
const error = new Error(`${tool} aborted`);
error.name = "AbortError";
return error;
}
function isAbortError(error: unknown): boolean {
return (
typeof error === "object" &&
error !== null &&
(error as { name?: string }).name === "AbortError"
);
}
function throwIfAborted(signal: AbortSignal | undefined, tool: string): void {
if (signal?.aborted) throw abortError(tool);
}
async function captureElementScreenshot(
cdp: SharedCdpRunner,
tabId: number,
target: CdpTarget,
backendNodeId: number,
frameId?: string,
signal?: AbortSignal,
): Promise<{ image_base64: string; width: number; height: number } | RpcError> {
if (signal?.aborted) return cancelled("screenshot");
const geometry = await resolveNodeGeometry(
cdp,
tabId,
{ target, backendNodeId, ...(frameId ? { frameId } : {}) },
{ scrollIntoView: true },
);
if (isRpcError(geometry)) return geometry;
if (signal?.aborted) return cancelled("screenshot");
const rect = geometry.topBounds;
const scrollErr = await scrollNodeIntoView(cdp, tabId, backendNodeId);
if (scrollErr) return scrollErr;
const rectOrErr = await nodeBoundingRect(cdp, tabId, backendNodeId);
if (isRpcError(rectOrErr)) return rectOrErr;
try {
const shot = await cdp.send<{ data?: string }>(tabId, "Page.captureScreenshot", {
format: "png",
clip: {
x: rect.x,
y: rect.y,
width: rect.width,
height: rect.height,
x: rectOrErr.x,
y: rectOrErr.y,
width: rectOrErr.width,
height: rectOrErr.height,
scale: 1,
},
});
if (signal?.aborted) return cancelled("screenshot");
const image_base64 = shot.data ?? "";
if (!image_base64) {
return { code: "cdp_failed", message: "Page.captureScreenshot returned no data" };
}
const dims = parsePngDimensions(image_base64) ?? {
width: Math.round(rect.width),
height: Math.round(rect.height),
width: Math.round(rectOrErr.width),
height: Math.round(rectOrErr.height),
};
return { image_base64, width: dims.width, height: dims.height };
} catch (err) {
@@ -232,15 +188,11 @@ async function captureFullTabPng(
deps: ScreenshotDeps,
ctx: SessionContext,
target: ResolvedTargetTab,
signal?: AbortSignal,
): Promise<string | RpcError> {
if (signal?.aborted) return cancelled("screenshot");
try {
const dataUrl = await deps.captureApi.captureVisibleTab(target.windowId, { format: "png" });
if (signal?.aborted) return cancelled("screenshot");
return stripDataUrlPrefix(dataUrl);
} catch (primaryErr) {
if (signal?.aborted) return cancelled("screenshot");
const cdp = deps.cdp;
if (!cdp) {
return {
@@ -250,15 +202,12 @@ async function captureFullTabPng(
}
let fallbackMsg: string;
try {
if (signal?.aborted) return cancelled("screenshot");
cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
await cdp.ensureAttachedToUrl?.(target.tabId, target.url);
if (signal?.aborted) return cancelled("screenshot");
const shot = await cdp.send<{ data?: string }>(target.tabId, "Page.captureScreenshot", {
format: "png",
fromSurface: true,
});
if (signal?.aborted) return cancelled("screenshot");
if (shot.data) return shot.data;
fallbackMsg = "Page.captureScreenshot returned no data";
} catch (fallbackErr) {
@@ -277,9 +226,7 @@ export async function handleScreenshot(
manager: SessionManager,
params: ScreenshotParams,
deps: ScreenshotDeps = defaultScreenshotDeps(),
signal?: AbortSignal,
): Promise<ScreenshotResult | RpcError> {
if (signal?.aborted) return cancelled("screenshot");
const ctxOrErr = lookupSession(manager, params, "screenshot");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const ctx = ctxOrErr;
@@ -291,7 +238,6 @@ export async function handleScreenshot(
"screenshot",
);
if (isRpcError(target)) return target;
if (signal?.aborted) return cancelled("screenshot");
const dialogCursor = deps.cdp ? markDialogCursor(deps.cdp, target.tabId) : 0;
const withShotDialogs = <T extends object>(result: T) =>
deps.cdp ? attachDialogs(deps.cdp, target.tabId, dialogCursor, result) : result;
@@ -303,30 +249,15 @@ export async function handleScreenshot(
}
const node = resolveSnapshotRef(ctx, ref, target.tabId);
if (isRpcError(node)) return node;
if (signal?.aborted) return cancelled("screenshot");
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
await deps.cdp.ensureAttachedToUrl?.(target.tabId, target.url);
if (signal?.aborted) return cancelled("screenshot");
const cdp = deps.cdp;
const nodeTarget = {
tabId: target.tabId,
...(node.cdpSessionId ? { sessionId: node.cdpSessionId } : {}),
};
const captured = await withOverlaysHiddenForCapture(
target.tabId,
() =>
captureElementScreenshot(
cdp,
target.tabId,
nodeTarget,
node.backendNodeId,
node.frameId,
signal,
),
() => captureElementScreenshot(cdp, target.tabId, node.backendNodeId),
deps.sendToTab,
);
if (isRpcError(captured)) return captured;
if (signal?.aborted) return cancelled("screenshot");
return withShotDialogs({
image_base64: captured.image_base64,
width: captured.width,
@@ -346,11 +277,10 @@ export async function handleScreenshot(
const captured = await withOverlaysHiddenForCapture(
target.tabId,
() => captureFullTabPng(deps, ctx, target, signal),
() => captureFullTabPng(deps, ctx, target),
deps.sendToTab,
);
if (isRpcError(captured)) return captured;
if (signal?.aborted) return cancelled("screenshot");
const image_base64 = captured;
const dims = parsePngDimensions(image_base64) ?? { width: 0, height: 0 };
return withShotDialogs({
@@ -376,7 +306,6 @@ export type CdpRunner = SharedCdpRunner;
/** Subset of CDP `AXNode` we care about — see `Accessibility.AXNode`. */
export interface CdpAxNode {
nodeId: string;
frameId?: string;
parentId?: string;
backendDOMNodeId?: number;
ignored?: boolean;
@@ -419,12 +348,15 @@ function isSensitive(capturedNode: CapturedNode | undefined): boolean {
const type = (attrs.type ?? "").toLowerCase();
if (type === "password") return true;
const autocomplete = (attrs.autocomplete ?? "").toLowerCase();
return (
autocomplete.startsWith("cc-") ||
if (autocomplete.startsWith("cc-")) return true;
if (
autocomplete === "one-time-code" ||
autocomplete === "current-password" ||
autocomplete === "new-password"
);
) {
return true;
}
return false;
}
interface AxNodeSignals {
@@ -623,7 +555,7 @@ function iframeNameFor(node: CapturedNode): string | undefined {
function isRenderableIframeControl(node: CapturedNode): boolean {
const tag = normalizeTag(node.tag);
if (!IFRAME_RENDERABLE_TAGS.has(tag)) return false;
if (!node.localRect && !node.rect) return false;
if (!node.rect) return false;
if (node.pointerEvents === "none") return false;
if ((node.attrs.type ?? "").toLowerCase() === "hidden") return false;
@@ -795,12 +727,7 @@ function applyCapturedSignals(
signals: VomNodeDomSignals,
): VomNode {
if (!capturedNode) return node;
const attrs =
node.sensitive && Object.prototype.hasOwnProperty.call(capturedNode.attrs, "value")
? Object.fromEntries(
Object.entries(capturedNode.attrs).filter(([name]) => name.toLowerCase() !== "value"),
)
: capturedNode.attrs;
const attrs = capturedNode.attrs;
const text = cleanAttr(capturedNode.textContent);
const nearbyText = nearbyTextFor(capturedNode, signals.childrenByParentId);
const placeholder = cleanAttr(capturedNode.formPlaceholder) ?? cleanAttr(attrs.placeholder);
@@ -867,7 +794,8 @@ function vomNodeFromCaptured(
signals: VomNodeDomSignals,
): VomNode {
const sensitive = isSensitive(capturedNode);
const value = sensitive ? undefined : capturedNode.formValue;
const value =
capturedNode.formValue ?? (sensitive ? (capturedNode.attrs.value ?? "") : undefined);
const tag = normalizeTag(capturedNode.tag);
const node = applyCapturedSignals(
{
@@ -903,17 +831,8 @@ function axVomNode(
): VomNode {
const role = axString(axNode.role);
const capturedValue = capturedNode?.formValue;
const sourceValue = capturedValue ?? axValue(axNode.value);
const value = capturedValue ?? axValue(axNode.value);
const signalsForNode = axSignals.get(axNode.nodeId);
const sensitive = isSensitive(capturedNode) || signalsForNode?.sensitive === true;
const value = sensitive ? undefined : sourceValue;
const inputState =
inputStateFor(capturedNode, sourceValue) ??
(sensitive && ["textbox", "searchbox"].includes(role?.toLowerCase() ?? "")
? sourceValue === undefined || sourceValue === ""
? "empty"
: "filled"
: undefined);
let name = FORM_CONTROL_TAGS.has(normalizeTag(capturedNode?.tag))
? formControlName(capturedNode, axString(axNode.name), axSiblingLabel(axNode, axById), signals)
: (axString(axNode.name) ?? signalsForNode?.aggregatedText);
@@ -929,8 +848,8 @@ function axVomNode(
position: capturedNode?.position || "static",
pointerEvents: capturedNode?.pointerEvents || "auto",
modal: isModalSignal(axNode, capturedNode),
sensitive,
inputState,
sensitive: isSensitive(capturedNode) || signalsForNode?.sensitive === true,
inputState: inputStateFor(capturedNode, value),
};
if (role) node.role = role;
if (name) node.name = name;
@@ -1193,8 +1112,6 @@ export interface BuildVomSceneOptions {
pageUrl?: string;
}
export type VomFrameDocument = CapturedFrameDocument<CdpAxNode>;
function externalHrefHost(
href: string | undefined,
pageUrl: string | undefined,
@@ -1288,119 +1205,6 @@ export function buildVomScene(
};
}
export function buildFrameVomScene(
documents: VomFrameDocument[],
captured: CapturedViewModel,
options: BuildVomSceneOptions = {},
): VomScene {
const documentByFrameId = new Map(documents.map((document) => [document.frameId, document]));
const rootFrameId =
captured.rootFrameId ?? documents.find((document) => !document.parentFrameId)?.frameId;
const orderedFrameIds = [
...(rootFrameId ? [rootFrameId] : []),
...documents.map((document) => document.frameId),
].filter((frameId, index, all) => all.indexOf(frameId) === index);
const localScenes = new Map<string, VomScene>();
for (const frameId of orderedFrameIds) {
const document = documentByFrameId.get(frameId);
if (!document) continue;
localScenes.set(
frameId,
buildVomScene(
document.axNodes,
{
viewport: captured.viewport,
nodes: document.domNodes,
iframeNodes: new Map(),
surfaceProbes: frameId === rootFrameId ? captured.surfaceProbes : [],
excludedBackendNodeIds:
frameId === rootFrameId ? captured.excludedBackendNodeIds : new Set(),
},
{ pageUrl: document.url ?? options.pageUrl },
),
);
}
const sceneIdByFrameAndBackend = new Map<string, number>();
let nextSceneId = 1;
for (const [frameId, scene] of localScenes) {
for (const node of scene.nodes) {
sceneIdByFrameAndBackend.set(`${frameId}:${node.id}`, nextSceneId);
nextSceneId += 1;
}
}
const nodes: VomNode[] = [];
const surfaces: CondSurface[] = [];
const activeScopeBlocks: ActiveScopeBlock[] = [];
for (const [frameId, scene] of localScenes) {
const document = documentByFrameId.get(frameId);
const ownerBackendNodeId = document?.ownerBackendNodeId;
const parentFrameId = document?.parentFrameId;
const ownerSceneId =
ownerBackendNodeId !== undefined && parentFrameId
? sceneIdByFrameAndBackend.get(`${parentFrameId}:${ownerBackendNodeId}`)
: undefined;
if (frameId !== rootFrameId && ownerSceneId === undefined) continue;
const suppressedRoots = new Set(
frameId === rootFrameId
? []
: scene.nodes
.filter(
(node) =>
node.parentId === null &&
["rootwebarea", "webarea"].includes(node.role?.toLowerCase() ?? ""),
)
.map((node) => node.id),
);
for (const node of scene.nodes) {
if (suppressedRoots.has(node.id)) continue;
const sceneNodeId = sceneIdByFrameAndBackend.get(`${frameId}:${node.id}`) as number;
let parentId =
node.parentId !== null
? sceneIdByFrameAndBackend.get(`${frameId}:${node.parentId}`)
: ownerSceneId;
if (node.parentId !== null && suppressedRoots.has(node.parentId)) parentId = ownerSceneId;
const remapId = (id: number) => sceneIdByFrameAndBackend.get(`${frameId}:${id}`);
nodes.push({
...node,
id: sceneNodeId,
backendNodeId: node.id,
frameId,
contextScopeId: document?.contextScopeId ?? frameId,
parentId: parentId ?? null,
...(node.domParentId !== undefined
? { domParentId: node.domParentId === null ? null : (remapId(node.domParentId) ?? null) }
: {}),
...(node.domAncestorIds
? { domAncestorIds: node.domAncestorIds.flatMap((id) => remapId(id) ?? []) }
: {}),
});
}
for (const surface of scene.surfaces ?? []) {
const triggerId = sceneIdByFrameAndBackend.get(`${frameId}:${surface.triggerId}`);
if (triggerId !== undefined) surfaces.push({ ...surface, triggerId });
}
for (const block of scene.activeScopeBlocks ?? []) {
const triggerId = sceneIdByFrameAndBackend.get(`${frameId}:${block.triggerId}`);
if (triggerId !== undefined) activeScopeBlocks.push({ ...block, triggerId });
}
}
return {
viewport: captured.viewport,
nodes,
...(rootFrameId ? { rootFrameId } : {}),
...(surfaces.length > 0 ? { surfaces } : {}),
...(activeScopeBlocks.length > 0 ? { activeScopeBlocks } : {}),
};
}
export interface SnapshotDeps {
cdp: CdpRunner;
tabsApi: {
@@ -1460,9 +1264,7 @@ export async function handleGetHtml(
manager: SessionManager,
params: GetHtmlParams,
deps: SnapshotDeps = getDefaultDeps(),
signal?: AbortSignal,
): Promise<GetHtmlResult | RpcError> {
if (signal?.aborted) return cancelled("get_html");
const ctxOrErr = lookupSession(manager, params, "get_html");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const ctx = ctxOrErr;
@@ -1474,31 +1276,21 @@ export async function handleGetHtml(
"get_html",
);
if (isRpcError(target)) return target;
if (signal?.aborted) return cancelled("get_html");
const dialogCursor = markDialogCursor(deps.cdp, target.tabId);
const maxBytes =
params.max_bytes && params.max_bytes > 0 ? params.max_bytes : DEFAULT_GET_HTML_MAX_BYTES;
try {
throwIfAborted(signal, "get_html");
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
await deps.cdp.ensureAttachedToUrl?.(target.tabId, target.url);
throwIfAborted(signal, "get_html");
let html: string;
if (params.ref) {
const resolved = resolveSnapshotRef(ctx, params.ref, target.tabId);
if (isRpcError(resolved)) return resolved;
const resp = await sendToCdpTarget<{ outerHTML?: string }>(
deps.cdp,
{
tabId: target.tabId,
...(resolved.cdpSessionId ? { sessionId: resolved.cdpSessionId } : {}),
},
"DOM.getOuterHTML",
{ backendNodeId: resolved.backendNodeId },
);
throwIfAborted(signal, "get_html");
const resp = await deps.cdp.send<{ outerHTML?: string }>(target.tabId, "DOM.getOuterHTML", {
backendNodeId: resolved.backendNodeId,
});
html = resp.outerHTML ?? "";
} else {
const doc = await deps.cdp.send<{ root?: { nodeId?: number } }>(
@@ -1506,7 +1298,6 @@ export async function handleGetHtml(
"DOM.getDocument",
{ depth: 0 },
);
throwIfAborted(signal, "get_html");
const nodeId = doc.root?.nodeId;
if (typeof nodeId !== "number") {
return {
@@ -1517,7 +1308,6 @@ export async function handleGetHtml(
const resp = await deps.cdp.send<{ outerHTML?: string }>(target.tabId, "DOM.getOuterHTML", {
nodeId,
});
throwIfAborted(signal, "get_html");
html = resp.outerHTML ?? "";
}
const originalBytes = utf8ByteLength(html);
@@ -1529,7 +1319,6 @@ export async function handleGetHtml(
tab_id: target.tabId,
});
} catch (err) {
if (isAbortError(err)) return cancelled("get_html");
return {
code: "cdp_failed",
message: err instanceof Error ? err.message : String(err),
@@ -1537,93 +1326,6 @@ export async function handleGetHtml(
}
}
function emptyCapturedViewModel(viewport = { width: 0, height: 0 }): CapturedViewModel {
return { viewport, nodes: [], iframeNodes: new Map(), excludedBackendNodeIds: new Set() };
}
interface LayoutMetricsViewportReply {
cssLayoutViewport?: { clientWidth?: number; clientHeight?: number };
layoutViewport?: { clientWidth?: number; clientHeight?: number };
}
async function fallbackCapturedViewModel(
cdp: CdpRunner,
tabId: number,
signal?: AbortSignal,
): Promise<CapturedViewModel> {
throwIfAborted(signal, "observation");
let viewport = { width: 0, height: 0 };
try {
const metrics = await cdp.send<LayoutMetricsViewportReply>(tabId, "Page.getLayoutMetrics", {});
throwIfAborted(signal, "observation");
const source = metrics.cssLayoutViewport ?? metrics.layoutViewport ?? {};
viewport = {
width: source.clientWidth ?? 0,
height: source.clientHeight ?? 0,
};
} catch (error) {
if (isAbortError(error)) throw error;
}
const excludedBackendNodeIds = await collectOverlayExcludedBackendIds(cdp, tabId, signal);
throwIfAborted(signal, "observation");
return { ...emptyCapturedViewModel(viewport), excludedBackendNodeIds };
}
async function captureForVom(
cdp: CdpRunner,
tabId: number,
options: CaptureVomObservationOptions,
): Promise<CapturedViewModel> {
try {
return await captureViewModel(cdp, tabId, {
conditionalSurfaceProbe: options.conditionalSurfaceProbe ?? false,
hoverProbeBypassOverlay: options.hoverProbeBypassOverlay,
signal: options.signal,
});
} catch (error) {
if (isAbortError(error)) throw error;
return fallbackCapturedViewModel(cdp, tabId, options.signal);
}
}
export interface CaptureVomObservationOptions extends VomOptions {
conditionalSurfaceProbe?: boolean;
hoverProbeBypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>;
signal?: AbortSignal;
}
export async function captureVomObservation(
cdp: CdpRunner,
tabId: number,
url: string | undefined,
options: CaptureVomObservationOptions = {},
): Promise<CaptureVomObservationResult> {
throwIfAborted(options.signal, "observation");
await cdp.ensureAttachedToUrl?.(tabId, url);
throwIfAborted(options.signal, "observation");
const captured = await captureForVom(cdp, tabId, options);
throwIfAborted(options.signal, "observation");
const documents = await captureFrameData<CdpAxNode>(cdp, tabId, captured, options.signal);
throwIfAborted(options.signal, "observation");
const scene =
captured.rootFrameId && captured.frameNodes?.size
? buildFrameVomScene(documents, captured, { pageUrl: url })
: buildVomScene(documents[0]?.axNodes ?? [], captured, { pageUrl: url });
const rendered = renderVom(scene, {
maxDepth: options.maxDepth,
maxTokens: options.maxTokens,
redactValues: options.redactValues,
activeRegionPolicy: options.activeRegionPolicy,
});
throwIfAborted(options.signal, "observation");
return projectRecordSafeObservation({
rootFrameId: captured.rootFrameId ?? documents[0]?.frameId ?? "root",
frameDocuments: documents,
rendered,
surfaceProbes: captured.surfaceProbes,
});
}
async function handleVomObservation(
manager: SessionManager,
params: SnapshotParams | ObserveParams,
@@ -1631,9 +1333,7 @@ async function handleVomObservation(
effect: ToolEffect,
conditionalSurfaceProbe: boolean,
deps: SnapshotDeps = getDefaultDeps(),
signal?: AbortSignal,
): Promise<SnapshotResult | ObserveResult | RpcError> {
if (signal?.aborted) return cancelled(toolName);
const ctxOrErr = lookupSession(manager, params, toolName);
if (isRpcError(ctxOrErr)) return ctxOrErr;
const ctx = ctxOrErr;
@@ -1645,51 +1345,35 @@ async function handleVomObservation(
toolName,
);
if (isRpcError(target)) return target;
if (signal?.aborted) return cancelled(toolName);
const denied = enforceToolTargetScope(ctx, target, effect, toolName);
if (denied) return denied;
const dialogCursor = markDialogCursor(deps.cdp, target.tabId);
try {
throwIfAborted(signal, toolName);
deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
const effectiveConditionalSurfaceProbe =
deps.conditionalSurfaceProbe ?? conditionalSurfaceProbe;
const observation = await captureVomObservation(deps.cdp, target.tabId, target.url, {
const rendered = await captureVomObservation(deps.cdp, target.tabId, target.url, {
maxDepth: params.max_depth,
maxTokens: params.max_tokens,
activeRegionPolicy: true,
conditionalSurfaceProbe: effectiveConditionalSurfaceProbe,
hoverProbeBypassOverlay: deps.hoverProbeBypassOverlay,
signal,
});
throwIfAborted(signal, toolName);
const targetByFrameId = new Map(
observation.frames.map((frame) => [frame.frameId, frame.target]),
);
ctx.refStore.replace(
observation.refs.map((ref) => {
const refTarget = ref.frameId ? targetByFrameId.get(ref.frameId) : undefined;
return [
ref.ref,
{
backendNodeId: ref.backendNodeId,
tabId: target.tabId,
...(ref.frameId ? { frameId: ref.frameId } : {}),
...(refTarget?.sessionId ? { cdpSessionId: refTarget.sessionId } : {}),
},
] as const;
}),
rendered.refs.map(
(r) => [r.ref, { backendNodeId: r.backendNodeId, tabId: target.tabId }] as const,
),
);
return attachDialogs(deps.cdp, target.tabId, dialogCursor, {
text: observation.text,
ref_count: observation.refs.length,
text: rendered.text,
ref_count: rendered.refs.length,
tab_id: target.tabId,
truncated: observation.truncated,
truncated: rendered.truncated,
...(toolName === "observe" && (params as ObserveParams).debug_surfaces
? {
debug: {
surface_probes: (observation.surfaceProbes ?? []).map((probe) => ({
surface_probes: (rendered.surfaceProbes ?? []).map((probe) => ({
trigger_backend_node_id: probe.triggerBackendNodeId,
...(probe.triggerPoint ? { trigger_point: probe.triggerPoint } : {}),
trigger_action: probe.triggerAction,
@@ -1701,7 +1385,6 @@ async function handleVomObservation(
: {}),
});
} catch (err) {
if (isAbortError(err)) return cancelled(toolName);
return {
code: "cdp_failed",
message: err instanceof Error ? err.message : String(err),
@@ -1713,16 +1396,14 @@ export async function handleSnapshot(
manager: SessionManager,
params: SnapshotParams,
deps: SnapshotDeps = getDefaultDeps(),
signal?: AbortSignal,
): Promise<SnapshotResult | RpcError> {
return handleVomObservation(manager, params, "snapshot", "passive_read", false, deps, signal);
return handleVomObservation(manager, params, "snapshot", "passive_read", false, deps);
}
export async function handleObserve(
manager: SessionManager,
params: ObserveParams,
deps: SnapshotDeps = getDefaultDeps(),
signal?: AbortSignal,
): Promise<ObserveResult | RpcError> {
return handleVomObservation(manager, params, "observe", "transient_input", true, deps, signal);
return handleVomObservation(manager, params, "observe", "transient_input", true, deps);
}
+566 -96
View File
@@ -7,21 +7,53 @@ import {
isRecordQueryMessage,
isRecordStepMessage,
RECORD_CANCEL,
RECORD_CAPTURE_STATUS,
RECORD_START,
RECORD_STEP,
RECORD_STOP,
type RecordCancelMessage,
type RecordCaptureStatusMessage,
type RecordFinishMessage,
type RecordQueryResponse,
type RecordStartAck,
type RecordStartMessage,
type RecordStepPayload,
type RecordStopMessage,
} from "@/lib/record-bridge";
import {
armAllDocumentsForTab,
armDocumentCapture,
cancelDocumentCapture,
type ArmedDocument,
listInjectableFrames,
type RecordFrameSendDeps,
stopDocumentCapture,
} from "@/lib/record-frame-arming";
import { DEFAULT_MAX_PAGE_TOKENS } from "@/lib/record-constants";
import {
applyTargetMatching,
buildTraceV3,
cancelPendingSettles,
captureAndRegisterObservation,
clearPendingRedirectLanding,
createObservationState,
flushPendingRedirectLanding,
flushPendingSettles,
inferMissingPostStates,
type RecordingObservationState,
rememberStepOnPage,
scheduleDraftSettle,
scheduleRedirectLandingFlush,
settleUnsettledDrafts,
} from "@/lib/record-observation";
import { appendRecordedPayload, observeRecordedNavigation } from "@/lib/recording-step-buffer";
import { reduceTraceSteps, resolveTraceStartUrl } from "@/lib/trace-reducer";
import type { SessionManager } from "@/session-manager/manager";
import { EXTENSION_VERSION } from "@/transport/handshake";
import type {
DraftTraceStep,
FrameCaptureFailure,
FrameCaptureInfo,
FrameCaptureStatus,
RecordAwaitParams,
RecordAwaitResult,
RecordStartParams,
@@ -29,6 +61,7 @@ import type {
RecordStopParams,
RecordStopResult,
RpcError,
StopReason,
Trace,
} from "@/transport/types";
import { handleNavigate } from "./navigation";
@@ -58,6 +91,37 @@ interface ActiveRecording {
currentUrl?: string;
pendingNavigation: boolean;
pendingNavigationDeadline?: number;
observation: RecordingObservationState;
stoppedBy: StopReason;
maxPageTokens: number;
redactValues: boolean;
/** Documents whose capture acknowledged START and still need STOP. */
armedDocuments: Map<string, ArmedDocument>;
frameCaptureFailures: FrameCaptureFailure[];
frameCaptureStatus: FrameCaptureStatus;
/** When false, ignore subframe rearm requests during finish. */
acceptingNewFrames: boolean;
/** Rearms already sending START when finish begins. */
rearmCallbacks: Set<Promise<void>>;
/** Navigation callbacks tracked from event receipt through action enqueue. */
navigationCallbacks: Set<Promise<void>>;
/** Synchronous intake gate closed only after finish drains to stability. */
acceptingNavigation: boolean;
/**
* Serializes step appends with navigation observation so a click is always
* in `steps` before a same-turn `webNavigation` tries to annotate it.
*/
actionQueue: Promise<void>;
/**
* CDP runner captured at `record_start`. The message listeners are attached
* once at service-worker startup, where no runner exists yet, so observation
* capture must go through the recording instead of the listener deps.
*/
cdp?: CdpRunner;
}
function enqueueRecordingAction(recording: ActiveRecording, task: () => Promise<void>): void {
recording.actionQueue = recording.actionQueue.then(task, task).catch(() => {});
}
const recordings = new Map<string, ActiveRecording>();
@@ -78,6 +142,9 @@ function sleep(ms: number): Promise<void> {
return new Promise((resolve) => setTimeout(resolve, ms));
}
/** Recording producer version mirrored into trace.recorder.bsk. */
export const BSK_TRACE_VERSION = EXTENSION_VERSION;
/** Injectable http(s) landing page when `tool.record_start` omits `url`. */
export const RECORD_DEFAULT_START_URL = "https://example.com/";
@@ -155,16 +222,95 @@ async function sendRecordStartWithAck(
}
function buildTrace(recording: ActiveRecording): Trace {
const { pages, steps } = reduceTraceSteps(recording.steps, recording.startUrl);
const startUrl = resolveTraceStartUrl(recording.steps, recording.startUrl, pages);
return {
const frameCapture: FrameCaptureInfo | undefined =
recording.frameCaptureFailures.length > 0 || recording.frameCaptureStatus === "partial"
? {
status: recording.frameCaptureStatus,
...(recording.frameCaptureFailures.length > 0
? { failures: recording.frameCaptureFailures }
: {}),
}
: undefined;
return buildTraceV3({
obs: recording.observation,
steps: recording.steps,
startedAt: recording.startedAt,
...(recording.purpose ? { purpose: recording.purpose } : {}),
recorded_at: new Date().toISOString(),
started_at: recording.startedAt,
entry: { start_url: startUrl },
pages,
steps,
startUrl: recording.startUrl,
stoppedBy: recording.stoppedBy,
bskVersion: BSK_TRACE_VERSION,
...(frameCapture ? { frameCapture } : {}),
});
}
function recordFrameFailure(recording: ActiveRecording, failure: FrameCaptureFailure): void {
recording.frameCaptureFailures.push(failure);
recording.frameCaptureStatus = "partial";
}
function enrichRecordedStep(
recording: ActiveRecording,
step: RecordStepPayload,
): RecordStepPayload {
const enriched: RecordStepPayload = { ...step };
if (enriched.geometry && enriched.documentToken) {
const owner = recording.observation.documentTokenOwners.get(enriched.documentToken);
if (owner !== undefined) {
enriched.geometry = { ...enriched.geometry, ownerFrameBackendNodeId: owner };
}
}
if (!enriched.page_url && recording.currentUrl) {
enriched.page_url = recording.currentUrl;
}
return enriched;
}
async function publishCaptureStatus(
recording: ActiveRecording,
deps: RecordDeps,
): Promise<void> {
if (!deps.broadcastCaptureStatus) return;
const msg: RecordCaptureStatusMessage = {
type: RECORD_CAPTURE_STATUS,
status: recording.frameCaptureStatus,
...(recording.frameCaptureFailures.length > 0
? { failures: recording.frameCaptureFailures }
: {}),
};
try {
await deps.broadcastCaptureStatus(recording.tabId, msg);
} catch {
// Overlay is best-effort.
}
}
async function processRecordedStep(
recording: ActiveRecording,
deps: RecordDeps,
draftIndex: number,
): Promise<void> {
const cdp = recording.cdp ?? deps.cdp;
if (!cdp) return;
const draft = recording.steps[draftIndex];
if (!draft) return;
const stepId = draftIndex + 1;
if (draft.op === "navigate") {
draft.preStateId = recording.observation.lastSettled?.stateId;
} else {
applyTargetMatching(recording.observation, draft, stepId);
}
rememberStepOnPage(recording.observation, draft, stepId);
scheduleDraftSettle(
recording.observation,
draftIndex,
stepId,
cdp,
recording.tabId,
deps.tabsApi,
recording.steps,
);
}
export interface RecordDeps {
@@ -173,17 +319,67 @@ export interface RecordDeps {
tabId: number,
msg: RecordStartMessage | RecordStopMessage | RecordCancelMessage,
): Promise<unknown>;
sendToDocument?(
tabId: number,
documentId: string,
msg: RecordStartMessage | RecordStopMessage | RecordCancelMessage,
): Promise<unknown>;
getAllFrames?(tabId: number): Promise<
Array<{
frameId: number;
documentId?: string;
url?: string;
parentFrameId: number;
}>
>;
broadcastCaptureStatus?(
tabId: number,
msg: RecordCaptureStatusMessage,
): Promise<void>;
bypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>;
cdp?: CdpRunner;
signal?: AbortSignal;
}
function frameSendDeps(deps: RecordDeps): RecordFrameSendDeps {
return {
sendToDocument: (tabId, documentId, msg) => {
if (deps.sendToDocument) return deps.sendToDocument(tabId, documentId, msg);
return deps.sendToTab(tabId, msg);
},
getAllFrames: async (tabId) => {
if (deps.getAllFrames) return deps.getAllFrames(tabId);
if (!chrome.webNavigation?.getAllFrames) return [];
const frames = await chrome.webNavigation.getAllFrames({ tabId });
return (frames ?? []).map((frame) => ({
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
parentFrameId: frame.parentFrameId,
}));
},
};
}
let defaultDeps: RecordDeps | null = null;
function getDefaultDeps(): RecordDeps {
if (!defaultDeps) {
defaultDeps = {
tabsApi: chromeTabsApi,
sendToTab: (tabId, msg) => chrome.tabs.sendMessage(tabId, msg),
sendToDocument: (tabId, documentId, msg) =>
chrome.tabs.sendMessage(tabId, msg, { documentId }),
getAllFrames: async (tabId) => {
if (!chrome.webNavigation?.getAllFrames) return [];
const frames = await chrome.webNavigation.getAllFrames({ tabId });
return (frames ?? []).map((frame) => ({
frameId: frame.frameId,
documentId: frame.documentId,
url: frame.url,
parentFrameId: frame.parentFrameId,
}));
},
broadcastCaptureStatus: (tabId, msg) => chrome.tabs.sendMessage(tabId, msg, { frameId: 0 }),
};
}
return defaultDeps;
@@ -241,16 +437,29 @@ export function releaseBrowserObservationListenersIfIdle(): void {
detachBrowserObservation = null;
}
export function attachRecordStepListener(): () => void {
export function attachRecordStepListener(deps: RecordDeps = getDefaultDeps()): () => void {
const listener = (
message: unknown,
_sender: chrome.runtime.MessageSender,
sender: chrome.runtime.MessageSender,
_sendResponse: () => void,
) => {
if (!isRecordStepMessage(message)) return false;
for (const recording of recordings.values()) {
if (recording.requestId !== message.requestId) continue;
appendRecordedPayload(recording, message.step);
enqueueRecordingAction(recording, async () => {
const cdp = recording.cdp ?? deps.cdp;
await flushPendingRedirectLanding(
recording.observation,
recording.steps,
cdp,
recording.tabId,
deps.tabsApi,
{ cancelled: () => recording.settled },
);
const enriched = enrichRecordedStep(recording, message.step);
const draftIndex = appendRecordedPayload(recording, enriched);
if (draftIndex !== null) await processRecordedStep(recording, deps, draftIndex);
});
return false;
}
return false;
@@ -277,7 +486,11 @@ export function attachRecordFinishListener(deps: RecordDeps = getDefaultDeps()):
function findRecordingByTabId(tabId: number): ActiveRecording | null {
for (const recording of recordings.values()) {
if (recording.tabId === tabId && !recording.settled) return recording;
if (recording.settled) continue;
if (recording.tabId === tabId) return recording;
for (const armed of recording.armedDocuments.values()) {
if (armed.tabId === tabId) return recording;
}
}
return null;
}
@@ -325,71 +538,135 @@ async function clearRearmTimersForRecording(
}
}
async function stopRecordingOnAllAgentTabs(
async function stopRecordingOnAllDocuments(
recording: ActiveRecording,
deps: RecordDeps,
): Promise<void> {
recording.acceptingNewFrames = false;
const stopMsg: RecordStopMessage = { type: RECORD_STOP, requestId: recording.requestId };
let tabIds = [recording.tabId];
try {
const tabs = await deps.tabsApi.query({ windowId: recording.agentWindowId });
tabIds = [
...new Set([
recording.tabId,
...tabs.flatMap((tab) => (typeof tab.id === "number" ? [tab.id] : [])),
]),
];
} catch {
// Fall back to the current recording tab.
}
for (const tabId of tabIds) {
try {
const response = await deps.sendToTab(tabId, stopMsg);
if (tabId === recording.tabId && !isRecordStartAck(response)) {
throw new Error("content script did not confirm recorded steps");
}
} catch {
if (tabId === recording.tabId) {
throw new Error("failed to flush recorded steps");
}
const frameDeps = frameSendDeps(deps);
const armed = [...recording.armedDocuments.values()];
for (const doc of armed) {
const result = await stopDocumentCapture(doc, stopMsg, frameDeps);
if ("ok" in result) {
recording.armedDocuments.delete(doc.documentId);
continue;
}
if (deps.bypassOverlay) {
try {
await deps.bypassOverlay(tabId, false);
} catch {
// Best-effort cleanup.
}
recordFrameFailure(recording, result);
}
if (recording.armedDocuments.size > 0) {
for (const doc of [...recording.armedDocuments.values()]) {
recordFrameFailure(
recording,
{
reason: "flush_failed",
frame_id: doc.frameId,
document_id: doc.documentId,
url: doc.url,
detail: "document still armed after STOP",
},
);
}
recording.armedDocuments.clear();
}
await publishCaptureStatus(recording, deps);
if (deps.bypassOverlay) {
try {
await deps.bypassOverlay(recording.tabId, false);
} catch {
// Best-effort cleanup.
}
}
}
async function rearmRecording(
async function cancelRecordingOnAllDocuments(
recording: ActiveRecording,
deps: RecordDeps,
): Promise<void> {
const cancelMsg: RecordCancelMessage = { type: RECORD_CANCEL, requestId: recording.requestId };
const frameDeps = frameSendDeps(deps);
for (const doc of [...recording.armedDocuments.values()]) {
await cancelDocumentCapture(doc, cancelMsg, frameDeps);
}
recording.armedDocuments.clear();
}
async function rearmRecordingDocuments(
recording: ActiveRecording,
targetTabId: number,
deps: RecordDeps,
targetDocumentId?: string,
): Promise<boolean> {
// Do NOT toggle automation-bypass here: each retry used to increment the
// content-script counter, and a single stop decrement left the ControlOverlay
// stuck with pointer-events:none (page usable, Interrupt dead). RecordOverlay
// already hides the control chrome while activeRecord is set.
for (let attempt = 0; attempt < RECORD_REARM_MAX_ATTEMPTS; attempt += 1) {
const startMsg: RecordStartMessage = {
type: RECORD_START,
requestId: recording.requestId,
startedAtMs: recording.startedAtMs,
};
try {
await sendRecordStartWithAck(targetTabId, startMsg, deps.sendToTab);
recording.tabId = targetTabId;
return true;
} catch {
let resolveTracked!: () => void;
const tracked = new Promise<void>((resolve) => {
resolveTracked = resolve;
});
recording.rearmCallbacks.add(tracked);
const frameDeps = frameSendDeps(deps);
try {
const attemptLimit = targetDocumentId ? 1 : RECORD_REARM_MAX_ATTEMPTS;
for (let attempt = 0; attempt < attemptLimit; attempt += 1) {
if (recording.settled || recording.finishing || !recording.acceptingNewFrames) return false;
const { injectable, failures } = await listInjectableFrames(targetTabId, frameDeps);
for (const failure of failures) recordFrameFailure(recording, failure);
const targets = targetDocumentId
? injectable.filter((frame) => frame.documentId === targetDocumentId)
: injectable.filter((frame) => !recording.armedDocuments.has(frame.documentId ?? ""));
if (targets.length === 0) {
if (targetDocumentId) {
recordFrameFailure(recording, {
reason: "rearm_failed",
frame_id: 0,
document_id: targetDocumentId,
detail: "document not injectable",
});
await publishCaptureStatus(recording, deps);
return false;
}
recording.tabId = targetTabId;
return injectable.some((frame) => recording.armedDocuments.has(frame.documentId ?? ""));
}
let armedAny = false;
for (const frame of targets) {
const startMsg: RecordStartMessage = {
type: RECORD_START,
requestId: recording.requestId,
startedAtMs: recording.startedAtMs,
showOverlay: frame.frameId === 0,
frameMode: frame.frameId === 0 ? "top" : "child",
};
const result = await armDocumentCapture(targetTabId, frame, startMsg, frameDeps);
if (result.armed) {
recording.armedDocuments.set(result.armed.documentId, result.armed);
armedAny = true;
}
if (result.failure) recordFrameFailure(recording, result.failure);
}
if (armedAny) {
recording.tabId = targetTabId;
await publishCaptureStatus(recording, deps);
if (deps.cdp && recording.acceptingNewFrames) {
void captureAndRegisterObservation(
recording.observation,
deps.cdp,
targetTabId,
deps.tabsApi,
).catch(() => {});
}
return true;
}
if (attempt + 1 < RECORD_REARM_MAX_ATTEMPTS) {
await sleep(RECORD_REARM_RETRY_DELAY_MS);
}
}
await publishCaptureStatus(recording, deps);
return false;
} finally {
recording.rearmCallbacks.delete(tracked);
resolveTracked();
}
return false;
}
function scheduleRearmForTab(tabId: number, deps: RecordDeps): void {
@@ -401,7 +678,7 @@ function scheduleRearmForTab(tabId: number, deps: RecordDeps): void {
rearmTimers.delete(tabId);
void (async () => {
const current = await findRecordingForTab(tabId, deps);
if (current) await rearmRecording(current, tabId, deps);
if (current) await rearmRecordingDocuments(current, tabId, deps);
})();
}, RECORD_REARM_DEBOUNCE_MS),
);
@@ -432,11 +709,86 @@ export function attachRecordTabListener(deps: RecordDeps = getDefaultDeps()): ()
}
export function attachRecordNavigationListener(deps: RecordDeps = getDefaultDeps()): () => void {
const observeMainFrame = (tabId: number, url?: string, causedByAction?: boolean) => {
const observeMainFrame = (
tabId: number,
url?: string,
causedByAction?: boolean,
transitionType?: string,
transitionQualifiers?: string[],
) => {
if (!url) return;
const direct = findRecordingByTabId(tabId);
const candidates = direct
? direct.acceptingNavigation
? [direct]
: []
: [...recordings.values()].filter(
(recording) => !recording.settled && recording.acceptingNavigation,
);
if (candidates.length === 0) return;
let resolveTracked!: () => void;
const tracked = new Promise<void>((resolve) => {
resolveTracked = resolve;
});
for (const candidate of candidates) candidate.navigationCallbacks.add(tracked);
void (async () => {
const recording = await findRecordingForTab(tabId, deps);
if (recording && url) {
observeRecordedNavigation(recording, url, causedByAction);
try {
const recording = await findRecordingForTab(tabId, deps);
if (!recording || !recording.acceptingNavigation || !candidates.includes(recording)) {
return;
}
enqueueRecordingAction(recording, async () => {
const result = observeRecordedNavigation(
recording,
url,
causedByAction,
transitionType,
transitionQualifiers,
);
const cdp = recording.cdp ?? deps.cdp;
if (!cdp) return;
if (result.kind === "coalesce_redirect") {
scheduleRedirectLandingFlush(
recording.observation,
recording.steps,
cdp,
tabId,
deps.tabsApi,
result.url,
{ cancelled: () => recording.settled },
);
return;
}
if (result.kind === "noop") return;
// A concrete navigation supersedes any in-flight redirect coalesce.
clearPendingRedirectLanding(recording.observation);
const draftIndex = result.index;
const lastDraft = recording.steps[draftIndex];
if (!lastDraft) return;
if (result.kind === "appended") {
lastDraft.preStateId = recording.observation.lastSettled?.stateId;
rememberStepOnPage(recording.observation, lastDraft, draftIndex + 1);
}
scheduleDraftSettle(
recording.observation,
draftIndex,
draftIndex + 1,
cdp,
tabId,
deps.tabsApi,
recording.steps,
);
});
} finally {
for (const candidate of candidates) candidate.navigationCallbacks.delete(tracked);
resolveTracked();
}
})();
};
@@ -447,29 +799,36 @@ export function attachRecordNavigationListener(deps: RecordDeps = getDefaultDeps
})();
};
const onSubFrameComplete = (tabId: number, documentId?: string) => {
if (!documentId) return;
void (async () => {
const recording = await findRecordingForTab(tabId, deps);
if (!recording || !recording.acceptingNewFrames) return;
await rearmRecordingDocuments(recording, tabId, deps, documentId);
})();
};
if (chrome.webNavigation?.onCompleted) {
const completedListener = (
details: chrome.webNavigation.WebNavigationFramedCallbackDetails,
) => {
if (details.frameId !== 0) return;
onMainFrameComplete(details.tabId, details.url);
if (details.frameId === 0) {
onMainFrameComplete(details.tabId, details.url);
return;
}
onSubFrameComplete(details.tabId, details.documentId);
};
const committedListener = (
details: chrome.webNavigation.WebNavigationTransitionCallbackDetails,
) => {
if (details.frameId !== 0) return;
const causedByAction =
details.transitionType === "link" || details.transitionType === "form_submit"
? true
: details.transitionType === "typed" ||
details.transitionType === "auto_bookmark" ||
details.transitionType === "generated" ||
details.transitionType === "keyword" ||
details.transitionType === "keyword_generated" ||
details.transitionType === "reload"
? false
: undefined;
observeMainFrame(details.tabId, details.url, causedByAction);
observeMainFrame(
details.tabId,
details.url,
undefined,
details.transitionType,
details.transitionQualifiers,
);
};
chrome.webNavigation.onCompleted.addListener(completedListener);
chrome.webNavigation.onCommitted?.addListener(committedListener);
@@ -487,6 +846,46 @@ export function attachRecordNavigationListener(deps: RecordDeps = getDefaultDeps
return () => chrome.tabs.onUpdated.removeListener(listener);
}
const MAX_FINISH_DRAIN_ROUNDS = 10;
async function drainRecordingToStability(
recording: ActiveRecording,
deps: RecordDeps,
): Promise<boolean> {
const cdp = recording.cdp ?? deps.cdp;
for (let round = 0; round < MAX_FINISH_DRAIN_ROUNDS; round += 1) {
await Promise.all([...recording.navigationCallbacks]);
const actionTail = recording.actionQueue;
await actionTail;
await flushPendingRedirectLanding(
recording.observation,
recording.steps,
cdp,
recording.tabId,
deps.tabsApi,
{ cancelled: () => recording.settled },
);
await flushPendingSettles(recording.observation);
const redirectTail = recording.observation.redirectFlushQueue;
const settleTail = recording.observation.settleQueue;
await Promise.all([redirectTail, settleTail]);
if (
recording.navigationCallbacks.size === 0 &&
recording.actionQueue === actionTail &&
recording.observation.redirectFlushQueue === redirectTail &&
recording.observation.settleQueue === settleTail
) {
// No event or promise continuation can interleave with this synchronous
// check-and-close, so work accepted before the cutoff is fully drained.
recording.acceptingNavigation = false;
return true;
}
}
console.warn("[bsk record] navigation/action queues did not stabilize at stop");
return false;
}
export function attachRecordQueryListener(deps: RecordDeps = getDefaultDeps()): () => void {
const listener = (
message: unknown,
@@ -505,11 +904,15 @@ export function attachRecordQueryListener(deps: RecordDeps = getDefaultDeps()):
sendResponse({ active: false });
return;
}
await rearmRecording(recording, tabId, deps);
await rearmRecordingDocuments(recording, tabId, deps, sender.documentId);
sendResponse({
active: true,
requestId: recording.requestId,
startedAtMs: recording.startedAtMs,
captureStatus: recording.frameCaptureStatus,
...(recording.frameCaptureFailures.length > 0
? { captureFailures: recording.frameCaptureFailures }
: {}),
});
})();
return true;
@@ -527,23 +930,46 @@ async function finishRecordingByRequest(
if (recording.requestId !== requestId || recording.settled) continue;
const match = await findRecordingForTab(tabId, deps);
if (match !== recording) continue;
await finishRecording(sessionId, deps);
await finishRecording(sessionId, deps, "user_finish");
return;
}
}
async function finishRecording(sessionId: string, deps: RecordDeps): Promise<Trace | null> {
async function finishRecording(
sessionId: string,
deps: RecordDeps,
stoppedBy: StopReason,
): Promise<Trace | null> {
const recording = recordings.get(sessionId);
if (!recording || recording.settled || recording.finishing) return null;
if (!recording || recording.settled) return null;
if (recording.finishing) return recording.finishPromise;
recording.finishing = true;
recording.stoppedBy = stoppedBy;
await clearRearmTimersForRecording(recording, deps);
await Promise.all([...recording.rearmCallbacks]);
try {
await stopRecordingOnAllAgentTabs(recording, deps);
await stopRecordingOnAllDocuments(recording, deps);
} catch {
recording.finishing = false;
return null;
}
if (!(await drainRecordingToStability(recording, deps))) {
recording.finishing = false;
return null;
}
cancelPendingSettles(recording.observation);
inferMissingPostStates(recording.steps);
const cdp = recording.cdp ?? deps.cdp;
if (cdp) {
await settleUnsettledDrafts(
recording.observation,
cdp,
recording.tabId,
deps.tabsApi,
recording.steps,
);
}
recording.settled = true;
recordings.delete(sessionId);
@@ -582,6 +1008,8 @@ export async function handleRecordStart(
});
const navigateUrl = params.url ?? RECORD_DEFAULT_START_URL;
const startedAtMs = Date.now();
const maxPageTokens = params.max_page_tokens;
const redactValues = params.redact_values ?? false;
recordings.set(params.session_id, {
requestId,
tabId: target.tabId,
@@ -599,23 +1027,30 @@ export async function handleRecordStart(
currentUrl: navigateUrl,
pendingNavigation: false,
pendingNavigationDeadline: undefined,
observation: createObservationState({ maxPageTokens, redactValues }),
stoppedBy: "user_finish",
maxPageTokens: maxPageTokens ?? DEFAULT_MAX_PAGE_TOKENS,
redactValues,
armedDocuments: new Map(),
frameCaptureFailures: [],
frameCaptureStatus: "complete",
acceptingNewFrames: true,
rearmCallbacks: new Set(),
navigationCallbacks: new Set(),
acceptingNavigation: true,
actionQueue: Promise.resolve(),
...(deps.cdp ? { cdp: deps.cdp } : {}),
});
// Observe navigations for the whole recording lifetime; attach before
// optional navigate so the destination load can rearm capture.
ensureBrowserObservationListeners(deps);
const abortPending = async (notifyContent: boolean) => {
const pending = recordings.get(params.session_id);
recordings.delete(params.session_id);
releaseBrowserObservationListenersIfIdle();
if (notifyContent) {
try {
await deps.sendToTab(target.tabId, {
type: RECORD_CANCEL,
requestId,
});
} catch {
// Content may never have received RECORD_START.
}
if (notifyContent && pending) {
await cancelRecordingOnAllDocuments(pending, deps);
}
if (deps.bypassOverlay) {
try {
@@ -718,10 +1153,28 @@ export async function handleRecordStart(
if (cancelled) return cancelled;
}
const startMsg: RecordStartMessage = { type: RECORD_START, requestId, startedAtMs };
const startMsg: RecordStartMessage = {
type: RECORD_START,
requestId,
startedAtMs,
showOverlay: true,
frameMode: "top",
};
try {
await sendRecordStartWithAck(target.tabId, startMsg, deps.sendToTab);
const frameDeps = frameSendDeps(deps);
const { armed, failures } = await armAllDocumentsForTab(target.tabId, startMsg, frameDeps);
for (const failure of failures) recordFrameFailure(active, failure);
for (const doc of armed) active.armedDocuments.set(doc.documentId, doc);
if (active.armedDocuments.size === 0) {
await abortPending(true);
return {
code: "protocol_error",
message:
"failed to start recording in any frame — reload the BrowserSkill extension, then retry",
};
}
await publishCaptureStatus(active, deps);
} catch {
await abortPending(true);
return {
@@ -731,6 +1184,23 @@ export async function handleRecordStart(
};
}
if (deps.cdp) {
const activeRecording = recordings.get(params.session_id);
if (activeRecording) {
try {
await captureAndRegisterObservation(
activeRecording.observation,
deps.cdp,
target.tabId,
deps.tabsApi,
startUrl,
);
} catch {
// Proceed without initial observation; steps may be dropped by reducer.
}
}
}
{
const cancelled = await abortIfCancelled(true);
if (cancelled) return cancelled;
@@ -755,7 +1225,7 @@ export async function handleRecordStop(
};
}
const trace = await finishRecording(params.session_id, deps);
const trace = await finishRecording(params.session_id, deps, "cli_stop");
if (!trace) {
return {
code: "protocol_error",
+1 -41
View File
@@ -1,6 +1,5 @@
import { type SessionManager, SessionStartCleanupError } from "@/session-manager/manager";
import type { SessionManager } from "@/session-manager/manager";
import type { RpcError } from "@/transport/types";
import { rpcError } from "./errors";
import { clearRecordingForSession } from "./record";
import { isRpcError } from "./shared";
import { returnBorrowedTab, type TabManagementDeps } from "./tabs";
@@ -59,10 +58,6 @@ export interface SessionStartResult {
agent_window_id?: number;
}
export interface SessionStartDeps {
signal?: AbortSignal;
}
export interface SessionStopParams {
session_id: string;
}
@@ -76,7 +71,6 @@ export interface SessionStopResult {
}
export interface SessionStopDeps {
signal?: AbortSignal;
cdp?: {
detachSession(sessionId: string): Promise<void>;
};
@@ -97,7 +91,6 @@ export interface SessionStopDeps {
export async function handleSessionStart(
manager: SessionManager,
params: SessionStartParams,
deps: SessionStartDeps = {},
): Promise<SessionStartResult | RpcError> {
if (!params?.session_id) {
return {
@@ -111,23 +104,9 @@ export async function handleSessionStart(
const ctx = await manager.start(params.session_id, {
size: sizeOrErr,
focused: params.focused,
signal: deps.signal,
});
return { agent_window_id: ctx.agentWindowId };
} catch (err) {
if (err instanceof SessionStartCleanupError) {
return rpcError("protocol_error", "cleanup_failed", err.message, {
resource_type: "agent_window",
resource_id: err.windowId,
});
}
if (
typeof err === "object" &&
err !== null &&
(err as { name?: string }).name === "AbortError"
) {
return { code: "cancelled", message: "session_start aborted" };
}
// chrome.windows.create / SessionManager failures are not CDP
// failures (§4.5 reserves cdp_failed for raw CDP errors). Surface
// them as protocol_error so the CLI maps to the right exit code
@@ -178,9 +157,6 @@ export async function handleSessionStop(
message: `session ${params.session_id} unknown`,
};
}
if (deps.signal?.aborted) {
return { code: "cancelled", message: "session_stop aborted before teardown" };
}
// Step 1: auto-return borrowed tabs. Iterate over a snapshot of the
// ids so deletions during iteration do not break the Map iterator.
@@ -191,7 +167,6 @@ export async function handleSessionStop(
try {
const tabManagement = {
...(deps.tabManagement ?? {}),
signal: deps.signal,
isAgentWindowId:
deps.tabManagement?.isAgentWindowId ??
((windowId: number) => manager.findByWindowId(windowId) !== null),
@@ -219,17 +194,6 @@ export async function handleSessionStop(
}
}
if (deps.signal?.aborted) {
const nonCancelFailures = returnFailures?.filter((failure) => failure.code !== "cancelled");
if (nonCancelFailures && nonCancelFailures.length > 0) {
return {
...(returnedTabIds.length > 0 ? { returned_tab_ids: returnedTabIds } : {}),
return_failures: nonCancelFailures,
};
}
return { code: "cancelled", message: "session_stop aborted during tab return" };
}
const result: SessionStopResult = {};
if (returnedTabIds.length > 0) result.returned_tab_ids = returnedTabIds;
if (returnFailures && returnFailures.length > 0) {
@@ -249,10 +213,6 @@ export async function handleSessionStop(
// Step 3: detach CDP sessions this session opened (no-op if none).
await deps.cdp?.detachSession(params.session_id);
if (deps.signal?.aborted) {
return { code: "cancelled", message: "session_stop aborted before window close" };
}
// Step 4: close the Agent Window and drop the context.
await manager.stop(params.session_id);
+2 -24
View File
@@ -6,8 +6,7 @@
// exactly the same sandbox + visibility rules as the M6 observation
// handlers (review parity).
import type { CdpDebuggee, DialogCursor } from "@/browser-driver/chromium-cdp";
import type { CdpFrameGraph, CdpTarget } from "@/browser-driver/frame-graph";
import type { DialogCursor } from "@/browser-driver/chromium-cdp";
import type { SessionContext, SessionManager } from "@/session-manager/manager";
import { normaliseRef } from "@/session-manager/ref-store";
import type { ConsoleResult, JavaScriptDialogInfo, RpcError } from "@/transport/types";
@@ -52,11 +51,9 @@ export type { DialogCursor };
*/
export interface CdpRunner {
send<T = unknown>(tabId: number, method: string, params?: object): Promise<T>;
sendToTarget?<T = unknown>(target: CdpTarget, method: string, params?: object): Promise<T>;
getFrameGraph?(tabId: number): Promise<CdpFrameGraph>;
ensureAttachedToUrl?(tabId: number, expectedUrl: string | undefined): Promise<void>;
trackSessionTab?(sessionId: string, tabId: number): void;
onEvent?(handler: (source: CdpDebuggee, method: string, params: unknown) => void): {
onEvent?(handler: (source: chrome.debugger.Debuggee, method: string, params: unknown) => void): {
dispose(): void;
};
dialogCursor?(tabId: number): DialogCursor;
@@ -71,25 +68,6 @@ export interface CdpRunner {
): ConsoleResult;
}
export function sendToCdpTarget<T = unknown>(
cdp: CdpRunner,
target: CdpTarget,
method: string,
params?: object,
): Promise<T> {
if (target.sessionId && cdp.sendToTarget) {
return cdp.sendToTarget<T>(target, method, params);
}
return cdp.send<T>(target.tabId, method, params);
}
export function cdpRunnerForTarget(cdp: CdpRunner, target: CdpTarget): CdpRunner {
return {
send: <T = unknown>(_tabId: number, method: string, params?: object) =>
sendToCdpTarget<T>(cdp, target, method, params),
};
}
export interface BufferedReadBounds {
since: number | undefined;
limit: number;
+5 -17
View File
@@ -1,5 +1,5 @@
// Snapshot ref resolution — normalise `@eN` / `eN`, look up
// compound CDP node identity via the session RefStore, and build stable
// `backendNodeId` via the session RefStore, and build stable
// `ref_not_found` errors for hard-failure tool paths.
import type { SessionContext } from "@/session-manager/manager";
@@ -10,13 +10,6 @@ import { rpcError } from "./errors";
export interface SnapshotRefLookup {
backendNodeId: number;
refKey: string;
frameId?: string;
cdpSessionId?: string;
}
function refEntryForTab(ctx: SessionContext, refKey: string, tabId: number) {
const entry = ctx.refStore.resolveEntry(refKey);
return entry?.tabId === tabId ? entry : null;
}
/**
@@ -30,14 +23,9 @@ export function lookupSnapshotRef(
tabId: number,
): SnapshotRefLookup | null {
const refKey = normaliseRef(ref);
const entry = refEntryForTab(ctx, refKey, tabId);
if (!entry) return null;
return {
backendNodeId: entry.backendNodeId,
refKey,
...(entry.frameId ? { frameId: entry.frameId } : {}),
...(entry.cdpSessionId ? { cdpSessionId: entry.cdpSessionId } : {}),
};
const backendNodeId = ctx.refStore.resolve(refKey, { tabId });
if (backendNodeId === null) return null;
return { backendNodeId, refKey };
}
/**
@@ -49,7 +37,7 @@ export function resolveSnapshotRef(
ctx: SessionContext,
ref: string,
tabId: number,
): SnapshotRefLookup | RpcError {
): { backendNodeId: number; refKey: string } | RpcError {
const looked = lookupSnapshotRef(ctx, ref, tabId);
if (looked === null) {
return rpcError(
+7 -115
View File
@@ -157,7 +157,6 @@ export interface ChromeWindowsApi {
get(windowId: number): Promise<chrome.windows.Window>;
getLastFocused(filters?: chrome.windows.QueryOptions): Promise<chrome.windows.Window>;
create(props: chrome.windows.CreateData): Promise<chrome.windows.Window | undefined>;
remove(windowId: number): Promise<void>;
}
export const chromeWindowsApi: ChromeWindowsApi = {
@@ -165,7 +164,6 @@ export const chromeWindowsApi: ChromeWindowsApi = {
getLastFocused: (filters) =>
filters ? chrome.windows.getLastFocused(filters) : chrome.windows.getLastFocused(),
create: (p) => chrome.windows.create(p),
remove: (id) => chrome.windows.remove(id),
};
export interface AgentOverlayResetApi {
@@ -217,9 +215,7 @@ export async function handleTabList(
manager: SessionManager,
params: TabListParams,
api: ChromeTabsApi = chromeTabsApi,
signal?: AbortSignal,
): Promise<TabListResult | RpcError> {
if (signal?.aborted) return { code: "cancelled", message: "tab_list aborted" };
if (!params || typeof params.session_id !== "string" || params.session_id.length === 0) {
return {
code: "invalid_params",
@@ -253,7 +249,6 @@ export async function handleTabList(
const myAgentWindowId = ctx.agentWindowId;
const allTabs = await api.query({});
if (signal?.aborted) return { code: "cancelled", message: "tab_list aborted" };
const tabs: TabInfo[] = [];
for (const t of allTabs) {
if (typeof t.id !== "number") continue;
@@ -387,12 +382,7 @@ async function createTabAndCleanup(
try {
await getTabsApi(deps).remove(tab.id);
} catch (cleanupErr) {
return rpcError(
"protocol_error",
"cleanup_failed",
`tab_create aborted but cleanup of tab ${tab.id} failed: ${describeError(cleanupErr)}`,
{ resource_type: "tab", resource_id: tab.id },
);
console.debug("[bsk tab_create] cleanup after abort failed", cleanupErr);
}
}
return { code: "cancelled", message: "tab_create aborted" };
@@ -703,17 +693,7 @@ async function moveTabForBorrow(
try {
await tabsApi.move(tabId, { windowId: originalWindowId, index: originalIndex });
} catch (rollbackErr) {
return rpcError(
"protocol_error",
"cleanup_failed",
`tab_borrow aborted but rollback of tab ${tabId} to window ${originalWindowId} failed: ${describeError(rollbackErr)}`,
{
resource_type: "tab",
resource_id: tabId,
original_window_id: originalWindowId,
original_index: originalIndex,
},
);
console.debug("[bsk tab_borrow] rollback move failed", rollbackErr);
}
return { code: "cancelled", message: "tab_borrow aborted" };
}
@@ -845,12 +825,6 @@ export interface ReturnOutcome {
fallback: boolean;
}
interface FallbackWindowTarget {
windowId: number;
index: number;
created: boolean;
}
function describeError(err: unknown): string {
return err instanceof Error ? err.message : String(err);
}
@@ -859,13 +833,10 @@ async function chooseFallbackWindow(
ctx: SessionContext,
windowsApi: ChromeWindowsApi,
isAgentWindowId: (windowId: number) => boolean,
signal?: AbortSignal,
): Promise<FallbackWindowTarget | RpcError> {
): Promise<{ windowId: number; index: number } | RpcError> {
let lastFocusedError: unknown;
try {
const last = await windowsApi.getLastFocused({ windowTypes: ["normal"] });
const cancelled = aborted(signal, "tab_return");
if (cancelled) return cancelled;
const lastId = typeof last?.id === "number" ? last.id : null;
// Never relocate a user tab into *any* session's Agent Window: this
// session's is excluded explicitly, and other sessions' are excluded
@@ -875,15 +846,12 @@ async function chooseFallbackWindow(
// session write to it and, worse, see it destroyed when that session
// stops and closes its window.
if (lastId !== null && lastId !== ctx.agentWindowId && !isAgentWindowId(lastId)) {
return { windowId: lastId, index: -1, created: false };
return { windowId: lastId, index: -1 };
}
} catch (err) {
lastFocusedError = err;
}
const cancelled = aborted(signal, "tab_return");
if (cancelled) return cancelled;
try {
const created = await windowsApi.create({
url: NEW_TAB_DEFAULT_URL,
@@ -895,20 +863,7 @@ async function chooseFallbackWindow(
message: "tab_return: failed to create fallback window",
};
}
if (aborted(signal, "tab_return")) {
try {
await windowsApi.remove(created.id);
} catch (cleanupErr) {
return rpcError(
"protocol_error",
"cleanup_failed",
`tab_return aborted but cleanup of fallback window ${created.id} failed: ${describeError(cleanupErr)}`,
{ resource_type: "window", resource_id: created.id },
);
}
return { code: "cancelled", message: "tab_return aborted" };
}
return { windowId: created.id, index: 0, created: true };
return { windowId: created.id, index: 0 };
} catch (err) {
const suffix = lastFocusedError
? ` (after getLastFocused failed: ${describeError(lastFocusedError)})`
@@ -920,25 +875,6 @@ async function chooseFallbackWindow(
}
}
async function cleanupUnusedFallbackWindow(
windowsApi: ChromeWindowsApi,
target: FallbackWindowTarget,
reason: string,
): Promise<RpcError | null> {
if (!target.created) return null;
try {
await windowsApi.remove(target.windowId);
return null;
} catch (cleanupErr) {
return rpcError(
"protocol_error",
"cleanup_failed",
`${reason}; cleanup of fallback window ${target.windowId} failed: ${describeError(cleanupErr)}`,
{ resource_type: "window", resource_id: target.windowId },
);
}
}
function resetAgentOverlaysInReturnedTab(
ctx: SessionContext,
tabId: number,
@@ -975,13 +911,10 @@ export async function returnBorrowedTab(
const tabsApi = getTabsApi(deps);
const windowsApi = getWindowsApi(deps);
const isAgentWindowId = getIsAgentWindowId(deps);
const alreadyCancelled = aborted(deps.signal, "tab_return");
if (alreadyCancelled) return alreadyCancelled;
let targetWindowId = entry.originalWindowId;
let targetIndex = entry.originalIndex;
let fallback = false;
let fallbackTarget: FallbackWindowTarget | null = null;
// Check the original window is still around.
let originalAlive = true;
@@ -991,30 +924,14 @@ export async function returnBorrowedTab(
console.debug("[bsk tab_return] original window gone, falling back", err);
originalAlive = false;
}
const cancelledAfterLookup = aborted(deps.signal, "tab_return");
if (cancelledAfterLookup) return cancelledAfterLookup;
if (!originalAlive) {
fallback = true;
const target = await chooseFallbackWindow(ctx, windowsApi, isAgentWindowId, deps.signal);
const target = await chooseFallbackWindow(ctx, windowsApi, isAgentWindowId);
if ("code" in target) return target;
fallbackTarget = target;
targetWindowId = target.windowId;
targetIndex = target.index;
}
const cancelledBeforeMove = aborted(deps.signal, "tab_return");
if (cancelledBeforeMove) {
if (fallbackTarget) {
const cleanupError = await cleanupUnusedFallbackWindow(
windowsApi,
fallbackTarget,
"tab_return aborted before moving the borrowed tab",
);
if (cleanupError) return cleanupError;
}
return cancelledBeforeMove;
}
try {
const moved = await tabsApi.move(tabId, {
windowId: targetWindowId,
@@ -1030,33 +947,14 @@ export async function returnBorrowedTab(
fallback,
};
} catch (err) {
if (fallbackTarget) {
const cleanupError = await cleanupUnusedFallbackWindow(
windowsApi,
fallbackTarget,
`tab_return could not move tab ${tabId}`,
);
if (cleanupError) return cleanupError;
}
const cancelledAfterMoveFailure = aborted(deps.signal, "tab_return");
if (cancelledAfterMoveFailure) return cancelledAfterMoveFailure;
if (!fallback) {
const target = await chooseFallbackWindow(ctx, windowsApi, isAgentWindowId, deps.signal);
const target = await chooseFallbackWindow(ctx, windowsApi, isAgentWindowId);
if ("code" in target) {
return {
code: "cdp_failed",
message: `tab_return: chrome.tabs.move failed: ${describeError(err)}; fallback failed: ${target.message}`,
};
}
const cancelledBeforeFallbackMove = aborted(deps.signal, "tab_return");
if (cancelledBeforeFallbackMove) {
const cleanupError = await cleanupUnusedFallbackWindow(
windowsApi,
target,
"tab_return aborted before fallback move",
);
return cleanupError ?? cancelledBeforeFallbackMove;
}
try {
const moved = await tabsApi.move(tabId, {
windowId: target.windowId,
@@ -1072,12 +970,6 @@ export async function returnBorrowedTab(
fallback: true,
};
} catch (fallbackErr) {
const cleanupError = await cleanupUnusedFallbackWindow(
windowsApi,
target,
`tab_return fallback move for tab ${tabId} failed`,
);
if (cleanupError) return cleanupError;
return {
code: "cdp_failed",
message: `tab_return: chrome.tabs.move failed: ${describeError(err)}; fallback move failed: ${describeError(fallbackErr)}`,
@@ -18,8 +18,6 @@ function fakeSnapshotReply() {
"auto",
"type",
"password",
"value",
"hunter2",
];
const i = (s: string) => S.indexOf(s);
return {
@@ -31,8 +29,7 @@ function fakeSnapshotReply() {
nodeType: [1, 1, 1, 1],
nodeName: [i("html"), i("body"), i("div"), i("input")],
backendNodeId: [10, 11, 12, 13],
attributes: [[], [], [], [i("type"), i("password"), i("value"), i("hunter2")]],
inputValue: { index: [3], value: [i("hunter2")] },
attributes: [[], [], [], [i("type"), i("password")]],
},
layout: {
nodeIndex: [1, 2, 3],
@@ -220,8 +217,7 @@ function makeCdp(snapshot: unknown) {
{
state: "filled",
sensitive: true,
defaultValue: "hunter2",
value: "hunter2",
defaultValue: "",
placeholder: "secret",
},
],
@@ -266,11 +262,10 @@ describe("captureViewModel", () => {
expect(input).toMatchObject({
formState: "filled",
formDefaultValue: "",
formPlaceholder: "secret",
attrs: { type: "password" },
});
expect(input?.formValue).toBeUndefined();
expect(input?.formDefaultValue).toBeUndefined();
expect(cdp.send).toHaveBeenCalledWith(
4,
"Runtime.evaluate",
@@ -682,7 +677,7 @@ describe("captureViewModel", () => {
expect(nodes.find((n) => n.backendNodeId === 11)?.cursor).toBe("auto");
});
it("subtracts scroll offset to make rects viewport-relative", async () => {
it("preserves document rects while deriving viewport rects after scroll", async () => {
const cdp = {
send: vi.fn(async (_t: number, method: string) => {
if (method === "DOMSnapshot.enable") return {};
@@ -696,9 +691,10 @@ describe("captureViewModel", () => {
}) as unknown as <T>(tabId: number, method: string, params?: object) => Promise<T>,
};
const { nodes } = await captureViewModel(cdp, 4);
const div = nodes.find((n) => n.backendNodeId === 12);
expect(div?.localRect?.y).toBe(-200);
expect(div?.rect).toMatchObject({ y: 0, h: 600 });
const input = nodes.find((n) => n.backendNodeId === 13);
expect(input?.documentRect).toEqual({ x: 400, y: 300, w: 200, h: 40 });
expect(input?.localRect).toEqual({ x: 400, y: 100, w: 200, h: 40 });
expect(input?.rect).toEqual({ x: 400, y: 100, w: 200, h: 40 });
});
it("normalizes device-pixel bounds by devicePixelRatio", async () => {
@@ -842,7 +838,6 @@ describe("captureViewModel", () => {
strings: S,
documents: [
{
frameId: "main-frame",
scrollOffsetX: 0,
scrollOffsetY: 0,
nodes: {
@@ -870,7 +865,6 @@ describe("captureViewModel", () => {
},
// documents[1]: the iframe's sub-document with a login input
{
frameId: "child-frame",
scrollOffsetX: 0,
scrollOffsetY: 0,
nodes: {
@@ -900,14 +894,7 @@ describe("captureViewModel", () => {
throw new Error(method);
}) as unknown as <T>(tabId: number, method: string, params?: object) => Promise<T>,
};
const {
nodes,
iframeNodes,
frameNodes,
frameOwnerBackendNodeIds,
frameParentIds,
rootFrameId,
} = await captureViewModel(cdp, 4);
const { nodes, iframeNodes } = await captureViewModel(cdp, 4);
// Main frame has 4 nodes; iframe is included
expect(nodes.find((n) => n.backendNodeId === 13)?.tag).toBe("iframe");
// iframeNodes keyed by the <iframe> element's backendNodeId=13
@@ -920,13 +907,9 @@ describe("captureViewModel", () => {
expect(input?.ownerFrameBackendNodeId).toBe(13);
expect(input?.localRect).toEqual({ x: 0, y: 0, w: 200, h: 40 });
expect(input?.rect).toEqual({ x: 100, y: 300, w: 200, h: 40 });
expect(rootFrameId).toBe("main-frame");
expect(frameNodes?.get("child-frame")).toBe(subNodes);
expect(frameOwnerBackendNodeIds?.get("child-frame")).toBe(13);
expect(frameParentIds?.get("child-frame")).toBe("main-frame");
});
it("captures iframe documents without requiring owner-frame geometry", async () => {
it("clips iframe sub-document rects to the iframe viewport in top coordinates", async () => {
const S = [
"html",
"body",
@@ -954,10 +937,16 @@ describe("captureViewModel", () => {
contentDocumentIndex: { index: [2], value: [1] },
},
layout: {
nodeIndex: [1],
styles: [[i("static"), i("auto")]],
bounds: [[0, 0, 1000, 800]],
paintOrders: [0],
nodeIndex: [1, 2],
styles: [
[i("static"), i("auto")],
[i("static"), i("auto")],
],
bounds: [
[0, 0, 1000, 800],
[950, 760, 200, 100],
],
paintOrders: [0, 1],
},
},
{
@@ -995,7 +984,7 @@ describe("captureViewModel", () => {
const input = iframeNodes.get(13)?.find((n) => n.backendNodeId === 101);
expect(input?.localRect).toEqual({ x: 20, y: 20, w: 120, h: 60 });
expect(input?.rect).toBeNull();
expect(input?.rect).toEqual({ x: 970, y: 780, w: 30, h: 20 });
});
it("recursively normalizes nested iframe sub-documents", async () => {
@@ -1090,10 +1079,8 @@ describe("captureViewModel", () => {
const nestedIframe = iframeNodes.get(13)?.find((n) => n.backendNodeId === 23);
const input = iframeNodes.get(23)?.find((n) => n.backendNodeId === 31);
expect(nestedIframe?.localRect).toEqual({ x: 5, y: 6, w: 100, h: 80 });
expect(nestedIframe?.rect).toEqual({ x: 15, y: 26, w: 100, h: 80 });
expect(input?.ownerFrameBackendNodeId).toBe(23);
expect(input?.localRect).toEqual({ x: 1, y: 2, w: 40, h: 20 });
expect(input?.rect).toEqual({ x: 16, y: 28, w: 40, h: 20 });
});
@@ -1,135 +0,0 @@
import { describe, expect, it, vi } from "vitest";
import type { CdpRunner } from "../../shared";
import type { CapturedNode, CapturedViewModel } from "../capture";
import { captureFrameData } from "../frame-capture";
function ownerNode(backendNodeId: number, x: number): CapturedNode {
return {
backendNodeId,
parentBackendNodeId: 1,
frameId: "main",
ownerFrameBackendNodeId: null,
tag: "iframe",
attrs: {},
rect: { x, y: 100, w: 300, h: 200 },
localRect: { x, y: 100, w: 300, h: 200 },
paintOrder: 1,
position: "static",
pointerEvents: "auto",
};
}
function childSnapshot(frameId: string, backendNodeId: number) {
const strings = [frameId, "body", "button", "static", "auto", "pointer"];
return {
strings,
documents: [
{
frameId,
nodes: {
parentIndex: [-1, 0],
nodeName: [1, 2],
backendNodeId: [backendNodeId - 1, backendNodeId],
attributes: [[], []],
},
layout: {
nodeIndex: [0, 1],
styles: [
[3, 4, 4],
[3, 4, 5],
],
bounds: [
[0, 0, 300, 200],
[10, 20, 100, 40],
],
paintOrders: [0, 1],
},
},
],
};
}
describe("captureFrameData", () => {
it("captures and positions multiple OOPIF documents missing from the root snapshot", async () => {
const leftOwner = ownerNode(10, 50);
const rightOwner = ownerNode(20, 500);
const captured: CapturedViewModel = {
nodes: [leftOwner, rightOwner],
viewport: { width: 1000, height: 800 },
iframeNodes: new Map(),
frameNodes: new Map([["main", [leftOwner, rightOwner]]]),
frameOwnerBackendNodeIds: new Map(),
rootFrameId: "main",
excludedBackendNodeIds: new Set(),
};
const sendToTarget = vi.fn(async (target, method) => {
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 300, clientHeight: 200, pageX: 0, pageY: 0 } };
}
if (method === "DOMSnapshot.enable" || method === "Accessibility.enable") return {};
if (method === "DOMSnapshot.captureSnapshot") {
return target.sessionId === "left-session"
? childSnapshot("left", 101)
: childSnapshot("right", 201);
}
if (method === "Accessibility.getFullAXTree") return { nodes: [] };
throw new Error(`unexpected ${method}`);
});
const cdp: CdpRunner = {
send: vi.fn(async (_tabId, method, params) => {
if (method === "Accessibility.enable") return {};
if (method === "Accessibility.getFullAXTree") return { nodes: [] };
if (method === "DOM.getBoxModel") {
const backendNodeId = (params as { backendNodeId?: number })?.backendNodeId;
const x = backendNodeId === 10 ? 50 : 500;
return { model: { content: [x, 100, x + 300, 100, x + 300, 300, x, 300] } };
}
if (method === "Page.getLayoutMetrics") {
return { cssLayoutViewport: { clientWidth: 1000, clientHeight: 800 } };
}
throw new Error(`unexpected root ${method}`);
}) as CdpRunner["send"],
sendToTarget: sendToTarget as unknown as NonNullable<CdpRunner["sendToTarget"]>,
getFrameGraph: vi.fn(async () => ({
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "left",
parentFrameId: "main",
ownerBackendNodeId: 10,
target: { tabId: 4, sessionId: "left-session" },
},
{
frameId: "right",
parentFrameId: "main",
ownerBackendNodeId: 20,
target: { tabId: 4, sessionId: "right-session" },
},
],
})),
};
const trees = await captureFrameData(cdp, 4, captured);
expect(trees.map((tree) => tree.frameId)).toEqual(["main", "left", "right"]);
expect(
captured.frameNodes?.get("left")?.find((node) => node.backendNodeId === 101)?.rect,
).toEqual({ x: 60, y: 120, w: 100, h: 40 });
expect(
captured.frameNodes?.get("right")?.find((node) => node.backendNodeId === 201)?.rect,
).toEqual({ x: 510, y: 120, w: 100, h: 40 });
expect(captured.frameOwnerBackendNodeIds).toEqual(
new Map([
["left", 10],
["right", 20],
]),
);
expect(captured.frameParentIds).toEqual(
new Map([
["left", "main"],
["right", "main"],
]),
);
});
});
@@ -1,174 +0,0 @@
import { describe, expect, it } from "vitest";
import type { CdpFrameGraph } from "@/browser-driver/frame-graph";
import type { CapturedNode, CapturedViewModel } from "../capture";
import { buildFrameDocuments, type FrameAxBatch, type FrameOwnedAxNode } from "../frame-document";
function domNode(frameId: string, backendNodeId: number): CapturedNode {
return {
backendNodeId,
parentBackendNodeId: null,
frameId,
tag: "div",
attrs: {},
rect: { x: 0, y: 0, w: 20, h: 20 },
paintOrder: 0,
position: "static",
pointerEvents: "auto",
};
}
function captured(frameNodes: Map<string, CapturedNode[]>): CapturedViewModel {
return {
nodes: frameNodes.get("main") ?? [],
viewport: { width: 800, height: 600 },
iframeNodes: new Map(),
frameNodes,
rootFrameId: "main",
excludedBackendNodeIds: new Set(),
};
}
describe("buildFrameDocuments", () => {
it("partitions mixed AX results by node ownership and cuts cross-frame parent edges", () => {
const graph: CdpFrameGraph = {
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{ frameId: "child", parentFrameId: "main", target: { tabId: 4 } },
],
};
const batches: FrameAxBatch<FrameOwnedAxNode>[] = [
{
frame: graph.frames[0],
nodes: [
{ nodeId: "root", frameId: "main", backendDOMNodeId: 1, childIds: ["h"] },
{
nodeId: "h",
frameId: "child",
parentId: "root",
backendDOMNodeId: 100,
childIds: ["virtual"],
},
{ nodeId: "virtual", parentId: "h" },
],
},
];
const documents = buildFrameDocuments(
graph,
batches,
captured(
new Map([
["main", [domNode("main", 1)]],
["child", [domNode("child", 100)]],
]),
),
);
expect(documents.find((document) => document.frameId === "main")?.axNodes).toHaveLength(1);
const childNodes = documents.find((document) => document.frameId === "child")?.axNodes ?? [];
expect(childNodes.map((node) => node.nodeId)).toEqual(["h", "virtual"]);
expect(childNodes.find((node) => node.nodeId === "h")?.parentId).toBeUndefined();
expect(childNodes.find((node) => node.nodeId === "virtual")?.frameId).toBe("child");
});
it("deduplicates overlapping same-target AX results using the strongest ownership", () => {
const graph: CdpFrameGraph = {
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{ frameId: "child", parentFrameId: "main", target: { tabId: 4 } },
],
};
const childNode = { nodeId: "shared", backendDOMNodeId: 100 };
const batches: FrameAxBatch<FrameOwnedAxNode>[] = [
{
frame: graph.frames[0],
nodes: [{ ...childNode, frameId: "child" }],
},
{
frame: graph.frames[1],
nodes: [childNode],
},
];
const documents = buildFrameDocuments(
graph,
batches,
captured(
new Map([
["main", [domNode("main", 1)]],
["child", [domNode("child", 100)]],
]),
),
);
expect(documents.find((document) => document.frameId === "main")?.axNodes).toEqual([]);
expect(documents.find((document) => document.frameId === "child")?.axNodes).toHaveLength(1);
});
it("uses target-scoped backend ownership when OOPIF backend ids collide", () => {
const graph: CdpFrameGraph = {
rootFrameId: "main",
frames: [
{ frameId: "main", target: { tabId: 4 } },
{
frameId: "left",
parentFrameId: "main",
target: { tabId: 4, sessionId: "left-session" },
},
{
frameId: "right",
parentFrameId: "main",
target: { tabId: 4, sessionId: "right-session" },
},
],
};
const batches: FrameAxBatch<FrameOwnedAxNode>[] = graph.frames.slice(1).map((frame) => ({
frame,
nodes: [{ nodeId: `${frame.frameId}-node`, backendDOMNodeId: 7 }],
}));
const documents = buildFrameDocuments(
graph,
batches,
captured(
new Map([
["main", []],
["left", [domNode("left", 7)]],
["right", [domNode("right", 7)]],
]),
),
);
expect(documents.find((document) => document.frameId === "left")?.axNodes[0]?.nodeId).toBe(
"left-node",
);
expect(documents.find((document) => document.frameId === "right")?.axNodes[0]?.nodeId).toBe(
"right-node",
);
});
it("uses captured frame parent identity instead of guessing from backend ids", () => {
const frameNodes = new Map([
["main", [domNode("main", 1)]],
["left", [domNode("left", 7)]],
["right", [domNode("right", 7)]],
["child", [domNode("child", 9)]],
]);
const capture = captured(frameNodes);
capture.frameOwnerBackendNodeIds = new Map([["child", 7]]);
capture.frameParentIds = new Map([["child", "left"]]);
const frames = ["main", "left", "right", "child"].map((frameId) => ({
frame: { frameId, target: { tabId: 4 } },
nodes: [],
}));
const documents = buildFrameDocuments(null, frames, capture);
expect(documents.find((document) => document.frameId === "child")).toMatchObject({
parentFrameId: "left",
ownerBackendNodeId: 7,
});
});
});
+76 -207
View File
@@ -7,7 +7,6 @@
import type { Rect, Viewport } from "@browser-skill/vom";
import { evaluateHoverTrigger } from "@/lib/hover-trigger-policy";
import { isOverlayHostNode, OVERLAY_HOST_SELECTOR } from "../../lib/overlay-bridge";
import { childFrameProjection, type GeometryProjection, projectRectToViewport } from "../geometry";
import type { CdpRunner } from "../shared";
const REQUESTED_STYLES = ["position", "pointer-events", "cursor"] as const;
@@ -18,15 +17,16 @@ const STYLE_COL = Object.fromEntries(
export interface CapturedNode {
backendNodeId: number;
parentBackendNodeId: number | null;
frameId?: string;
/** Owning iframe backend node id; `null` for the top-level document. */
ownerFrameBackendNodeId?: number | null;
tag: string;
attrs: Record<string, string>;
/** Top-level viewport-relative CSS px, clipped to the owning frame viewport. */
rect: Rect | null;
/** Frame-local viewport-relative CSS px before top-level projection. */
/** Frame-local viewport-relative CSS px before top-level translation/clipping. */
localRect?: Rect | null;
/** Frame-local document-relative CSS px as reported by DOMSnapshot. */
documentRect?: Rect | null;
paintOrder: number;
position: string;
pointerEvents: string;
@@ -59,11 +59,6 @@ export interface CapturedViewModel {
nodes: CapturedNode[];
viewport: Viewport;
iframeNodes: CapturedIframeNodes;
frameNodes?: Map<string, CapturedNode[]>;
frameOwnerBackendNodeIds?: Map<string, number>;
/** Explicit DOMSnapshot frame ancestry; never inferred from backend node ids. */
frameParentIds?: Map<string, string>;
rootFrameId?: string;
surfaceProbes?: CapturedSurfaceProbe[];
/** Backend node ids belonging to the agent overlay host + its shadow subtree. */
excludedBackendNodeIds: Set<number>;
@@ -72,25 +67,6 @@ export interface CapturedViewModel {
export interface CaptureViewModelOptions {
conditionalSurfaceProbe?: boolean;
hoverProbeBypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>;
signal?: AbortSignal;
}
function captureAbortError(): Error {
const error = new Error("observation aborted");
error.name = "AbortError";
return error;
}
function isAbortError(error: unknown): boolean {
return (
typeof error === "object" &&
error !== null &&
(error as { name?: string }).name === "AbortError"
);
}
function throwIfAborted(signal: AbortSignal | undefined): void {
if (signal?.aborted) throw captureAbortError();
}
/** Sparse array format Chrome uses for infrequently-set per-node fields. */
@@ -99,12 +75,7 @@ interface SparseArray {
value: number[];
}
interface RareBooleanData {
index: number[];
}
interface SnapshotDocument {
frameId?: string | number;
scrollOffsetX?: number;
scrollOffsetY?: number;
nodes?: {
@@ -119,10 +90,6 @@ interface SnapshotDocument {
nodeValue?: number[];
/** Maps node array index → index into `documents[]` for frame content. */
contentDocumentIndex?: SparseArray;
inputValue?: SparseArray;
textValue?: SparseArray;
inputChecked?: RareBooleanData;
optionSelected?: RareBooleanData;
};
layout?: {
nodeIndex?: number[];
@@ -132,12 +99,6 @@ interface SnapshotDocument {
};
}
function snapshotFrameId(document: SnapshotDocument, strings: string[]): string | undefined {
if (typeof document.frameId === "string") return document.frameId || undefined;
if (typeof document.frameId === "number") return str(strings, document.frameId) || undefined;
return undefined;
}
interface SnapshotReply {
strings?: string[];
documents?: SnapshotDocument[];
@@ -157,25 +118,6 @@ function isFormControlTag(tag: string): boolean {
return tag === "input" || tag === "textarea" || tag === "select";
}
function isSensitiveFormControl(node: Pick<CapturedNode, "tag" | "attrs">): boolean {
return node.tag === "input" && (node.attrs.type ?? "").toLowerCase() === "password";
}
function snapshotFormState(
value: string | undefined,
defaultValue: string,
hasDefaultValue: boolean,
sensitive: boolean,
): CapturedNode["formState"] {
if (sensitive) {
if (value === undefined && !hasDefaultValue) return undefined;
return (value ?? defaultValue) === "" ? "empty" : "filled";
}
if (value === undefined) return undefined;
if (value === "") return "empty";
return value === defaultValue ? "default" : "filled";
}
const MAX_FORM_ENRICH_CONTROLS = 250;
interface CapturedFormState {
@@ -207,8 +149,9 @@ function formStateBatchExpression(maxControls: number): string {
return {
state,
sensitive,
defaultValue,
placeholder,
...(sensitive ? {} : { value: rawValue, defaultValue }),
...(sensitive ? {} : { value: rawValue }),
};
};
const collect = (doc) => {
@@ -250,14 +193,10 @@ function applyFormStates(nodes: CapturedNode[], states: CapturedFormState[]): vo
index += 1;
if (!state) continue;
node.formState = state.state;
node.formDefaultValue = state.defaultValue ?? "";
node.formPlaceholder = state.placeholder ?? "";
if (state.sensitive || isSensitiveFormControl(node)) {
delete node.formValue;
delete node.formDefaultValue;
delete node.attrs.value;
} else {
node.formDefaultValue = state.defaultValue ?? "";
if (state.value !== undefined) node.formValue = state.value;
if (!state.sensitive && state.value !== undefined) {
node.formValue = state.value;
}
}
}
@@ -266,25 +205,21 @@ async function enrichFormControlStates(
cdp: CdpRunner,
tabId: number,
frameNodeGroups: CapturedNode[][],
signal?: AbortSignal,
): Promise<void> {
const hasControls = frameNodeGroups.some((nodes) =>
nodes.some((node) => isFormControlTag(node.tag)),
);
if (!hasControls) return;
try {
throwIfAborted(signal);
const result = await cdp.send<RuntimeEvaluateReply>(tabId, "Runtime.evaluate", {
expression: formStateBatchExpression(MAX_FORM_ENRICH_CONTROLS),
returnByValue: true,
});
throwIfAborted(signal);
const frameStates = flattenFrameFormStates(runtimeValue<CapturedFrameFormState>(result));
for (let i = 0; i < frameNodeGroups.length; i += 1) {
applyFormStates(frameNodeGroups[i], frameStates[i] ?? []);
}
} catch (error) {
if (isAbortError(error)) throw error;
} catch {
// Best-effort enrichment. DOMSnapshot/AX data still carries the nodes.
}
}
@@ -316,9 +251,9 @@ interface ParseDocumentResult {
}
interface FrameContext {
frameId?: string;
ownerFrameBackendNodeId: number | null;
projection: GeometryProjection | null;
originInTop: { x: number; y: number };
clipRectInTop: Rect;
scrollX: number;
scrollY: number;
}
@@ -337,6 +272,38 @@ function devicePixelRatio(metrics: LayoutMetricsReply): number {
return dpr >= 1 ? dpr : 1;
}
function viewportRect(viewport: Viewport): Rect {
return { x: 0, y: 0, w: Math.max(0, viewport.width), h: Math.max(0, viewport.height) };
}
function intersectRects(a: Rect, b: Rect): Rect | null {
const x1 = Math.max(a.x, b.x);
const y1 = Math.max(a.y, b.y);
const x2 = Math.min(a.x + a.w, b.x + b.w);
const y2 = Math.min(a.y + a.h, b.y + b.h);
if (x2 <= x1 || y2 <= y1) return null;
return { x: x1, y: y1, w: x2 - x1, h: y2 - y1 };
}
function rectInTop(localRect: Rect, context: FrameContext): Rect | null {
const translated = {
x: context.originInTop.x + localRect.x,
y: context.originInTop.y + localRect.y,
w: localRect.w,
h: localRect.h,
};
return intersectRects(translated, context.clipRectInTop);
}
function unclipRectInTop(localRect: Rect, context: FrameContext): Rect {
return {
x: context.originInTop.x + localRect.x,
y: context.originInTop.y + localRect.y,
w: localRect.w,
h: localRect.h,
};
}
function sparseIndexMap(sparse: SparseArray | undefined): Map<number, number> {
const out = new Map<number, number>();
if (!sparse?.index || !sparse.value) return out;
@@ -460,20 +427,8 @@ function hoverStateExpression(): string {
})()`;
}
async function wait(ms: number, signal?: AbortSignal): Promise<void> {
throwIfAborted(signal);
await new Promise<void>((resolve, reject) => {
const timer = setTimeout(() => {
signal?.removeEventListener("abort", onAbort);
resolve();
}, ms);
const onAbort = () => {
clearTimeout(timer);
signal?.removeEventListener("abort", onAbort);
reject(captureAbortError());
};
signal?.addEventListener("abort", onAbort, { once: true });
});
async function wait(ms: number): Promise<void> {
await new Promise((resolve) => setTimeout(resolve, ms));
}
async function clearHover(cdp: CdpRunner, tabId: number): Promise<void> {
@@ -645,51 +600,41 @@ async function probeHoverSurfaces(
): Promise<CapturedSurfaceProbe[]> {
const started = Date.now();
try {
throwIfAborted(options.signal);
const cssScan = await cdp.send<RuntimeEvaluateReply>(tabId, "Runtime.evaluate", {
expression: hoverCssTriggerScanExpression(),
returnByValue: true,
});
throwIfAborted(options.signal);
const cssHoverPoints = runtimeValue<Array<{ x: number; y: number }>>(cssScan) ?? [];
const candidates = buildHoverCandidates(nodes, cssHoverPoints);
if (candidates.length === 0) return [];
const results: CapturedSurfaceProbe[] = [];
const seen = new Set<number>();
throwIfAborted(options.signal);
await options.hoverProbeBypassOverlay?.(tabId, true).catch(() => undefined);
try {
throwIfAborted(options.signal);
for (const candidate of candidates.slice(0, MAX_HOVER_TRIGGERS)) {
throwIfAborted(options.signal);
if (Date.now() - started > MAX_HOVER_PROBE_MS) break;
if (results.length >= MAX_HOVER_SURFACES) break;
if (seen.has(candidate.backendNodeId)) continue;
try {
await clearHover(cdp, tabId);
throwIfAborted(options.signal);
await wait(HOVER_SETTLE_MS, options.signal);
await wait(HOVER_SETTLE_MS);
const baselineReply = await cdp.send<RuntimeEvaluateReply>(tabId, "Runtime.evaluate", {
expression: hoverStateExpression(),
returnByValue: true,
});
throwIfAborted(options.signal);
const baselineItems = runtimeValue<HoverRuntimeItem[]>(baselineReply) ?? [];
throwIfAborted(options.signal);
await cdp.send(tabId, "Input.dispatchMouseEvent", {
type: "mouseMoved",
x: candidate.x,
y: candidate.y,
});
throwIfAborted(options.signal);
await wait(HOVER_SETTLE_MS, options.signal);
await wait(HOVER_SETTLE_MS);
const collected = await cdp.send<RuntimeEvaluateReply>(tabId, "Runtime.evaluate", {
expression: hoverStateExpression(),
returnByValue: true,
});
throwIfAborted(options.signal);
const subItems = diffHoverItems(
baselineItems,
runtimeValue<HoverRuntimeItem[]>(collected) ?? [],
@@ -703,8 +648,7 @@ async function probeHoverSurfaces(
subItems,
confidence: confidenceForHover(candidate, subItems),
});
} catch (error) {
if (isAbortError(error)) throw error;
} catch {
continue;
} finally {
await clearHover(cdp, tabId);
@@ -715,7 +659,6 @@ async function probeHoverSurfaces(
}
return results;
} catch (err) {
if (isAbortError(err)) throw err;
console.debug("[bsk capture] hover surface probe failed", err);
return [];
}
@@ -728,16 +671,13 @@ async function probeHoverSurfaces(
export async function collectOverlayExcludedBackendIds(
cdp: CdpRunner,
tabId: number,
signal?: AbortSignal,
): Promise<Set<number>> {
const excluded = new Set<number>();
try {
throwIfAborted(signal);
const doc = await cdp.send<{ root?: { nodeId?: number } }>(tabId, "DOM.getDocument", {
depth: 0,
pierce: true,
});
throwIfAborted(signal);
const rootNodeId = doc.root?.nodeId;
if (typeof rootNodeId !== "number") return excluded;
@@ -745,7 +685,6 @@ export async function collectOverlayExcludedBackendIds(
nodeId: rootNodeId,
selector: OVERLAY_HOST_SELECTOR,
});
throwIfAborted(signal);
if (typeof found.nodeId !== "number" || found.nodeId === 0) return excluded;
const described = await cdp.send<{ node?: CdpDomNode }>(tabId, "DOM.describeNode", {
@@ -753,10 +692,8 @@ export async function collectOverlayExcludedBackendIds(
depth: -1,
pierce: true,
});
throwIfAborted(signal);
collectBackendIdsFromDomNode(described.node, excluded);
} catch (err) {
if (isAbortError(err)) throw err;
console.debug("[bsk capture] overlay exclusion fallback failed", err);
}
return excluded;
@@ -817,10 +754,6 @@ function parseDocumentNodes(
const posCol = STYLE_COL.position;
const peCol = STYLE_COL["pointer-events"];
const cursorCol = STYLE_COL.cursor;
const inputValues = sparseIndexMap(dn.inputValue);
const textValues = sparseIndexMap(dn.textValue);
const checkedInputs = new Set(dn.inputChecked?.index ?? []);
const selectedOptions = new Set(dn.optionSelected?.index ?? []);
// Collect visible text from #text child nodes so element CapturedNodes
// carry a textContent value usable as a button/link label fallback.
@@ -857,6 +790,7 @@ function parseDocumentNodes(
let rect: Rect | null = null;
let localRect: Rect | null = null;
let documentRect: Rect | null = null;
let paintOrder = 0;
let position = "static";
let pointerEvents = "auto";
@@ -865,16 +799,19 @@ function parseDocumentNodes(
if (li !== undefined) {
const b = dl?.bounds?.[li];
if (b && b.length >= 4 && b[2] > 0 && b[3] > 0) {
localRect = {
x: b[0] / dpr - context.scrollX,
y: b[1] / dpr - context.scrollY,
documentRect = {
x: b[0] / dpr,
y: b[1] / dpr,
w: b[2] / dpr,
h: b[3] / dpr,
};
const bounds = context.projection
? projectRectToViewport(localRect, context.projection)
: null;
rect = bounds ? { x: bounds.x, y: bounds.y, w: bounds.width, h: bounds.height } : null;
localRect = {
x: documentRect.x - context.scrollX,
y: documentRect.y - context.scrollY,
w: documentRect.w,
h: documentRect.h,
};
rect = rectInTop(localRect, context);
}
paintOrder = dl?.paintOrders?.[li] ?? 0;
const styleRow = dl?.styles?.[li] ?? [];
@@ -889,44 +826,20 @@ function parseDocumentNodes(
const textContent = nodeTextContent.get(n);
const rawFormValueIndex = tag === "textarea" ? textValues.get(n) : inputValues.get(n);
const rawFormValue =
rawFormValueIndex !== undefined ? str(strings, rawFormValueIndex) : undefined;
const sensitive = isSensitiveFormControl({ tag, attrs });
const formDefaultValue = attrs.value ?? "";
const formValue = sensitive ? undefined : rawFormValue;
const formState = snapshotFormState(
rawFormValue,
formDefaultValue,
Object.prototype.hasOwnProperty.call(attrs, "value"),
sensitive,
);
if (sensitive) delete attrs.value;
nodes.push({
backendNodeId,
parentBackendNodeId,
...(context.frameId ? { frameId: context.frameId } : {}),
ownerFrameBackendNodeId: context.ownerFrameBackendNodeId,
tag,
attrs,
rect,
localRect,
documentRect,
paintOrder,
position,
pointerEvents,
cursor,
textContent,
...(formValue !== undefined ? { formValue } : {}),
...(tag === "input" || tag === "textarea"
? {
formPlaceholder: attrs.placeholder ?? "",
...(!sensitive ? { formDefaultValue } : {}),
...(formState ? { formState } : {}),
}
: {}),
...(checkedInputs.has(n) ? { formValue: "true", formState: "filled" } : {}),
...(selectedOptions.has(n) ? { formValue: attrs.value ?? textContent ?? "" } : {}),
});
}
return { nodes, excludedBackendNodeIds };
@@ -934,8 +847,6 @@ function parseDocumentNodes(
interface ParseFrameDocumentsResult {
iframeNodes: CapturedIframeNodes;
frameOwnerBackendNodeIds: Map<string, number>;
frameParentIds: Map<string, string>;
excludedBackendNodeIds: Set<number>;
}
@@ -949,14 +860,10 @@ function parseChildFrameDocuments(
visited = new Set<number>(),
): ParseFrameDocumentsResult {
const iframeNodes: CapturedIframeNodes = new Map();
const frameOwnerBackendNodeIds = new Map<string, number>();
const frameParentIds = new Map<string, string>();
const excludedBackendNodeIds = new Set<number>();
const parentDoc = documents[parentDocIndex];
const cdi = sparseIndexMap(parentDoc?.nodes?.contentDocumentIndex);
if (cdi.size === 0) {
return { iframeNodes, frameOwnerBackendNodeIds, frameParentIds, excludedBackendNodeIds };
}
if (cdi.size === 0) return { iframeNodes, excludedBackendNodeIds };
const parentBackendIds = parentDoc?.nodes?.backendNodeId ?? [];
const parentNodeByBackendId = new Map(parentNodes.map((node) => [node.backendNodeId, node]));
@@ -968,15 +875,19 @@ function parseChildFrameDocuments(
const iframeBackendId = parentBackendIds[nodeArrayIdx];
if (iframeBackendId === undefined) continue;
const iframeNode = parentNodeByBackendId.get(iframeBackendId);
const projection =
parentContext.projection && iframeNode?.localRect
? childFrameProjection(parentContext.projection, iframeNode.localRect)
: null;
if (!iframeNode?.localRect) continue;
const iframeRectInTop = unclipRectInTop(iframeNode.localRect, parentContext);
const childClip = intersectRects(iframeRectInTop, parentContext.clipRectInTop);
if (!childClip) {
iframeNodes.set(iframeBackendId, []);
continue;
}
const childContext: FrameContext = {
frameId: snapshotFrameId(childDoc, strings),
ownerFrameBackendNodeId: iframeBackendId,
projection,
originInTop: { x: iframeRectInTop.x, y: iframeRectInTop.y },
clipRectInTop: childClip,
scrollX: childDoc.scrollOffsetX ?? 0,
scrollY: childDoc.scrollOffsetY ?? 0,
};
@@ -984,10 +895,6 @@ function parseChildFrameDocuments(
nextVisited.add(childDocIndex);
const parsed = parseDocumentNodes(childDoc, strings, dpr, childContext);
iframeNodes.set(iframeBackendId, parsed.nodes);
if (childContext.frameId) {
frameOwnerBackendNodeIds.set(childContext.frameId, iframeBackendId);
if (parentContext.frameId) frameParentIds.set(childContext.frameId, parentContext.frameId);
}
for (const id of parsed.excludedBackendNodeIds) excludedBackendNodeIds.add(id);
const nested = parseChildFrameDocuments(
@@ -1002,16 +909,10 @@ function parseChildFrameDocuments(
for (const [nestedFrameId, nestedNodes] of nested.iframeNodes) {
iframeNodes.set(nestedFrameId, nestedNodes);
}
for (const [frameId, ownerBackendNodeId] of nested.frameOwnerBackendNodeIds) {
frameOwnerBackendNodeIds.set(frameId, ownerBackendNodeId);
}
for (const [frameId, parentFrameId] of nested.frameParentIds) {
frameParentIds.set(frameId, parentFrameId);
}
for (const id of nested.excludedBackendNodeIds) excludedBackendNodeIds.add(id);
}
return { iframeNodes, frameOwnerBackendNodeIds, frameParentIds, excludedBackendNodeIds };
return { iframeNodes, excludedBackendNodeIds };
}
export async function captureViewModel(
@@ -1019,9 +920,7 @@ export async function captureViewModel(
tabId: number,
options: CaptureViewModelOptions = {},
): Promise<CapturedViewModel> {
throwIfAborted(options.signal);
const metrics = await cdp.send<LayoutMetricsReply>(tabId, "Page.getLayoutMetrics", {});
throwIfAborted(options.signal);
const dpr = devicePixelRatio(metrics);
const vpSrc = metrics.cssLayoutViewport ?? metrics.layoutViewport ?? {};
const viewport: Viewport = {
@@ -1032,13 +931,11 @@ export async function captureViewModel(
const scrollY = vpSrc.pageY ?? 0;
await cdp.send(tabId, "DOMSnapshot.enable", {});
throwIfAborted(options.signal);
const snap = await cdp.send<SnapshotReply>(tabId, "DOMSnapshot.captureSnapshot", {
computedStyles: REQUESTED_STYLES,
includePaintOrder: true,
includeDOMRects: true,
});
throwIfAborted(options.signal);
const strings = snap.strings ?? [];
const documents = snap.documents ?? [];
@@ -1054,13 +951,9 @@ export async function captureViewModel(
}
const topContext: FrameContext = {
frameId: snapshotFrameId(doc0, strings),
ownerFrameBackendNodeId: null,
projection: {
sourceClips: [],
edges: [],
topViewport: viewport,
},
originInTop: { x: 0, y: 0 },
clipRectInTop: viewportRect(viewport),
scrollX,
scrollY,
};
@@ -1074,35 +967,11 @@ export async function captureViewModel(
excludedBackendNodeIds.add(id);
}
await enrichFormControlStates(cdp, tabId, [nodes, ...iframeNodes.values()], options.signal);
throwIfAborted(options.signal);
await enrichFormControlStates(cdp, tabId, [nodes, ...iframeNodes.values()]);
const surfaceProbes = options.conditionalSurfaceProbe
? await probeHoverSurfaces(cdp, tabId, nodes, options)
: [];
throwIfAborted(options.signal);
const frameNodes = new Map<string, CapturedNode[]>();
if (topContext.frameId) frameNodes.set(topContext.frameId, nodes);
for (const iframeFrameNodes of iframeNodes.values()) {
const frameId = iframeFrameNodes.find((node) => node.frameId)?.frameId;
if (frameId) frameNodes.set(frameId, iframeFrameNodes);
}
for (const [frameId, ownerBackendNodeId] of frameParsed.frameOwnerBackendNodeIds) {
if (!frameNodes.has(frameId)) {
frameNodes.set(frameId, iframeNodes.get(ownerBackendNodeId) ?? []);
}
}
return {
nodes,
viewport,
iframeNodes,
frameNodes,
frameOwnerBackendNodeIds: frameParsed.frameOwnerBackendNodeIds,
frameParentIds: frameParsed.frameParentIds,
...(topContext.frameId ? { rootFrameId: topContext.frameId } : {}),
surfaceProbes,
excludedBackendNodeIds,
};
return { nodes, viewport, iframeNodes, surfaceProbes, excludedBackendNodeIds };
}
@@ -1,179 +0,0 @@
import type { CdpFrame, CdpFrameGraph } from "@/browser-driver/frame-graph";
import { resolveFrameProjection } from "../frame-geometry";
import { type GeometryProjection, projectRectToViewport } from "../geometry";
import { type CdpRunner, cdpRunnerForTarget, sendToCdpTarget } from "../shared";
import { type CapturedNode, type CapturedViewModel, captureViewModel } from "./capture";
import {
buildFrameDocuments,
type FrameAxBatch,
type FrameDocument,
type FrameOwnedAxNode,
} from "./frame-document";
export type FrameAxNode = FrameOwnedAxNode;
export type CapturedFrameDocument<T extends FrameAxNode> = FrameDocument<T>;
function throwIfAborted(signal?: AbortSignal): void {
if (signal?.aborted) throw new DOMException("observation aborted", "AbortError");
}
function isAbortError(err: unknown): boolean {
return err instanceof Error && err.name === "AbortError";
}
async function discoverFrameGraph(cdp: CdpRunner, tabId: number): Promise<CdpFrameGraph | null> {
if (!cdp.getFrameGraph) return null;
try {
return await cdp.getFrameGraph(tabId);
} catch (err) {
console.debug("[bsk observation] frame graph capture failed", err);
return null;
}
}
function transformFrameNodes(
nodes: CapturedNode[],
projection: GeometryProjection,
): CapturedNode[] {
return nodes.map((node) => ({
...node,
rect: (() => {
const bounds = projectRectToViewport(node.rect, projection);
return bounds ? { x: bounds.x, y: bounds.y, w: bounds.width, h: bounds.height } : null;
})(),
}));
}
function frameOwnerNode(captured: CapturedViewModel, frame: CdpFrame): CapturedNode | undefined {
if (frame.ownerBackendNodeId === undefined) return undefined;
const parentNodes = frame.parentFrameId
? captured.frameNodes?.get(frame.parentFrameId)
: captured.nodes;
return parentNodes?.find((node) => node.backendNodeId === frame.ownerBackendNodeId);
}
async function captureMissingFrameDocuments(
cdp: CdpRunner,
tabId: number,
graph: CdpFrameGraph,
captured: CapturedViewModel,
signal?: AbortSignal,
): Promise<void> {
if (!captured.frameNodes) captured.frameNodes = new Map();
if (!captured.frameOwnerBackendNodeIds) captured.frameOwnerBackendNodeIds = new Map();
if (!captured.frameParentIds) captured.frameParentIds = new Map();
for (const frame of graph.frames) {
if (!frame.target.sessionId || captured.frameNodes.has(frame.frameId)) continue;
const owner = frameOwnerNode(captured, frame);
if (!owner?.rect || frame.ownerBackendNodeId === undefined) continue;
try {
const child = await captureViewModel(cdpRunnerForTarget(cdp, frame.target), tabId, {
signal,
});
const projection = await resolveFrameProjection(cdp, graph, frame.frameId);
if (!projection) continue;
const rootFrameId = child.rootFrameId ?? frame.frameId;
const childFrames = child.frameNodes ?? new Map<string, CapturedNode[]>();
if (!childFrames.has(rootFrameId)) childFrames.set(rootFrameId, child.nodes);
if (rootFrameId !== frame.frameId && !childFrames.has(frame.frameId)) {
childFrames.set(
frame.frameId,
child.nodes.map((node) => ({ ...node, frameId: frame.frameId })),
);
}
for (const [childFrameId, nodes] of childFrames) {
captured.frameNodes.set(childFrameId, transformFrameNodes(nodes, projection));
}
captured.iframeNodes.set(
frame.ownerBackendNodeId,
captured.frameNodes.get(frame.frameId) ?? [],
);
captured.frameOwnerBackendNodeIds.set(frame.frameId, frame.ownerBackendNodeId);
if (frame.parentFrameId) captured.frameParentIds.set(frame.frameId, frame.parentFrameId);
for (const [childFrameId, ownerBackendNodeId] of child.frameOwnerBackendNodeIds ?? []) {
captured.frameOwnerBackendNodeIds.set(childFrameId, ownerBackendNodeId);
}
for (const [childFrameId, parentFrameId] of child.frameParentIds ?? []) {
captured.frameParentIds.set(childFrameId, parentFrameId);
}
} catch (err) {
if (isAbortError(err)) throw err;
console.debug("[bsk observation] child frame DOM capture failed", {
frameId: frame.frameId,
err,
});
}
}
}
async function captureAxTrees<T extends FrameAxNode>(
cdp: CdpRunner,
tabId: number,
discoveredFrames: CdpFrame[],
captured: CapturedViewModel,
signal?: AbortSignal,
): Promise<FrameAxBatch<T>[]> {
let frames = [...discoveredFrames];
const capturedFrameIds = [...(captured.frameNodes?.keys() ?? [])];
if (frames.length === 0) {
const rootFrameId = captured.rootFrameId ?? capturedFrameIds[0] ?? "root";
frames = [{ frameId: rootFrameId, target: { tabId } }];
}
for (const frameId of capturedFrameIds) {
if (!frames.some((frame) => frame.frameId === frameId)) {
frames.push({ frameId, target: { tabId } });
}
}
let firstFailure: unknown;
const trees = await Promise.all(
frames.map(async (frame): Promise<FrameAxBatch<T>> => {
try {
throwIfAborted(signal);
await sendToCdpTarget(cdp, frame.target, "Accessibility.enable", {});
const result = await sendToCdpTarget<{ nodes?: T[] }>(
cdp,
frame.target,
"Accessibility.getFullAXTree",
frame.frameId === "root" ? {} : { frameId: frame.frameId },
);
throwIfAborted(signal);
return {
frame,
nodes: result.nodes ?? [],
};
} catch (err) {
if (isAbortError(err)) throw err;
firstFailure ??= err;
console.debug("[bsk observation] frame AX capture failed", {
frameId: frame.frameId,
err,
});
return { frame, nodes: [] };
}
}),
);
const capturedNodeCount = [...captured.iframeNodes.values()].reduce(
(count, nodes) => count + nodes.length,
captured.nodes.length,
);
if (firstFailure && capturedNodeCount === 0 && trees.every((tree) => tree.nodes.length === 0)) {
throw firstFailure;
}
return trees;
}
export async function captureFrameData<T extends FrameAxNode>(
cdp: CdpRunner,
tabId: number,
captured: CapturedViewModel,
signal?: AbortSignal,
): Promise<CapturedFrameDocument<T>[]> {
const graph = await discoverFrameGraph(cdp, tabId);
const frames = graph?.frames ?? [];
if (graph) await captureMissingFrameDocuments(cdp, tabId, graph, captured, signal);
throwIfAborted(signal);
const batches = await captureAxTrees<T>(cdp, tabId, frames, captured, signal);
return buildFrameDocuments(graph, batches, captured);
}
@@ -1,173 +0,0 @@
import {
type CdpFrame,
type CdpFrameGraph,
type CdpTarget,
cdpTargetKey,
} from "@/browser-driver/frame-graph";
import type { CapturedNode, CapturedViewModel } from "./capture";
export interface FrameOwnedAxNode {
nodeId: string;
frameId?: string;
parentId?: string;
childIds?: string[];
backendDOMNodeId?: number;
}
export interface FrameAxBatch<T extends FrameOwnedAxNode> {
frame: CdpFrame;
nodes: T[];
}
export interface FrameDocument<T extends FrameOwnedAxNode> extends CdpFrame {
contextScopeId: string;
axNodes: T[];
domNodes: CapturedNode[];
}
interface Ownership {
frameId: string;
strength: number;
}
interface OwnedCandidate<T> {
node: T;
target: CdpTarget;
ownership: Ownership;
}
function targetNodeKey(target: CdpTarget, nodeId: string): string {
return `${cdpTargetKey(target)}:${nodeId}`;
}
function targetBackendKey(target: CdpTarget, backendNodeId: number): string {
return `${cdpTargetKey(target)}:${backendNodeId}`;
}
function frameList<T extends FrameOwnedAxNode>(
graph: CdpFrameGraph | null,
batches: FrameAxBatch<T>[],
captured: CapturedViewModel,
): CdpFrame[] {
const frames = new Map<string, CdpFrame>();
for (const frame of graph?.frames ?? []) frames.set(frame.frameId, frame);
for (const batch of batches) {
if (!frames.has(batch.frame.frameId)) frames.set(batch.frame.frameId, batch.frame);
}
for (const frameId of captured.frameNodes?.keys() ?? []) {
if (!frames.has(frameId))
frames.set(frameId, { frameId, target: { tabId: batches[0]?.frame.target.tabId ?? 0 } });
}
return [...frames.values()].map((frame) => {
const ownerBackendNodeId =
frame.ownerBackendNodeId ?? captured.frameOwnerBackendNodeIds?.get(frame.frameId);
const parentFrameId = frame.parentFrameId ?? captured.frameParentIds?.get(frame.frameId);
return {
...frame,
...(ownerBackendNodeId !== undefined ? { ownerBackendNodeId } : {}),
...(parentFrameId ? { parentFrameId } : {}),
};
});
}
export function buildFrameDocuments<T extends FrameOwnedAxNode>(
graph: CdpFrameGraph | null,
batches: FrameAxBatch<T>[],
captured: CapturedViewModel,
): FrameDocument<T>[] {
const frames = frameList(graph, batches, captured);
const frameById = new Map(frames.map((frame) => [frame.frameId, frame]));
const rootFrameId = graph?.rootFrameId ?? captured.rootFrameId ?? frames[0]?.frameId;
const domNodesForFrame = (frameId: string): CapturedNode[] =>
captured.frameNodes?.get(frameId) ?? (frameId === rootFrameId ? captured.nodes : []);
const backendOwner = new Map<string, string>();
for (const frame of frames) {
for (const node of domNodesForFrame(frame.frameId)) {
backendOwner.set(targetBackendKey(frame.target, node.backendNodeId), frame.frameId);
}
}
const candidates = new Map<string, OwnedCandidate<T>>();
for (const batch of batches) {
const nodeById = new Map(batch.nodes.map((node) => [node.nodeId, node]));
const ownershipByNodeId = new Map<string, Ownership>();
const resolving = new Set<string>();
const resolveOwnership = (node: T): Ownership => {
const cached = ownershipByNodeId.get(node.nodeId);
if (cached) return cached;
if (node.frameId && frameById.has(node.frameId)) {
const ownership = { frameId: node.frameId, strength: 4 };
ownershipByNodeId.set(node.nodeId, ownership);
return ownership;
}
if (typeof node.backendDOMNodeId === "number") {
const frameId = backendOwner.get(
targetBackendKey(batch.frame.target, node.backendDOMNodeId),
);
if (frameId) {
const ownership = { frameId, strength: 3 };
ownershipByNodeId.set(node.nodeId, ownership);
return ownership;
}
}
if (node.parentId && !resolving.has(node.nodeId)) {
const parent = nodeById.get(node.parentId);
if (parent) {
resolving.add(node.nodeId);
const inherited = resolveOwnership(parent);
resolving.delete(node.nodeId);
const ownership = {
frameId: inherited.frameId,
strength: Math.min(2, inherited.strength),
};
ownershipByNodeId.set(node.nodeId, ownership);
return ownership;
}
}
const ownership = { frameId: batch.frame.frameId, strength: 1 };
ownershipByNodeId.set(node.nodeId, ownership);
return ownership;
};
for (const node of batch.nodes) {
const ownership = resolveOwnership(node);
const key = targetNodeKey(batch.frame.target, node.nodeId);
const existing = candidates.get(key);
if (!existing || ownership.strength > existing.ownership.strength) {
candidates.set(key, { node, target: batch.frame.target, ownership });
}
}
}
const ownershipByTargetNode = new Map(
[...candidates].map(([key, candidate]) => [key, candidate.ownership.frameId]),
);
const axNodesByFrame = new Map<string, T[]>();
for (const candidate of candidates.values()) {
const { node, target, ownership } = candidate;
const parentFrameId = node.parentId
? ownershipByTargetNode.get(targetNodeKey(target, node.parentId))
: undefined;
const childIds = node.childIds?.filter(
(childId) => ownershipByTargetNode.get(targetNodeKey(target, childId)) === ownership.frameId,
);
const ownedNode = {
...node,
frameId: ownership.frameId,
...(parentFrameId === ownership.frameId
? { parentId: node.parentId }
: { parentId: undefined }),
...(childIds ? { childIds } : {}),
};
const nodes = axNodesByFrame.get(ownership.frameId) ?? [];
nodes.push(ownedNode);
axNodesByFrame.set(ownership.frameId, nodes);
}
return frames.map((frame) => ({
...frame,
contextScopeId: frame.frameId,
axNodes: axNodesByFrame.get(frame.frameId) ?? [],
domNodes: domNodesForFrame(frame.frameId),
}));
}
@@ -1,77 +0,0 @@
import type { Rect, RenderedRef, VomResult } from "@browser-skill/vom";
import type { CdpTarget } from "@/browser-driver/frame-graph";
import type { CapturedSurfaceProbe } from "./capture";
import type { CapturedFrameDocument, FrameAxNode } from "./frame-capture";
/** Frame identity needed to resolve an `@eN` that lives in an iframe. */
export interface CaptureVomFrame {
frameId: string;
target: CdpTarget;
parentFrameId?: string;
ownerBackendNodeId?: number;
}
/** Allowlisted geometry used to match a recorded action to a rendered ref. */
export interface CaptureVomMatchNode {
frameId: string;
backendNodeId: number;
tag: string;
/** Top-level viewport-relative geometry. */
rect: Rect | null;
/** Frame-local viewport-relative geometry, when available. */
localRect?: Rect | null;
}
/**
* Raw AX/DOM trees are never returned. When `redactValues` is enabled by the
* caller, form values in `text` are masked as well.
*/
export interface CaptureVomObservationResult {
text: string;
refs: RenderedRef[];
truncated: boolean;
rootFrameId: string;
frames: CaptureVomFrame[];
matchNodes: CaptureVomMatchNode[];
surfaceProbes?: CapturedSurfaceProbe[];
}
function projectFrames(documents: CapturedFrameDocument<FrameAxNode>[]): CaptureVomFrame[] {
return documents.map((document) => ({
frameId: document.frameId,
target: document.target,
...(document.parentFrameId ? { parentFrameId: document.parentFrameId } : {}),
...(document.ownerBackendNodeId !== undefined
? { ownerBackendNodeId: document.ownerBackendNodeId }
: {}),
}));
}
function projectMatchNodes(documents: CapturedFrameDocument<FrameAxNode>[]): CaptureVomMatchNode[] {
return documents.flatMap((document) =>
document.domNodes.map((node) => ({
frameId: document.frameId,
backendNodeId: node.backendNodeId,
tag: node.tag,
rect: node.rect,
...(node.localRect !== undefined ? { localRect: node.localRect } : {}),
})),
);
}
export function projectRecordSafeObservation(input: {
rootFrameId: string;
frameDocuments: CapturedFrameDocument<FrameAxNode>[];
rendered: VomResult;
surfaceProbes?: CapturedSurfaceProbe[];
}): CaptureVomObservationResult {
return {
text: input.rendered.text,
refs: input.rendered.refs,
truncated: input.rendered.truncated,
rootFrameId: input.rootFrameId,
frames: projectFrames(input.frameDocuments),
matchNodes: projectMatchNodes(input.frameDocuments),
surfaceProbes: input.surfaceProbes,
};
}
-3
View File
@@ -47,9 +47,7 @@ export async function handleWindowResize(
manager: SessionManager,
params: WindowResizeParams,
api: WindowResizeApi = chromeWindowResizeApi,
signal?: AbortSignal,
): Promise<WindowResizeResult | RpcError> {
if (signal?.aborted) return { code: "cancelled", message: "window_resize aborted" };
const ctxOrErr = lookupSession(manager, params, "window_resize");
if (isRpcError(ctxOrErr)) return ctxOrErr;
const ctx = ctxOrErr;
@@ -66,7 +64,6 @@ export async function handleWindowResize(
}
try {
if (signal?.aborted) return { code: "cancelled", message: "window_resize aborted" };
await api.update(ctx.agentWindowId, {
width: sizeOrErr.width,
height: sizeOrErr.height,
@@ -65,6 +65,11 @@ function deferredFakeTransport(): { transport: Transport; emit: (frame: Protocol
}
describe("performHandshake", () => {
it("advertises the Trace v3 compatibility boundary", () => {
expect(PROTOCOL_VERSION).toBe("1.1");
expect(MIN_COMPATIBLE_PROTOCOL).toBe("1.1");
});
it("sends system.handshake with identity and both compat fields", async () => {
let sentFrame: ProtocolFrame | null = null;
const transport = fakeTransport((req) => {
@@ -74,9 +79,9 @@ describe("performHandshake", () => {
result: {
server: "browser-skill-daemon",
version: "0.1.0",
protocol_version: "1.0",
protocol_version: "1.1",
min_compatible_peer: "0.0.0",
min_compatible_protocol: "1.0",
min_compatible_protocol: "1.1",
},
};
});
@@ -102,7 +107,7 @@ describe("performHandshake", () => {
},
});
expect(outcome.result.min_compatible_peer).toBe("0.0.0");
expect(outcome.result.min_compatible_protocol).toBe("1.0");
expect(outcome.result.min_compatible_protocol).toBe("1.1");
});
it("accepts legacy daemon reply with only min_compatible_peer", async () => {
@@ -129,8 +134,8 @@ describe("performHandshake", () => {
const response = {
server: "browser-skill-daemon",
version: "0.1.0",
protocol_version: "1.0",
min_compatible_protocol: "1.0",
protocol_version: "1.1",
min_compatible_protocol: "1.1",
} satisfies HandshakeResult;
const transport = fakeTransport((req) => ({
id: (req as { id: string }).id,
@@ -145,7 +150,7 @@ describe("performHandshake", () => {
});
expect(outcome.result.min_compatible_peer).toBeUndefined();
expect(outcome.result.min_compatible_protocol).toBe("1.0");
expect(outcome.result.min_compatible_protocol).toBe("1.1");
});
it("rejects when the daemon responds with an error", async () => {
@@ -178,9 +183,9 @@ describe("performHandshake", () => {
result: {
server: "browser-skill-daemon",
version: "0.1.0",
protocol_version: "1.0",
protocol_version: "1.1",
min_compatible_peer: "0.0.0",
min_compatible_protocol: "1.0",
min_compatible_protocol: "1.1",
},
});
emit({
@@ -188,14 +193,14 @@ describe("performHandshake", () => {
result: {
server: "browser-skill-daemon",
version: "0.1.0",
protocol_version: "1.0",
protocol_version: "1.1",
min_compatible_peer: "0.0.0",
min_compatible_protocol: "1.0",
min_compatible_protocol: "1.1",
},
});
await expect(pending).resolves.toMatchObject({
result: { server: "browser-skill-daemon", protocol_version: "1.0" },
result: { server: "browser-skill-daemon", protocol_version: "1.1" },
});
});
+2 -2
View File
@@ -7,7 +7,7 @@ import type {
ResponseFrame,
} from "./types";
export const PROTOCOL_VERSION = "1.0";
export const PROTOCOL_VERSION = "1.1";
/**
* Extension semver, injected at build time from `package.json` via
* Vite's `define` (see `wxt.config.ts` and `vitest.config.ts`).
@@ -18,7 +18,7 @@ export const EXTENSION_VERSION: string =
* Lowest **protocol** version this extension accepts (e.g. `"1.0"`).
* Must stay in sync with daemon `MIN_COMPATIBLE_PROTOCOL`.
*/
export const MIN_COMPATIBLE_PROTOCOL = "1.0";
export const MIN_COMPATIBLE_PROTOCOL = "1.1";
/**
* **Deprecated** — legacy app-semver floor for wire compat with old
* daemons. New code sends `"0.0.0"`; compat decisions ignore this.
+150 -21
View File
@@ -3,6 +3,8 @@
// stay structural (interface, not class) so the same JSON parses on both
// sides without extra adapters.
import type { CaptureTargetDescriptor } from "@/lib/describe-target";
export type RpcId = string;
export type ErrorCode =
@@ -32,8 +34,7 @@ export type RpcErrorReason =
| "tab_not_active"
| "restricted_tab_url"
| "borrow_conflict"
| "screenshot_capture_failed"
| "cleanup_failed";
| "screenshot_capture_failed";
export interface RpcErrorData {
reason?: RpcErrorReason;
@@ -657,26 +658,72 @@ export interface EmulateResult {
}
// --------------------------------------------------------------------------
// Semantic record payloads — mirror bsk-protocol record.rs (Trace v2)
// Semantic record payloads — mirror bsk-protocol record.rs (Trace v3)
// --------------------------------------------------------------------------
export const TRACE_VERSION = 3;
export const VOM_FORMAT_VERSION = 1;
export interface TargetDescriptor {
ref?: string;
role?: string;
name?: string;
tag: string;
name_attr?: string;
placeholder?: string;
nearby_label?: string;
ctx?: string;
unmatched?: boolean;
}
export interface TraceEntry {
start_url: string;
}
export interface PageRef {
export interface RecorderInfo {
bsk: string;
vom: number;
}
export type FrameCaptureStatus = "complete" | "partial";
export type FrameCaptureFailureReason =
| "not_injectable"
| "arm_failed"
| "rearm_failed"
| "drain_failed"
| "flush_failed";
export interface FrameCaptureFailure {
reason: FrameCaptureFailureReason;
frame_id: number;
document_id?: string;
url?: string;
detail?: string;
}
export interface FrameCaptureInfo {
status: FrameCaptureStatus;
failures?: FrameCaptureFailure[];
}
export type StopReason = "user_finish" | "cli_stop";
export interface TraceState {
id: string;
url: string;
title?: string;
/** Wire-only: full page observation text. */
body?: string;
/** Disk-only: filename under the bundle `pages/` directory. */
page?: string;
truncated?: boolean;
}
export interface StepResult {
state: string;
}
export interface StepCommon {
id: number;
state: string;
result: StepResult;
}
export interface SelectedOption {
@@ -684,66 +731,140 @@ export interface SelectedOption {
label?: string;
}
export interface StepEffect {
navigated_to: string;
export type NavigationCause =
| "user_typed"
| "link"
| "form_submit"
| "reload"
| "history"
| "script"
| "browser";
export type FillCommit = "enter" | "suggestion" | "blur";
/** Geometry reported by the content script at capture time. */
export interface CaptureGeometry {
rect: { x: number; y: number; w: number; h: number };
scrollX: number;
scrollY: number;
position: string;
tag: string;
ownerFrameBackendNodeId?: number | null;
}
export interface StepCommon {
id: number;
page: string;
effect?: StepEffect;
}
/** Capture/buffer draft before v2 reduction. */
/** Capture/buffer draft before v3 reduction. */
export type DraftTraceStep =
| {
op: "click";
target: TargetDescriptor;
captureTarget?: CaptureTargetDescriptor;
geometry?: CaptureGeometry;
navigated_to?: string;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "hover";
target: TargetDescriptor;
captureTarget?: CaptureTargetDescriptor;
geometry?: CaptureGeometry;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "fill";
target: TargetDescriptor;
captureTarget?: CaptureTargetDescriptor;
geometry?: CaptureGeometry;
value: string;
commit?: FillCommit;
redacted?: boolean;
navigated_to?: string;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "press";
key: string;
target?: TargetDescriptor;
captureTarget?: CaptureTargetDescriptor;
geometry?: CaptureGeometry;
modifiers?: KeyModifier[];
navigated_to?: string;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "select";
target: TargetDescriptor;
captureTarget?: CaptureTargetDescriptor;
geometry?: CaptureGeometry;
values: string[];
labels?: string[];
navigated_to?: string;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "scroll";
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
}
| {
op: "navigate";
url: string;
cause?: NavigationCause;
page_url?: string;
title?: string;
preStateId?: string;
postStateId?: string;
observationBody?: string;
observationHash?: string;
truncated?: boolean;
/** Raw webNavigation transition metadata for cause mapping. */
transitionType?: string;
transitionQualifiers?: string[];
};
/** Exported record-only step (trace v2). */
/** Exported record-only step (trace v3). */
export type Step =
| ({ op: "navigate" } & StepCommon & { to: string })
| ({ op: "navigate" } & StepCommon & { to: string; cause: NavigationCause })
| ({ op: "click" } & StepCommon & { target: TargetDescriptor })
| ({ op: "hover" } & StepCommon & { target: TargetDescriptor })
| ({ op: "fill" } & StepCommon & {
target: TargetDescriptor;
value: string;
commit: FillCommit;
redacted?: boolean;
})
| ({ op: "select" } & StepCommon & {
@@ -754,14 +875,20 @@ export type Step =
key: string;
modifiers?: KeyModifier[];
target?: TargetDescriptor;
});
})
| ({ op: "scroll" } & StepCommon);
export interface Trace {
version: number;
recorded_at: string;
started_at?: string;
purpose?: string;
stopped_by: StopReason;
entry: TraceEntry;
pages: PageRef[];
recorder: RecorderInfo;
/** Present when one or more frames could not be armed, rearmed, or flushed. */
frame_capture?: FrameCaptureInfo;
states: TraceState[];
steps: Step[];
}
@@ -770,6 +897,8 @@ export interface RecordStartParams {
tab_id?: number;
url?: string;
purpose?: string;
max_page_tokens?: number;
redact_values?: boolean;
}
export interface RecordStartResult {
+4 -2
View File
@@ -25,8 +25,10 @@ export default defineConfig({
name: "BrowserSkill",
description:
"Let AI agents use your logged-in browser in a separate Agent Window—without interrupting your work. Powered by the bsk CLI.",
// Flat debugger sessions are required to address out-of-process iframes.
minimum_chrome_version: "125",
// Chrome 116 is the first version where WebSocket activity resets the
// service-worker idle timer, which the 20s heartbeat relies on to
// keep the daemon link alive.
minimum_chrome_version: "116",
permissions: [
"alarms",
"debugger",
-5
View File
@@ -39,11 +39,6 @@
"**",
"!**/node_modules/**",
"!**/dist/**",
"!packages/dsh-plugin-browserskill/lib/**",
"!packages/dsh-plugin-browserskill/src/skill-content.generated.ts",
"!packages/dsh-plugin-browserskill/src/client/bsk-ui.nomodule.css",
"!packages/dsh-plugin-browserskill/.bsk-ui.raw.css",
"!packages/dsh-plugin-browserskill/.bsk-ui.input.css",
"!**/target/**",
"!**/.wxt/**",
"!**/.output/**",
+14 -6
View File
@@ -265,19 +265,27 @@ retry — complete the task autonomously or stop gracefully.
### Recording — `bsk record`
Capture the user's own actions in the Agent Window to a `trace.json`, for later LLM-driven automation:
Capture the user's own actions in the Agent Window to a **trace v3 bundle**, for later LLM-driven automation:
```bash
bsk record start --browser <instance-id-or-label> [--url https://…] [--purpose "publish a wiki doc"] [--output trace.json]
bsk record start --browser <instance-id-or-label> \
[--url https://…] [--purpose "publish a wiki doc"] \
[--max-page-tokens 3000] [--redact-values] \
[--output trace]
# `--url` is optional; default https://example.com/ when omitted (must be http(s)).
# Blocks until the user clicks Finish in the recording panel, then writes ./trace.json and closes the window.
# Blocks until the user clicks Finish in the recording panel, then writes:
# trace/trace.json — action chain + state index (version 3)
# trace/pages/ — one `.vom.txt` observe snapshot per settled page state
bsk record stop [--output trace.json] # terminal fallback if the browser panel is unavailable
bsk record stop [--output trace] # terminal fallback if the browser panel is unavailable
```
- The trace is a **record-only action log** (a `pages[]` dictionary + `navigate`/`click`/`fill`/`select`/`press` steps with `target` descriptors). It records *what the user did*; deciding which inputs are variable is left to the executing agent.
- **Bundle layout:** `--output` is a directory (default `./trace`) containing `trace.json` and `pages/`. `trace.json` lists `states[]` (page observation ids) and `steps[]` (each step binds `state` = observe snapshot *before* the action and `result.state` = snapshot *after* settle). Page bodies live in `pages/sN.vom.txt` using the same VOM format as `bsk observe`.
- Each `states[]` entry is one **settled page observation** (captured after recording start, navigation landing, or action settle — not on a timer).
- `target.ref` values like `@e12` exist **only inside** the bundle for disambiguation; do **not** copy `@eN` refs into SKILL.md or agent runbooks — use visible names from the observation text instead.
- `--purpose` is optional context metadata; it does **not** change what gets captured.
- There is **no** `bsk replay` — to redo a flow, read the trace and reuse the existing `session` / `snapshot` / `@eN` / `click` / `fill` tools. Follow **Stop when the goal is met**.
- `--redact-values` masks all form values in page files as `[filled]` / `[empty]`.
- Recording captures interactions in the **top frame and injectable child frames** (same-origin and cross-origin where the extension can attach). Embedded frames that cannot be injected are omitted; the trace may include `frame_capture.status: "partial"` with failure details.
- Do **not** record on banking/SSO/password-manager pages; passwords are redacted but traces may still contain sensitive text.
## Error handling
+16 -42
View File
@@ -18,12 +18,6 @@ use crate::daemon::state::PROTOCOL_VERSION;
const EXTENSION_STORE_URL: &str =
"https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi";
/// Edge Add-ons listing for the browser-skill extension.
const EXTENSION_STORE_URL_EDGE: &str = "https://microsoftedge.microsoft.com/addons/detail/browserskill/emacgiaaaiojkkpkddmmdfhmokgmnikg";
/// Store listings highlighted in repair hints, in the order they appear.
const EXTENSION_STORE_URLS: [&str; 2] = [EXTENSION_STORE_URL, EXTENSION_STORE_URL_EDGE];
/// Status of a single doctor check. `Ok` / `Fail` are the legacy two
/// states; `NotApplicable` (review M2) is reported as "N/A" in human
/// output and as `"status": "na"` in `--json` output, so a check that
@@ -426,10 +420,7 @@ fn check_extension_connected(status: Option<&StatusResult>) -> CheckResult {
CheckResult::fail(
name,
"0 browsers connected",
format!(
"install the extension from {EXTENSION_STORE_URL} (Chrome) \
or {EXTENSION_STORE_URL_EDGE} (Edge) and load it in the browser"
),
format!("install the extension from {EXTENSION_STORE_URL} and load it in Chromium"),
)
}
}
@@ -437,14 +428,17 @@ fn check_extension_connected(status: Option<&StatusResult>) -> CheckResult {
/// Highlight known URLs in repair hints for terminal output. Plain
/// text is preserved in `--json` and in stored [`CheckResult::hint`].
fn style_hint(hint: &str) -> String {
let mut styled = hint.to_string();
for url in EXTENSION_STORE_URLS {
if !styled.contains(url) {
continue;
}
styled = styled.replace(url, &style(url).cyan().bold().underlined().to_string());
if !hint.contains(EXTENSION_STORE_URL) {
return hint.to_string();
}
styled
hint.replace(
EXTENSION_STORE_URL,
&style(EXTENSION_STORE_URL)
.cyan()
.bold()
.underlined()
.to_string(),
)
}
fn render_human(checks: &[CheckResult]) {
@@ -486,7 +480,7 @@ mod m2_tests {
fn fake_status(browsers: Vec<BrowserStatusEntry>, skew: Vec<VersionSkewEntry>) -> StatusResult {
StatusResult {
daemon_version: env!("CARGO_PKG_VERSION").into(),
protocol_version: "1.0".into(),
protocol_version: "1.1".into(),
pid: 1,
uptime_secs: 0,
ws_port: 0,
@@ -510,26 +504,6 @@ mod m2_tests {
hint.contains(EXTENSION_STORE_URL),
"hint should include Chrome Web Store URL: {hint}"
);
assert!(
hint.contains(EXTENSION_STORE_URL_EDGE),
"hint should include Edge Add-ons URL: {hint}"
);
}
#[test]
fn style_hint_preserves_every_store_url() {
let hint = check_extension_connected(Some(&fake_status(Vec::new(), Vec::new())))
.hint
.expect("extension disconnected should include a hint");
let styled = style_hint(&hint);
// Whether or not the terminal accepts colors, styling must never drop or
// mangle a URL the user has to click.
for url in EXTENSION_STORE_URLS {
assert!(
styled.contains(url),
"styled hint should still contain {url}: {styled}"
);
}
}
#[test]
@@ -571,7 +545,7 @@ mod m2_tests {
session_count: 0,
connected_at_ms: 1,
version_skew: false,
extension_protocol_version: "1.0".into(),
extension_protocol_version: "1.1".into(),
}],
Vec::new(),
);
@@ -592,7 +566,7 @@ mod m2_tests {
session_count: 0,
connected_at_ms: 1,
version_skew: true,
extension_protocol_version: "1.1".into(),
extension_protocol_version: "1.2".into(),
}],
vec![VersionSkewEntry {
instance_id: "alpha".into(),
@@ -600,8 +574,8 @@ mod m2_tests {
label: "Personal".into(),
server_version: env!("CARGO_PKG_VERSION").into(),
client_version: "0.0.9".into(),
server_protocol_version: "1.0".into(),
client_protocol_version: "1.1".into(),
server_protocol_version: "1.1".into(),
client_protocol_version: "1.2".into(),
}],
);
let check = check_browsers_protocol_compatible(Some(&status));
+604 -50
View File
@@ -1,15 +1,17 @@
//! `bsk record start|stop` — capture user actions in the Agent Window.
use std::fs;
use std::path::PathBuf;
use std::collections::HashSet;
use std::fs::{self, OpenOptions};
use std::io;
use std::path::{Path, PathBuf};
use std::time::Duration;
use anyhow::Context;
use bsk_protocol::Method;
use bsk_protocol::tools::{
RecordAwaitParams, RecordAwaitResult, RecordStartParams, RecordStartResult, RecordStopParams,
RecordStopResult, Trace,
FrameCaptureStatus, RecordAwaitParams, RecordAwaitResult, RecordStartParams, RecordStartResult,
RecordStopParams, RecordStopResult, TRACE_VERSION, Trace, TraceState,
};
use bsk_protocol::{ErrorCode, Method};
use clap::{Args, Subcommand};
use crate::cli::TOOL_IPC_TIMEOUT;
@@ -18,10 +20,34 @@ use crate::cli::ensure_daemon::ensure_daemon;
use crate::cli::error::{CliError, Format};
use crate::cli::record_state;
use crate::cli::session::{SessionStartOptions, start_session, stop_session};
use crate::daemon::ipc::STOP_IN_PROGRESS_REASON;
/// Max wait for the user to click 结束 in the browser (24 hours).
const RECORD_AWAIT_TIMEOUT_MS: u32 = 86_400_000;
/// Tear down the recording session, tolerating the teardown the other half of
/// the pair already started. `record start` blocks in `record_await` and stops
/// the session once the trace arrives; the documented `record stop` fallback
/// stops it too. Whichever loses that race must not report a failure — the
/// trace is already written by then.
fn stop_recording_session(sock: PathBuf, session_id: &str) -> Result<(), CliError> {
match stop_session(sock, session_id) {
Ok(_) => Ok(()),
Err(err) if session_teardown_raced(&err) => Ok(()),
Err(err) => Err(err),
}
}
fn session_teardown_raced(err: &CliError) -> bool {
if err.code() == Some(ErrorCode::NotFound) {
return true;
}
err.data()
.and_then(|data| data.get("reason"))
.and_then(serde_json::Value::as_str)
== Some(STOP_IN_PROGRESS_REASON)
}
/// `bsk record …` subcommand tree.
#[derive(Debug, Clone, Args)]
pub struct RecordCmd {
@@ -58,15 +84,25 @@ pub struct RecordStartArgs {
#[arg(long)]
pub purpose: Option<String>,
/// Output path for the trace JSON (default `./trace.json`).
#[arg(long, default_value = "trace.json")]
/// Max VOM tokens per page observation file (default 3000).
#[arg(long = "max-page-tokens")]
pub max_page_tokens: Option<u32>,
/// Redact all form values in page observations (`[filled]` only).
#[arg(long = "redact-values")]
pub redact_values: bool,
/// Output directory for the trace bundle (default `./trace`).
/// Writes `<dir>/trace.json` and `<dir>/pages/*.vom.txt`.
#[arg(long, default_value = "trace")]
pub output: PathBuf,
}
#[derive(Debug, Clone, Args)]
pub struct RecordStopArgs {
/// Output path for the trace JSON (default `./trace.json`).
#[arg(long, default_value = "trace.json")]
/// Output directory for the trace bundle (default `./trace`).
/// Writes `<dir>/trace.json` and `<dir>/pages/*.vom.txt`.
#[arg(long, default_value = "trace")]
pub output: PathBuf,
}
@@ -98,6 +134,8 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
tab_id: args.tab_id,
url: args.url,
purpose: args.purpose.clone(),
max_page_tokens: args.max_page_tokens,
redact_values: Some(args.redact_values),
};
let start_result = business_rpc::call::<RecordStartParams, RecordStartResult>(
info.sock_path.clone(),
@@ -142,13 +180,13 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
// Keep write/render inside a Result so `?` cannot skip session teardown.
let run_result: Result<(), CliError> = match await_result {
Ok(result) => (|| {
write_trace_file(&args.output, &result.trace)?;
render_finish(&result.trace, &args.output, format)
let pages_dir = write_trace_bundle(&args.output, &result.trace)?;
render_finish(&result.trace, &args.output, &pages_dir, format)
})(),
Err(err) => Err(err),
};
let session_stop_result = stop_session(info.sock_path, &session.session_id);
let session_stop_result = stop_recording_session(info.sock_path, &session.session_id);
record_state::clear();
run_result?;
@@ -161,22 +199,23 @@ fn dispatch_stop(args: RecordStopArgs, format: Format) -> Result<(), CliError> {
let info = ensure_daemon().context("ensure daemon is running")?;
let session_id = state.session_id.clone();
let params = RecordStopParams {
session_id: session_id.clone(),
};
let result = business_rpc::call::<RecordStopParams, RecordStopResult>(
info.sock_path.clone(),
"record-stop",
Method::ToolRecordStop,
Some(params),
TOOL_IPC_TIMEOUT,
)?;
let run_result: Result<(), CliError> = (|| {
let params = RecordStopParams {
session_id: session_id.clone(),
};
let result = business_rpc::call::<RecordStopParams, RecordStopResult>(
info.sock_path.clone(),
"record-stop",
Method::ToolRecordStop,
Some(params),
TOOL_IPC_TIMEOUT,
)?;
write_trace_file(&args.output, &result.trace)?;
render_stop(&result, &args.output, format)
let pages_dir = write_trace_bundle(&args.output, &result.trace)?;
render_stop(&result, &args.output, &pages_dir, format)
})();
let session_stop_result = stop_session(info.sock_path, &session_id);
let session_stop_result = stop_recording_session(info.sock_path, &session_id);
record_state::clear();
run_result?;
@@ -190,30 +229,313 @@ fn record_await_ipc_timeout(timeout_ms: u32) -> Duration {
.unwrap_or(Duration::from_secs(u64::from(timeout_ms / 1_000) + 15))
}
fn write_trace_file(output: &PathBuf, trace: &Trace) -> Result<(), CliError> {
let json = serde_json::to_string_pretty(trace)
.context("serialize trace JSON")
.map_err(CliError::Local)?;
if let Some(parent) = output.parent() {
if !parent.as_os_str().is_empty() {
fs::create_dir_all(parent)
.with_context(|| format!("create output directory {}", parent.display()))
.map_err(CliError::Local)?;
fn pages_dir_for_output(output_dir: &Path) -> PathBuf {
output_dir.join("pages")
}
fn trace_json_path(output_dir: &Path) -> PathBuf {
output_dir.join("trace.json")
}
fn is_canonical_state_id(id: &str) -> bool {
let Some(number) = id.strip_prefix('s') else {
return false;
};
let mut digits = number.bytes();
matches!(digits.next(), Some(b'1'..=b'9')) && digits.all(|byte| byte.is_ascii_digit())
}
fn validate_bundle_directory(path: &Path, description: &str) -> Result<(), CliError> {
match fs::symlink_metadata(path) {
Ok(metadata) if metadata.file_type().is_symlink() => Err(CliError::Local(anyhow::anyhow!(
"{description} {} must not be a symlink",
path.display()
))),
Ok(metadata) if metadata.is_dir() => Ok(()),
Ok(_) => Err(CliError::Local(anyhow::anyhow!(
"{description} {} is not a directory",
path.display()
))),
Err(err) if err.kind() == io::ErrorKind::NotFound => Ok(()),
Err(err) => Err(CliError::Local(
anyhow::Error::new(err).context(format!("inspect {description} {}", path.display())),
)),
}
}
fn validate_replaceable_file(path: &Path, description: &str) -> Result<(), CliError> {
match fs::symlink_metadata(path) {
Ok(metadata) if metadata.file_type().is_file() => Ok(()),
Ok(_) => Err(CliError::Local(anyhow::anyhow!(
"{description} {} is not a regular file",
path.display()
))),
Err(err) if err.kind() == io::ErrorKind::NotFound => Ok(()),
Err(err) => Err(CliError::Local(
anyhow::Error::new(err).context(format!("inspect {description} {}", path.display())),
)),
}
}
fn commit_staged_file(staged: &Path, target: &Path, backup: &Path) -> io::Result<Option<PathBuf>> {
let old_file = match fs::symlink_metadata(target) {
Ok(_) => {
fs::rename(target, backup)?;
Some(backup.to_path_buf())
}
Err(err) if err.kind() == io::ErrorKind::NotFound => None,
Err(err) => return Err(err),
};
if let Err(err) = fs::rename(staged, target) {
if let Some(old_file) = old_file.as_deref() {
if let Err(restore_err) = fs::rename(old_file, target) {
return Err(io::Error::new(
restore_err.kind(),
format!(
"{err}; additionally failed to restore {}: {restore_err}",
target.display()
),
));
}
}
return Err(err);
}
Ok(old_file)
}
fn rollback_installed_files(installed: &[(PathBuf, Option<PathBuf>)]) -> io::Result<()> {
let mut rollback_error = None;
for (target, backup) in installed.iter().rev() {
match fs::remove_file(target) {
Ok(()) => {}
Err(err) if err.kind() == io::ErrorKind::NotFound => {}
Err(err) => {
rollback_error.get_or_insert(err);
continue;
}
}
if let Some(backup) = backup {
if let Err(err) = fs::rename(backup, target) {
rollback_error.get_or_insert(err);
}
}
}
match rollback_error {
Some(err) => Err(err),
None => Ok(()),
}
}
/// Drop page files left by an earlier, longer recording without deleting the
/// directory itself: `record start` and the `record stop` fallback export the
/// same recording concurrently, and a `remove_dir_all` here would fail the
/// writer whose files the other one just erased.
fn prune_stale_pages(pages_dir: &Path, keep: &HashSet<String>) -> Result<(), CliError> {
let entries = match fs::read_dir(pages_dir) {
Ok(entries) => entries,
Err(err) if err.kind() == io::ErrorKind::NotFound => return Ok(()),
Err(err) => {
return Err(CliError::Local(
anyhow::Error::new(err).context(format!("read pages dir {}", pages_dir.display())),
));
}
};
for entry in entries.flatten() {
let name = entry.file_name();
let Some(name) = name.to_str() else { continue };
if !name.ends_with(".vom.txt") || keep.contains(name) {
continue;
}
match fs::remove_file(entry.path()) {
Ok(()) => {}
Err(err) if err.kind() == io::ErrorKind::NotFound => {}
Err(err) => {
return Err(CliError::Local(
anyhow::Error::new(err)
.context(format!("remove stale page {}", entry.path().display())),
));
}
}
}
fs::write(output, format!("{json}\n"))
.with_context(|| format!("write trace to {}", output.display()))
.map_err(CliError::Local)?;
Ok(())
}
fn render_finish(trace: &Trace, output: &PathBuf, format: Format) -> Result<(), CliError> {
fn write_trace_bundle(output_dir: &Path, trace: &Trace) -> Result<PathBuf, CliError> {
let pages_dir = pages_dir_for_output(output_dir);
let trace_path = trace_json_path(output_dir);
if trace.version != TRACE_VERSION {
return Err(CliError::Local(anyhow::anyhow!(
"trace version {} is not supported (expected {})",
trace.version,
TRACE_VERSION
)));
}
let mut state_ids = HashSet::with_capacity(trace.states.len());
for state in &trace.states {
if !is_canonical_state_id(&state.id) {
return Err(CliError::Local(anyhow::anyhow!(
"state id {:?} is not canonical (expected s1, s2, ...)",
state.id
)));
}
if !state_ids.insert(state.id.as_str()) {
return Err(CliError::Local(anyhow::anyhow!(
"duplicate state id {:?}",
state.id
)));
}
}
let mut disk_states: Vec<TraceState> = Vec::with_capacity(trace.states.len());
let mut page_writes = Vec::new();
let mut keep = HashSet::new();
for state in &trace.states {
let mut disk = state.clone();
if let Some(body) = state.body.as_deref() {
let filename = format!("{}.vom.txt", state.id);
let page_path = pages_dir.join(&filename);
if page_path.parent() != Some(pages_dir.as_path()) {
return Err(CliError::Local(anyhow::anyhow!(
"page path {} escapes pages directory {}",
page_path.display(),
pages_dir.display()
)));
}
keep.insert(filename.clone());
page_writes.push((filename.clone(), body));
disk.body = None;
disk.page = Some(filename);
}
disk_states.push(disk);
}
let mut disk_trace = trace.clone();
disk_trace.states = disk_states;
let json = serde_json::to_string_pretty(&disk_trace)
.context("serialize trace JSON")
.map_err(CliError::Local)?;
validate_bundle_directory(output_dir, "output bundle directory")?;
fs::create_dir_all(output_dir)
.with_context(|| format!("create output dir {}", output_dir.display()))
.map_err(CliError::Local)?;
validate_bundle_directory(&pages_dir, "pages directory")?;
// `record start` and `record stop` can export the same recording at once.
// Serialize writers without replacing the shared directory underneath one
// another; the lock is released when this file handle is dropped.
let lock_path = output_dir.join(".bsk-record-export.lock");
let lock_file = OpenOptions::new()
.read(true)
.write(true)
.create(true)
.truncate(false)
.open(&lock_path)
.with_context(|| format!("open export lock {}", lock_path.display()))
.map_err(CliError::Local)?;
fs2::FileExt::lock_exclusive(&lock_file)
.with_context(|| format!("lock export bundle {}", output_dir.display()))
.map_err(CliError::Local)?;
fs::create_dir_all(&pages_dir)
.with_context(|| format!("create pages dir {}", pages_dir.display()))
.map_err(CliError::Local)?;
for (filename, _) in &page_writes {
validate_replaceable_file(&pages_dir.join(filename), "page observation")?;
}
validate_replaceable_file(&trace_path, "trace JSON")?;
let transaction_id = uuid::Uuid::new_v4();
let staging_dir = output_dir.join(format!(".bsk-record-stage-{transaction_id}"));
let staging_pages = staging_dir.join("pages");
fs::create_dir_all(&staging_pages)
.with_context(|| format!("create staging dir {}", staging_pages.display()))
.map_err(CliError::Local)?;
let stage_result: Result<(), CliError> = (|| {
for (filename, body) in &page_writes {
let staged_path = staging_pages.join(filename);
fs::write(&staged_path, body)
.with_context(|| format!("write staged page observation {}", staged_path.display()))
.map_err(CliError::Local)?;
}
let staged_trace = staging_dir.join("trace.json");
fs::write(&staged_trace, format!("{json}\n"))
.with_context(|| format!("write staged trace to {}", staged_trace.display()))
.map_err(CliError::Local)?;
Ok(())
})();
if let Err(err) = stage_result {
let _ = fs::remove_dir_all(&staging_dir);
return Err(err);
}
let mut replacements: Vec<(PathBuf, PathBuf, PathBuf)> =
Vec::with_capacity(page_writes.len() + 1);
for (filename, _) in &page_writes {
replacements.push((
staging_pages.join(filename),
pages_dir.join(filename),
pages_dir.join(format!(".{filename}.{transaction_id}.bak")),
));
}
replacements.push((
staging_dir.join("trace.json"),
trace_path.clone(),
output_dir.join(format!(".trace.json.{transaction_id}.bak")),
));
let mut installed = Vec::with_capacity(replacements.len());
for (staged, target, backup) in replacements {
match commit_staged_file(&staged, &target, &backup) {
Ok(old_file) => installed.push((target, old_file)),
Err(err) => {
let rollback_result = rollback_installed_files(&installed);
let _ = fs::remove_dir_all(&staging_dir);
let message = match rollback_result {
Ok(()) => format!("commit replacement {}: {err}", target.display()),
Err(rollback_err) => format!(
"commit replacement {}: {err}; rollback also failed: {rollback_err}",
target.display()
),
};
return Err(CliError::Local(anyhow::anyhow!(message)));
}
}
}
for (_, backup) in &installed {
if let Some(backup) = backup {
let _ = fs::remove_file(backup);
}
}
let _ = fs::remove_dir_all(&staging_dir);
prune_stale_pages(&pages_dir, &keep)?;
Ok(pages_dir)
}
fn render_finish(
trace: &Trace,
output_dir: &PathBuf,
pages_dir: &PathBuf,
format: Format,
) -> Result<(), CliError> {
let trace_path = trace_json_path(output_dir);
match format {
Format::Json => {
println!(
"{}",
serde_json::to_string(&serde_json::json!({
"output": output,
"output": output_dir,
"trace_json": trace_path,
"pages_dir": pages_dir,
"trace": trace,
"window_closed": true,
}))
@@ -221,7 +543,20 @@ fn render_finish(trace: &Trace, output: &PathBuf, format: Format) -> Result<(),
);
}
Format::Human => {
println!("saved {} steps to {}", trace.steps.len(), output.display());
println!(
"saved {} steps and {} states to {} (+ {})",
trace.steps.len(),
trace.states.len(),
trace_path.display(),
pages_dir.display()
);
if let Some(frame_capture) = &trace.frame_capture {
if matches!(frame_capture.status, FrameCaptureStatus::Partial) {
eprintln!(
"warning: trace is partial — some embedded frames were not recorded (see frame_capture.failures in trace.json)"
);
}
}
}
}
Ok(())
@@ -230,33 +565,252 @@ fn render_finish(trace: &Trace, output: &PathBuf, format: Format) -> Result<(),
fn render_stop(
result: &RecordStopResult,
output: &PathBuf,
pages_dir: &PathBuf,
format: Format,
) -> Result<(), CliError> {
render_finish(&result.trace, output, format)
render_finish(&result.trace, output, pages_dir, format)
}
#[cfg(test)]
mod tests {
use super::*;
use bsk_protocol::tools::{
NavigationCause, RecorderInfo, Step, StepCommon, StepResult, StopReason, TraceEntry,
VOM_FORMAT_VERSION,
};
#[test]
fn default_output_is_trace_json() {
let args = RecordStopArgs {
output: PathBuf::from("trace.json"),
};
assert_eq!(args.output, PathBuf::from("trace.json"));
fn sample_trace(state_id: &str, body: &str) -> Trace {
Trace {
version: TRACE_VERSION,
recorded_at: "2026-08-10T02:12:55.360Z".into(),
stopped_by: StopReason::UserFinish,
entry: TraceEntry {
start_url: "https://example.com/".into(),
},
recorder: RecorderInfo {
bsk: "0.1.10".into(),
vom: VOM_FORMAT_VERSION,
},
frame_capture: None,
states: vec![TraceState {
id: state_id.into(),
url: "https://example.com/".into(),
title: Some("Example".into()),
body: Some(body.into()),
page: None,
truncated: false,
}],
steps: vec![Step::Navigate {
common: StepCommon {
id: 1,
state: state_id.into(),
result: StepResult {
state: state_id.into(),
},
},
to: "https://example.com/".into(),
cause: NavigationCause::UserTyped,
}],
purpose: None,
started_at: None,
}
}
#[test]
fn start_args_default_output_is_trace_json() {
fn concurrent_teardown_is_not_a_record_failure() {
let stopping = CliError::from_rpc(bsk_protocol::RpcError {
code: ErrorCode::Timeout,
message: "session stop is already in progress".into(),
data: Some(serde_json::json!({ "reason": STOP_IN_PROGRESS_REASON })),
});
assert!(session_teardown_raced(&stopping));
let gone = CliError::from_rpc(bsk_protocol::RpcError {
code: ErrorCode::NotFound,
message: "session is not registered".into(),
data: None,
});
assert!(session_teardown_raced(&gone));
}
#[test]
fn unrelated_stop_timeout_still_fails() {
let stalled = CliError::from_rpc(bsk_protocol::RpcError {
code: ErrorCode::Timeout,
message: "tool.session_stop timed out".into(),
data: None,
});
assert!(!session_teardown_raced(&stalled));
}
#[test]
fn default_output_is_trace_dir() {
let args = RecordStopArgs {
output: PathBuf::from("trace"),
};
assert_eq!(args.output, PathBuf::from("trace"));
}
#[test]
fn start_args_default_output_is_trace_dir() {
let args = RecordStartArgs {
browser: None,
tab_id: None,
url: None,
purpose: None,
output: PathBuf::from("trace.json"),
max_page_tokens: None,
redact_values: false,
output: PathBuf::from("trace"),
};
assert_eq!(args.output, PathBuf::from("trace.json"));
assert_eq!(args.output, PathBuf::from("trace"));
}
#[test]
fn write_trace_bundle_writes_dir_layout() {
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
let trace = sample_trace("s1", "@vom 1\nL1 page");
let pages_dir = write_trace_bundle(&output, &trace).unwrap();
let trace_path = output.join("trace.json");
assert!(trace_path.exists());
assert_eq!(pages_dir, output.join("pages"));
assert!(pages_dir.join("s1.vom.txt").exists());
let disk: serde_json::Value =
serde_json::from_str(&fs::read_to_string(&trace_path).unwrap()).unwrap();
assert_eq!(disk["states"][0]["page"], "s1.vom.txt");
assert!(disk["states"][0].get("body").is_none());
assert_eq!(
fs::read_to_string(pages_dir.join("s1.vom.txt")).unwrap(),
"@vom 1\nL1 page"
);
// Re-exporting drops pages from the previous run but keeps the
// directory in place for the concurrent writer.
fs::write(pages_dir.join("s9.vom.txt"), "stale").unwrap();
fs::write(pages_dir.join("notes.md"), "keep me").unwrap();
write_trace_bundle(&output, &trace).unwrap();
assert!(!pages_dir.join("s9.vom.txt").exists());
assert!(pages_dir.join("s1.vom.txt").exists());
assert!(pages_dir.join("notes.md").exists());
}
#[test]
fn write_trace_bundle_rejects_traversal_state_id() {
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
let trace = sample_trace("../escaped", "@vom 1\nL1 page");
let err = write_trace_bundle(&output, &trace).unwrap_err();
assert!(err.to_string().contains("state id"));
assert!(!output.join("escaped.vom.txt").exists());
assert!(!output.join("trace.json").exists());
}
#[test]
fn write_trace_bundle_rejects_duplicate_state_id() {
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
let mut trace = sample_trace("s1", "@vom 1\nL1 first");
trace.states.push(TraceState {
body: Some("@vom 1\nL1 duplicate".into()),
..trace.states[0].clone()
});
let err = write_trace_bundle(&output, &trace).unwrap_err();
assert!(err.to_string().contains("duplicate state id"));
assert!(!output.exists());
}
#[test]
fn unsupported_version_preserves_existing_bundle() {
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
let original_json = fs::read(output.join("trace.json")).unwrap();
let mut replacement = sample_trace("s2", "replacement");
replacement.version += 1;
let err = write_trace_bundle(&output, &replacement).unwrap_err();
assert!(err.to_string().contains("not supported"));
assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
assert_eq!(
fs::read_to_string(output.join("pages/s1.vom.txt")).unwrap(),
"original"
);
assert!(!output.join("pages/s2.vom.txt").exists());
}
#[test]
fn page_replacement_failure_preserves_existing_bundle() {
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
let original_json = fs::read(output.join("trace.json")).unwrap();
fs::write(output.join("pages/notes.md"), "keep me").unwrap();
fs::create_dir(output.join("pages/s2.vom.txt")).unwrap();
let err = write_trace_bundle(&output, &sample_trace("s2", "replacement")).unwrap_err();
assert!(err.to_string().contains("page"));
assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
assert_eq!(
fs::read_to_string(output.join("pages/s1.vom.txt")).unwrap(),
"original"
);
assert_eq!(
fs::read_to_string(output.join("pages/notes.md")).unwrap(),
"keep me"
);
}
#[cfg(unix)]
#[test]
fn write_trace_bundle_rejects_symlinked_pages_directory() {
use std::os::unix::fs::symlink;
let dir = tempfile::tempdir().unwrap();
let output = dir.path().join("flow");
let external_pages = dir.path().join("external-pages");
fs::create_dir_all(&output).unwrap();
fs::create_dir_all(&external_pages).unwrap();
fs::write(external_pages.join("s9.vom.txt"), "external").unwrap();
symlink(&external_pages, output.join("pages")).unwrap();
let err = write_trace_bundle(&output, &sample_trace("s1", "replacement")).unwrap_err();
assert!(err.to_string().contains("symlink"));
assert_eq!(
fs::read_to_string(external_pages.join("s9.vom.txt")).unwrap(),
"external"
);
assert!(!external_pages.join("s1.vom.txt").exists());
assert!(!output.join("trace.json").exists());
}
#[cfg(unix)]
#[test]
fn write_trace_bundle_rejects_symlinked_output_directory() {
use std::os::unix::fs::symlink;
let dir = tempfile::tempdir().unwrap();
let external_output = dir.path().join("external-output");
let output = dir.path().join("flow");
fs::create_dir_all(&external_output).unwrap();
fs::write(external_output.join("notes.md"), "external").unwrap();
symlink(&external_output, &output).unwrap();
let err = write_trace_bundle(&output, &sample_trace("s1", "replacement")).unwrap_err();
assert!(err.to_string().contains("symlink"));
assert_eq!(
fs::read_to_string(external_output.join("notes.md")).unwrap(),
"external"
);
assert!(!external_output.join("trace.json").exists());
assert!(!external_output.join("pages").exists());
}
#[test]
+27 -3
View File
@@ -141,9 +141,9 @@ pub fn info_for(code: ErrorCode) -> RenderInfo {
exit_code: 2,
},
ErrorCode::VersionTooOld => RenderInfo {
summary: "peer version is too old to communicate with this build",
summary: "the installed bsk CLI and BrowserSkill extension are incompatible",
hint: Some(
"upgrade both bsk CLI and the browser-skill extension so both sides satisfy `min_compatible_protocol`",
"update bsk and the BrowserSkill extension, reload the extension, then run `bsk daemon restart`",
),
exit_code: 5,
},
@@ -157,7 +157,7 @@ pub fn info_for(code: ErrorCode) -> RenderInfo {
ErrorCode::NoBrowserConnected => RenderInfo {
summary: "no browser is connected to the daemon",
hint: Some(
"open the browser-skill extension in your browser and wait for the popup to show \"connected\"",
"open the BrowserSkill extension and wait for it to show \"connected\"; if the extension is already open, update and reload it, then run `bsk daemon restart`",
),
exit_code: 1,
},
@@ -310,6 +310,30 @@ mod tests {
assert_eq!(exit_code_for(ErrorCode::VersionTooOld), 5);
}
#[test]
fn version_error_names_the_components_and_gives_update_steps() {
let info = info_for(ErrorCode::VersionTooOld);
assert!(info.summary.contains("bsk CLI"));
assert!(info.summary.contains("BrowserSkill extension"));
let hint = info
.hint
.expect("version mismatch should have a repair hint");
assert!(hint.contains("update bsk"));
assert!(hint.contains("reload the extension"));
assert!(hint.contains("bsk daemon restart"));
assert!(!hint.contains("min_compatible_protocol"));
}
#[test]
fn no_browser_hint_explains_version_mismatch_recovery() {
let hint = info_for(ErrorCode::NoBrowserConnected)
.hint
.expect("no browser should have a repair hint");
assert!(hint.contains("extension is already open"));
assert!(hint.contains("update"));
assert!(hint.contains("bsk daemon restart"));
}
/// Every code must carry a non-empty summary; an empty `summary`
/// would render `error: ` with nothing after it.
#[test]
+25 -92
View File
@@ -206,15 +206,15 @@ pub fn full_handler(status: DaemonStatus, state: Arc<DaemonState>) -> RpcHandler
Ok(v) => ResponseBody::Ok(v),
Err(e) => ResponseBody::Err(e),
},
Method::SessionStart => match handle_session_start(&state, rpc_id, params).await {
Method::SessionStart => match handle_session_start(&state, params).await {
Ok(v) => ResponseBody::Ok(v),
Err(e) => ResponseBody::Err(e),
},
Method::SessionStop => match handle_session_stop(&state, rpc_id, params).await {
Method::SessionStop => match handle_session_stop(&state, params).await {
Ok(v) => ResponseBody::Ok(v),
Err(e) => ResponseBody::Err(e),
},
Method::SessionStopAll => match handle_session_stop_all_rpc(&state, rpc_id).await {
Method::SessionStopAll => match handle_session_stop_all(&state).await {
Ok(v) => ResponseBody::Ok(v),
Err(e) => ResponseBody::Err(e),
},
@@ -434,9 +434,8 @@ async fn handle_wait_ms(
/// cancellation surfaces (M10.2 + review C2):
///
/// 1. [`AbortRegistry`] — answers daemon-local cancellable runners
/// (`tool.wait_ms` and `session.*` lifecycle calls). If a token is
/// registered we trip it and stop. Lifecycle handlers forward their
/// own WS cancel after preserving request-before-cancel ordering.
/// (currently `tool.wait_ms`). If a token is registered we trip it
/// and stop.
/// 2. [`super::inflight::ToolInflightRegistry`] — every IPC-tracked
/// `tool.*` RPC, registered the moment the IPC handler accepts the
/// request. Trips the entry's cancel token regardless of whether
@@ -444,10 +443,10 @@ async fn handle_wait_ms(
/// * **Queued** — the worker's pre-flight observes the cancelled
/// token and short-circuits with `cancelled` before any WS
/// frame leaves the daemon (review C2 fix).
/// * **Forwarded** — we push a WS-side `cancel { rpc_id: ws_rpc_id }`
/// frame so the extension's dispatcher can trip its
/// `AbortController`; the worker keeps the session busy until the
/// extension returns its final result or the cleanup timeout expires.
/// * **Forwarded** — the worker's `tokio::select!` returns
/// `cancelled` immediately, AND we additionally push a WS-side
/// `cancel { rpc_id: ws_rpc_id }` frame so the extension's
/// dispatcher can trip its `AbortController`.
///
/// Returns `{ cancelled }` reflecting whether either surface
/// matched. The RPC itself never errors — a cancelled tool surfaces
@@ -635,20 +634,7 @@ fn clamp_browser_wait(wait_ms: Option<u64>) -> Option<Duration> {
))
}
async fn handle_session_start(
state: &Arc<DaemonState>,
rpc_id: RpcId,
params: Value,
) -> Result<Value, RpcError> {
let abort_guard = state
.abort_registry
.register(rpc_id)
.map_err(|err| RpcError {
code: ErrorCode::ProtocolError,
message: format!("session.start cancellation registration failed: {err:?}"),
data: None,
})?;
let cancel = abort_guard.token().clone();
async fn handle_session_start(state: &Arc<DaemonState>, params: Value) -> Result<Value, RpcError> {
let params: CliSessionStartParams = if params.is_null() {
CliSessionStartParams {
browser_instance_id: None,
@@ -687,7 +673,6 @@ async fn handle_session_start(
},
state.config.extension_connect_wait,
DEFAULT_RPC_TIMEOUT,
Some(cancel),
)
.await
{
@@ -711,8 +696,6 @@ fn map_start_error(err: StartSessionError) -> RpcError {
StartSessionError::AmbiguousBrowserLabel { .. } => ErrorCode::InvalidParams,
StartSessionError::IdExhausted => ErrorCode::ProtocolError,
StartSessionError::Timeout => ErrorCode::Timeout,
StartSessionError::Cancelled => ErrorCode::Cancelled,
StartSessionError::CleanupFailed { .. } => ErrorCode::ProtocolError,
StartSessionError::TransportClosed => ErrorCode::ProtocolError,
StartSessionError::ExtensionError(inner) => inner.code,
};
@@ -728,16 +711,6 @@ fn map_start_error(err: StartSessionError) -> RpcError {
"label": label,
"instance_ids": instance_ids,
})),
StartSessionError::CleanupFailed {
session_id,
agent_window_id,
..
} => Some(serde_json::json!({
"reason": "cleanup_failed",
"session_id": session_id.0,
"agent_window_id": agent_window_id,
})),
StartSessionError::ExtensionError(inner) => inner.data.clone(),
_ => None,
};
RpcError {
@@ -747,20 +720,7 @@ fn map_start_error(err: StartSessionError) -> RpcError {
}
}
async fn handle_session_stop(
state: &Arc<DaemonState>,
rpc_id: RpcId,
params: Value,
) -> Result<Value, RpcError> {
let abort_guard = state
.abort_registry
.register(rpc_id)
.map_err(|err| RpcError {
code: ErrorCode::ProtocolError,
message: format!("session.stop cancellation registration failed: {err:?}"),
data: None,
})?;
let cancel = abort_guard.token().clone();
async fn handle_session_stop(state: &Arc<DaemonState>, params: Value) -> Result<Value, RpcError> {
let params: CliSessionStopParams = if params.is_null() {
CliSessionStopParams {
session_id: None,
@@ -774,7 +734,7 @@ async fn handle_session_stop(
})?
};
if params.all {
return handle_session_stop_all(state, Some(cancel)).await;
return handle_session_stop_all(state).await;
}
let session_id = match params.session_id {
Some(s) => SessionId(s),
@@ -793,7 +753,6 @@ async fn handle_session_stop(
&state.session_interrupts,
&session_id,
DEFAULT_SESSION_STOP_TIMEOUT,
Some(cancel),
)
.await
{
@@ -827,7 +786,19 @@ async fn handle_session_stop(
}
}
/// Marks a `session.stop` rejected because another teardown already owns the
/// session, so callers that race on purpose (`bsk record`) can tell it apart
/// from a browser that stopped responding.
pub const STOP_IN_PROGRESS_REASON: &str = "session_stop_in_progress";
fn map_stop_error(err: StopSessionError) -> RpcError {
if matches!(err, StopSessionError::Stopping) {
return RpcError {
code: ErrorCode::Timeout,
message: err.to_string(),
data: Some(serde_json::json!({ "reason": STOP_IN_PROGRESS_REASON })),
};
}
let code = match &err {
StopSessionError::NotFound => ErrorCode::NotFound,
StopSessionError::Stopping => ErrorCode::Timeout,
@@ -845,25 +816,7 @@ fn map_stop_error(err: StopSessionError) -> RpcError {
}
}
async fn handle_session_stop_all_rpc(
state: &Arc<DaemonState>,
rpc_id: RpcId,
) -> Result<Value, RpcError> {
let abort_guard = state
.abort_registry
.register(rpc_id)
.map_err(|err| RpcError {
code: ErrorCode::ProtocolError,
message: format!("session.stop_all cancellation registration failed: {err:?}"),
data: None,
})?;
handle_session_stop_all(state, Some(abort_guard.token().clone())).await
}
async fn handle_session_stop_all(
state: &Arc<DaemonState>,
cancel: Option<super::abort::AbortToken>,
) -> Result<Value, RpcError> {
async fn handle_session_stop_all(state: &Arc<DaemonState>) -> Result<Value, RpcError> {
let ids: Vec<SessionId> = state
.sessions
.snapshot()
@@ -875,16 +828,6 @@ async fn handle_session_stop_all(
let mut returned_tab_ids = Vec::new();
let mut return_failures = Vec::new();
for id in ids {
if cancel
.as_ref()
.is_some_and(super::abort::AbortToken::is_cancelled)
{
return Err(RpcError {
code: ErrorCode::Cancelled,
message: "session.stop_all was cancelled".into(),
data: None,
});
}
match stop_session(
&state.browsers,
&state.sessions,
@@ -892,7 +835,6 @@ async fn handle_session_stop_all(
&state.session_interrupts,
&id,
DEFAULT_SESSION_STOP_TIMEOUT,
cancel.clone(),
)
.await
{
@@ -918,15 +860,6 @@ async fn handle_session_stop_all(
returned_tab_ids.extend(stop.returned_tab_ids);
return_failures.extend(stop.return_failures);
}
Err(err)
if matches!(
&err,
StopSessionError::ExtensionError(inner)
if inner.code == ErrorCode::Cancelled
) =>
{
return Err(map_stop_error(err));
}
Err(err) => {
debug!(session = %id, ?err, "session.stop_all: failure (continuing)");
let rpc = map_stop_error(err);
+59 -274
View File
@@ -43,7 +43,6 @@ use serde_json::Value;
use tokio::sync::{mpsc, oneshot};
use tracing::{debug, warn};
use super::abort::AbortToken;
use super::browsers::BrowserRegistry;
use super::inflight::{PromoteOutcome, ToolInflightEntry};
use super::sessions::{SessionId, SessionRegistry};
@@ -59,11 +58,6 @@ pub const QUEUE_CAPACITY: usize = 64;
/// so tests can default it.
pub const DEFAULT_TOOL_TIMEOUT: Duration = Duration::from_secs(30);
/// Once a forwarded request is cancelled, keep the session queue busy while
/// the extension finishes handler-side compensation. This is deliberately
/// bounded so a wedged extension cannot pin the session forever.
pub const CANCEL_CLEANUP_TIMEOUT: Duration = Duration::from_secs(2);
/// Job submitted into a session queue. Carries everything the worker
/// needs to forward one RPC and one oneshot to deliver the answer.
pub struct ToolJob {
@@ -76,22 +70,14 @@ pub struct ToolJob {
/// the queued/forwarded state machine; the worker checks it on
/// pre-flight (so a cancel arriving while the job was still
/// queued short-circuits without ever touching the extension)
/// and observes `cancel.cancelled()` while awaiting the WS response.
/// A forwarded cancel keeps the worker busy until the extension's
/// final response arrives or the bounded cleanup timeout expires.
/// and selects on `cancel.cancelled()` while awaiting the WS
/// response (so a cancel arriving mid-flight unblocks the worker
/// the same way the WS-side cancel frame already did).
///
/// `None` for daemon-internal callers that do not flow through an
/// IPC request id (e.g. `session.stop`'s queued teardown call —
/// those already carry their own bespoke retry / abort path).
pub inflight: Option<Arc<ToolInflightEntry>>,
/// Cancellation token for session-lifecycle work registered in the
/// daemon-local abort registry. Unlike `inflight`, lifecycle callers
/// forward their own WS cancel after the request has been queued.
pub lifecycle_cancel: Option<AbortToken>,
/// How long cancellation keeps the worker busy while the extension
/// finishes compensation. Session teardown uses its full RPC budget
/// so daemon state is reconciled even after the CLI exits.
pub cancel_cleanup_timeout: Duration,
/// Owning session — used by `ToolInflightRegistry::cancel_session`
/// to drain queued jobs that have not yet been promoted to
/// "forwarded". Worker callers also have access to it via the
@@ -331,8 +317,6 @@ impl ToolQueueRegistry {
timeout,
false,
inflight,
None,
CANCEL_CLEANUP_TIMEOUT,
)
.await;
// A long-running request may outlive the idle threshold. Touching
@@ -371,8 +355,6 @@ impl ToolQueueRegistry {
timeout,
respond: respond_tx,
inflight,
lifecycle_cancel: None,
cancel_cleanup_timeout: CANCEL_CLEANUP_TIMEOUT,
session_id: sid.clone(),
};
match forward_one(sid, &self.browsers, &self.sessions, &job).await {
@@ -391,7 +373,6 @@ impl ToolQueueRegistry {
method: Method,
params: Value,
timeout: Duration,
cancel: Option<AbortToken>,
) -> Result<Value, DispatchError> {
let (sender, state) = {
let mut guard = self.queues.lock().expect("tool queue registry poisoned");
@@ -418,8 +399,6 @@ impl ToolQueueRegistry {
timeout,
true,
None,
cancel,
timeout,
)
.await
}
@@ -435,8 +414,6 @@ async fn dispatch_with_sender(
timeout: Duration,
wait_for_capacity: bool,
inflight: Option<Arc<ToolInflightEntry>>,
lifecycle_cancel: Option<AbortToken>,
cancel_cleanup_timeout: Duration,
) -> Result<Value, DispatchError> {
let (respond_tx, respond_rx) = oneshot::channel();
let job = ToolJob {
@@ -445,11 +422,8 @@ async fn dispatch_with_sender(
timeout,
respond: respond_tx,
inflight,
lifecycle_cancel,
cancel_cleanup_timeout,
session_id,
};
let lifecycle_cancellable = job.lifecycle_cancel.is_some();
if wait_for_capacity {
let waited = tokio::time::timeout(
timeout.saturating_add(Duration::from_secs(1)),
@@ -474,15 +448,8 @@ async fn dispatch_with_sender(
mpsc::error::TrySendError::Closed(_) => DispatchError::QueueClosed,
});
}
let mut response_timeout = timeout.saturating_add(Duration::from_secs(1));
if lifecycle_cancellable {
// A lifecycle cancel can arrive at the original RPC deadline and
// intentionally keeps the worker alive while extension cleanup
// settles. Do not let this outer waiter abandon that reconciliation
// early, which could reopen a daemon session whose window is gone.
response_timeout = response_timeout.saturating_add(cancel_cleanup_timeout);
}
let waited = tokio::time::timeout(response_timeout, respond_rx).await;
let waited =
tokio::time::timeout(timeout.saturating_add(Duration::from_secs(1)), respond_rx).await;
match waited {
Ok(Ok(Ok(v))) => Ok(v),
Ok(Ok(Err(rpc))) => Err(DispatchError::Rpc(rpc)),
@@ -520,17 +487,11 @@ async fn forward_one(
// / browser resolution work, and before any WS frame leaves the
// daemon (review C2). We keep the `tokio::select!` later so
// cancels arriving mid-WS-hop also unblock the worker.
if job
.inflight
.as_ref()
.is_some_and(|entry| entry.is_cancelled())
|| job
.lifecycle_cancel
.as_ref()
.is_some_and(AbortToken::is_cancelled)
if let Some(entry) = job.inflight.as_ref()
&& entry.is_cancelled()
{
return Err(cancelled_error(
job.inflight.as_deref(),
Some(entry),
"tool dispatch cancelled before forwarding",
));
}
@@ -594,20 +555,11 @@ async fn forward_one(
}
}
None => {
// Daemon-internal callers do not carry a tool-inflight entry.
// Session lifecycle jobs instead use the abort-registry token
// below; other internal callers keep the plain send path.
if job
.lifecycle_cancel
.as_ref()
.is_some_and(AbortToken::is_cancelled)
{
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(cancelled_error(
None,
"session lifecycle cancelled before forwarding",
));
}
// Daemon-internal callers (e.g. queued session.stop drain)
// don't carry an inflight entry; they cannot be cancelled
// through the IPC `cancel` surface, so a plain sink send
// is sufficient and keeps the existing transport-error
// behaviour.
if client.sink.send(request).is_err() {
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(RpcError {
@@ -616,90 +568,23 @@ async fn forward_one(
data: None,
});
}
job.lifecycle_cancel.clone()
None
}
};
let forward_lifecycle_cancel = || {
if job.lifecycle_cancel.is_none() {
return;
}
let cancel = Frame::Request(RequestFrame {
id: format!("cancel-{rpc_id}"),
method: Method::Cancel,
params: Some(serde_json::json!({ "rpc_id": rpc_id })),
});
if let Err(err) = client.sink.send(cancel) {
warn!(%rpc_id, ?err, "failed to forward session lifecycle cancel");
}
};
let on_abort: Option<&(dyn Fn() + Sync)> = job
.lifecycle_cancel
.as_ref()
.map(|_| &forward_lifecycle_cancel as &(dyn Fn() + Sync));
let waited = await_with_optional_cancel(
job.timeout,
job.cancel_cleanup_timeout,
waiter,
cancel_token.as_ref(),
on_abort,
)
.await;
let waited = await_with_optional_cancel(job.timeout, waiter, cancel_token.as_ref()).await;
let response = match waited {
WaitOutcome::Response(resp) => resp,
WaitOutcome::CancelledAfterResponse(resp) => {
if job.method == Method::ToolSessionStop
&& let ResponseBody::Ok(value) = resp.body
{
// Teardown crossed its final irreversible boundary before
// the cancel landed. Commit the real extension result so the
// daemon does not retain a session whose window is gone.
return Ok(value);
}
// Cancellation wins the external verdict, but only after the
// extension's original RPC has settled. Preserve any non-cancel
// error so compensation failures remain explicit instead of
// being hidden behind a generic cancelled result.
if let ResponseBody::Err(err) = resp.body
&& !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted)
{
return Err(err);
}
WaitOutcome::Cancelled => {
// Drop the WS waiter so a late extension reply is dropped
// cleanly (otherwise pending.resolve would log a stale
// entry). The CLI caller will get a synthesised
// `cancelled` here even if the extension never answers.
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(cancelled_error(
job.inflight.as_deref(),
"tool dispatch cancelled after extension cleanup",
"tool dispatch cancelled mid-flight",
));
}
WaitOutcome::TimedOutAfterResponse(resp) => match resp.body {
ResponseBody::Ok(value) if job.method == Method::ToolSessionStop => {
// The close crossed its irreversible boundary during the
// timeout cleanup grace. Reconcile daemon state instead of
// reopening a session whose window is already gone.
return Ok(value);
}
ResponseBody::Err(err)
if !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted) =>
{
return Err(err);
}
_ => {
return Err(RpcError {
code: ErrorCode::Timeout,
message: format!("tool RPC timed out after {:?}", job.timeout),
data: None,
});
}
},
WaitOutcome::CleanupTimeout => {
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(RpcError {
code: ErrorCode::Timeout,
message: format!(
"cancelled tool did not finish cleanup within {:?}",
job.cancel_cleanup_timeout
),
data: Some(serde_json::json!({ "reason": "cancel_cleanup_timeout" })),
});
}
WaitOutcome::WaiterClosed => {
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(RpcError {
@@ -726,57 +611,47 @@ async fn forward_one(
#[derive(Debug)]
enum WaitOutcome {
Response(bsk_protocol::ResponseFrame),
CancelledAfterResponse(bsk_protocol::ResponseFrame),
TimedOutAfterResponse(bsk_protocol::ResponseFrame),
CleanupTimeout,
Cancelled,
WaiterClosed,
Timeout,
}
async fn await_with_optional_cancel(
timeout: Duration,
cleanup_timeout: Duration,
mut waiter: oneshot::Receiver<bsk_protocol::ResponseFrame>,
waiter: oneshot::Receiver<bsk_protocol::ResponseFrame>,
cancel: Option<&super::abort::AbortToken>,
on_abort: Option<&(dyn Fn() + Sync)>,
) -> WaitOutcome {
match cancel {
Some(token) => {
let deadline = tokio::time::sleep(timeout);
tokio::pin!(deadline);
tokio::select! {
// Cancel keeps same-tick priority, but it no longer drops the
// waiter. The worker remains busy until the extension replies
// after compensation or the cleanup deadline expires.
biased;
_ = token.cancelled() => {
if let Some(on_abort) = on_abort {
on_abort();
}
match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
Ok(Ok(resp)) => WaitOutcome::CancelledAfterResponse(resp),
Ok(Err(_)) => WaitOutcome::WaiterClosed,
Err(_) => WaitOutcome::CleanupTimeout,
}
},
outcome = &mut waiter => match outcome {
Ok(resp) => WaitOutcome::Response(resp),
Err(_) => WaitOutcome::WaiterClosed,
},
_ = &mut deadline => {
let Some(on_abort) = on_abort else {
return WaitOutcome::Timeout;
};
on_abort();
match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
Ok(Ok(resp)) => WaitOutcome::TimedOutAfterResponse(resp),
Ok(Err(_)) => WaitOutcome::WaiterClosed,
Err(_) => WaitOutcome::CleanupTimeout,
}
},
}
}
None => match tokio::time::timeout(timeout, &mut waiter).await {
// Cancel wins if both ready: when a cancel notification and
// the extension's response are both observable on the same
// tokio tick, `biased;` polls `token.cancelled()` first and
// resolves the whole `select!` to `WaitOutcome::Cancelled`.
// The already-arrived tool result is intentionally dropped.
//
// This extends design §4.6 — which only specifies "CLI sends
// `cancel` on SIGINT and waits for the extension's
// `cancelled` reply (≤ 2s before forced exit)" — by pinning
// the same-tick race to a single, observable verdict.
// Without `biased;`, an extension racing a fast successful
// reply against a cancel notification could occasionally
// resolve as `ok` even though the caller had already moved on
// to compensation logic, leaving the agent trusting state the
// daemon had just been asked to roll back. Picking
// cancel-wins keeps the external observation rule simple
// ("once a cancel is in flight, the in-flight RPC's verdict
// is `cancelled`") and matches what the CLI / agent already
// assumes after firing SIGINT. Round 3 M1 / round 4 M2
// nail-down.
Some(token) => tokio::select! {
biased;
_ = token.cancelled() => WaitOutcome::Cancelled,
outcome = tokio::time::timeout(timeout, waiter) => match outcome {
Ok(Ok(resp)) => WaitOutcome::Response(resp),
Ok(Err(_)) => WaitOutcome::WaiterClosed,
Err(_) => WaitOutcome::Timeout,
},
},
None => match tokio::time::timeout(timeout, waiter).await {
Ok(Ok(resp)) => WaitOutcome::Response(resp),
Ok(Err(_)) => WaitOutcome::WaiterClosed,
Err(_) => WaitOutcome::Timeout,
@@ -858,17 +733,10 @@ mod await_with_optional_cancel_tests {
let (tx, rx) = oneshot::channel();
tx.send(dummy_response()).unwrap();
let outcome = await_with_optional_cancel(
Duration::from_secs(10),
Duration::from_secs(1),
rx,
Some(&token),
None,
)
.await;
let outcome = await_with_optional_cancel(Duration::from_secs(10), rx, Some(&token)).await;
assert!(
matches!(outcome, WaitOutcome::CancelledAfterResponse(_)),
"expected cancellation to win while retaining the ready response"
matches!(outcome, WaitOutcome::Cancelled),
"expected Cancelled when both arms are ready under biased; (got something else)"
);
}
@@ -881,14 +749,7 @@ mod await_with_optional_cancel_tests {
let (tx, rx) = oneshot::channel();
tx.send(dummy_response()).unwrap();
let outcome = await_with_optional_cancel(
Duration::from_secs(10),
Duration::from_secs(1),
rx,
Some(&token),
None,
)
.await;
let outcome = await_with_optional_cancel(Duration::from_secs(10), rx, Some(&token)).await;
match outcome {
WaitOutcome::Response(frame) => {
assert!(matches!(frame.body, ResponseBody::Ok(_)));
@@ -901,83 +762,9 @@ mod await_with_optional_cancel_tests {
async fn no_cancel_token_still_returns_response() {
let (tx, rx) = oneshot::channel();
tx.send(dummy_response()).unwrap();
let outcome = await_with_optional_cancel(
Duration::from_secs(10),
Duration::from_secs(1),
rx,
None,
None,
)
.await;
let outcome = await_with_optional_cancel(Duration::from_secs(10), rx, None).await;
assert!(matches!(outcome, WaitOutcome::Response(_)));
}
#[tokio::test]
async fn cancelled_waiter_stays_pending_until_cleanup_response_arrives() {
let token = AbortToken::new();
let (tx, rx) = oneshot::channel();
token.cancel();
let pending = tokio::spawn(async move {
await_with_optional_cancel(
Duration::from_secs(10),
Duration::from_secs(1),
rx,
Some(&token),
None,
)
.await
});
tokio::task::yield_now().await;
assert!(!pending.is_finished());
tx.send(dummy_response()).unwrap();
assert!(matches!(
pending.await.unwrap(),
WaitOutcome::CancelledAfterResponse(_)
));
}
#[tokio::test]
async fn cancelled_waiter_releases_after_cleanup_timeout() {
let token = AbortToken::new();
let (_tx, rx) = oneshot::channel();
token.cancel();
let outcome = await_with_optional_cancel(
Duration::from_secs(10),
Duration::from_millis(10),
rx,
Some(&token),
None,
)
.await;
assert!(matches!(outcome, WaitOutcome::CleanupTimeout));
}
#[tokio::test]
async fn lifecycle_timeout_forwards_abort_and_waits_for_cleanup_response() {
use std::sync::atomic::{AtomicBool, Ordering};
let token = AbortToken::new();
let (tx, rx) = oneshot::channel();
let abort_forwarded = AtomicBool::new(false);
let forward_abort = || abort_forwarded.store(true, Ordering::SeqCst);
tokio::spawn(async move {
tokio::time::sleep(Duration::from_millis(20)).await;
tx.send(dummy_response()).unwrap();
});
let outcome = await_with_optional_cancel(
Duration::from_millis(5),
Duration::from_millis(100),
rx,
Some(&token),
Some(&forward_abort),
)
.await;
assert!(abort_forwarded.load(Ordering::SeqCst));
assert!(matches!(outcome, WaitOutcome::TimedOutAfterResponse(_)));
}
}
#[cfg(test)]
@@ -1040,8 +827,6 @@ mod tool_job_session_id_tests {
timeout: Duration::from_secs(1),
respond: tx,
inflight: None,
lifecycle_cancel: None,
cancel_cleanup_timeout: CANCEL_CLEANUP_TIMEOUT,
session_id: SessionId("sess-A".into()),
};
assert_eq!(job.session_id.0, "sess-A");
+26 -223
View File
@@ -12,14 +12,13 @@ use bsk_protocol::system::{BrowserStatusEntry, SessionStatusEntry};
use bsk_protocol::tools::{
SessionStartParams, SessionStartResult, SessionStopParams, SessionStopResult,
};
use bsk_protocol::{ErrorCode, Frame, Method, RequestFrame, ResponseBody, RpcError, RpcId};
use bsk_protocol::{Frame, RequestFrame, ResponseBody, RpcError, RpcId};
use rand::Rng;
use serde::{Deserialize, Serialize};
use tokio::time::{Duration, timeout};
use super::abort::AbortToken;
use super::browsers::{BrowserClient, BrowserId, BrowserRegistry, SelectError};
use super::queue::{CANCEL_CLEANUP_TIMEOUT, DispatchError, ToolQueueRegistry};
use super::queue::{DispatchError, ToolQueueRegistry};
use super::session_interrupt::SessionInterruptRegistry;
#[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
@@ -340,14 +339,6 @@ pub enum StartSessionError {
IdExhausted,
#[error("session creation timed out waiting for extension")]
Timeout,
#[error("session creation was cancelled")]
Cancelled,
#[error("session creation cancellation cleanup failed for {session_id}: {message}")]
CleanupFailed {
session_id: SessionId,
agent_window_id: Option<i64>,
message: String,
},
#[error("extension rejected tool.session_start: {0:?}")]
ExtensionError(RpcError),
#[error("transport closed while waiting for extension response")]
@@ -363,8 +354,6 @@ impl StartSessionError {
StartSessionError::AmbiguousBrowserLabel { .. } => "invalid_params",
StartSessionError::IdExhausted => "protocol_error",
StartSessionError::Timeout => "timeout",
StartSessionError::Cancelled => "cancelled",
StartSessionError::CleanupFailed { .. } => "protocol_error",
StartSessionError::TransportClosed => "protocol_error",
StartSessionError::ExtensionError(err) => match err.code {
bsk_protocol::ErrorCode::Timeout => "timeout",
@@ -418,7 +407,6 @@ pub struct AgentWindowOptions {
/// On success the matching per-session dispatch queue (M6.5,
/// design §5) is also spawned so that subsequent `tool.*` RPCs serialise
/// against this session's Agent Window / ref-store.
#[allow(clippy::too_many_arguments)]
pub async fn start_session(
registry: &Arc<BrowserRegistry>,
sessions: &Arc<SessionRegistry>,
@@ -427,32 +415,24 @@ pub async fn start_session(
window: AgentWindowOptions,
connect_wait: Duration,
timeout_dur: Duration,
cancel: Option<AbortToken>,
) -> Result<Session, StartSessionError> {
let selection = registry.select_with_connect_wait(requested, connect_wait);
tokio::pin!(selection);
let selected = match cancel.as_ref() {
Some(token) => tokio::select! {
biased;
_ = token.cancelled() => return Err(StartSessionError::Cancelled),
selected = &mut selection => selected,
},
None => selection.await,
};
let client: Arc<BrowserClient> = selected.map_err(|e| match e {
SelectError::NoBrowserConnected => StartSessionError::NoBrowserConnected,
SelectError::MultipleBrowsersOnline => StartSessionError::MultipleBrowsersOnline {
browsers: snapshot_status_entries(registry, sessions),
},
SelectError::NotFound => StartSessionError::BrowserNotFound,
SelectError::AmbiguousLabel {
label,
instance_ids,
} => StartSessionError::AmbiguousBrowserLabel {
label,
instance_ids,
},
})?;
let client: Arc<BrowserClient> = registry
.select_with_connect_wait(requested, connect_wait)
.await
.map_err(|e| match e {
SelectError::NoBrowserConnected => StartSessionError::NoBrowserConnected,
SelectError::MultipleBrowsersOnline => StartSessionError::MultipleBrowsersOnline {
browsers: snapshot_status_entries(registry, sessions),
},
SelectError::NotFound => StartSessionError::BrowserNotFound,
SelectError::AmbiguousLabel {
label,
instance_ids,
} => StartSessionError::AmbiguousBrowserLabel {
label,
instance_ids,
},
})?;
let session_id = sessions
.reserve_id(client.id.clone(), SESSION_ID_MAX_RESERVE_ATTEMPTS, now_ms)
.ok_or(StartSessionError::IdExhausted)?;
@@ -473,71 +453,22 @@ pub async fn start_session(
let mut pending = client.pending.lock().unwrap();
pending.register(rpc_id.clone())
};
if cancel.as_ref().is_some_and(AbortToken::is_cancelled) {
sessions.cancel_reservation(&session_id);
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(StartSessionError::Cancelled);
}
if client.sink.send(Frame::Request(request)).is_err() {
sessions.cancel_reservation(&session_id);
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(StartSessionError::TransportClosed);
}
let mut waiter = waiter;
enum StartWaitOutcome {
Response(bsk_protocol::ResponseFrame),
WaiterClosed,
Cancelled,
Timeout,
}
let wait_outcome = match cancel.as_ref() {
Some(token) => {
let deadline = tokio::time::sleep(timeout_dur);
tokio::pin!(deadline);
tokio::select! {
biased;
_ = token.cancelled() => StartWaitOutcome::Cancelled,
response = &mut waiter => match response {
Ok(response) => StartWaitOutcome::Response(response),
Err(_) => StartWaitOutcome::WaiterClosed,
},
_ = &mut deadline => StartWaitOutcome::Timeout,
}
}
None => match timeout(timeout_dur, &mut waiter).await {
Ok(Ok(response)) => StartWaitOutcome::Response(response),
Ok(Err(_)) => StartWaitOutcome::WaiterClosed,
Err(_) => StartWaitOutcome::Timeout,
},
};
let response = match wait_outcome {
StartWaitOutcome::Response(response) => response,
StartWaitOutcome::WaiterClosed => {
let response = match timeout(timeout_dur, waiter).await {
Ok(Ok(resp)) => resp,
Ok(Err(_)) => {
sessions.cancel_reservation(&session_id);
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(StartSessionError::TransportClosed);
}
StartWaitOutcome::Cancelled => {
return Err(finish_aborted_start(
&client,
sessions,
&session_id,
&rpc_id,
waiter,
StartAbortReason::Cancelled,
)
.await);
}
StartWaitOutcome::Timeout => {
return Err(finish_aborted_start(
&client,
sessions,
&session_id,
&rpc_id,
waiter,
StartAbortReason::Timeout,
)
.await);
Err(_) => {
sessions.cancel_reservation(&session_id);
client.pending.lock().unwrap().cancel(&rpc_id);
return Err(StartSessionError::Timeout);
}
};
let agent_window_id = match response.body {
@@ -573,124 +504,6 @@ pub async fn start_session(
Ok(session)
}
#[derive(Clone, Copy)]
enum StartAbortReason {
Cancelled,
Timeout,
}
impl StartAbortReason {
fn error(self) -> StartSessionError {
match self {
Self::Cancelled => StartSessionError::Cancelled,
Self::Timeout => StartSessionError::Timeout,
}
}
}
async fn finish_aborted_start(
client: &Arc<BrowserClient>,
sessions: &Arc<SessionRegistry>,
session_id: &SessionId,
rpc_id: &RpcId,
mut waiter: tokio::sync::oneshot::Receiver<bsk_protocol::ResponseFrame>,
reason: StartAbortReason,
) -> StartSessionError {
sessions.cancel_reservation(session_id);
let cancel = Frame::Request(RequestFrame {
id: format!("cancel-{rpc_id}"),
method: Method::Cancel,
params: Some(serde_json::json!({ "rpc_id": rpc_id })),
});
if client.sink.send(cancel).is_err() {
client.pending.lock().unwrap().cancel(rpc_id);
return StartSessionError::TransportClosed;
}
match timeout(CANCEL_CLEANUP_TIMEOUT, &mut waiter).await {
Ok(Ok(response)) => match response.body {
ResponseBody::Err(err)
if matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted) =>
{
reason.error()
}
ResponseBody::Err(err) => StartSessionError::ExtensionError(err),
ResponseBody::Ok(value) => {
let agent_window_id = serde_json::from_value::<SessionStartResult>(value)
.ok()
.and_then(|result| result.agent_window_id);
match rollback_extension_session(client, session_id).await {
Ok(()) => reason.error(),
Err(message) => StartSessionError::CleanupFailed {
session_id: session_id.clone(),
agent_window_id,
message,
},
}
}
},
Ok(Err(_)) => {
client.pending.lock().unwrap().cancel(rpc_id);
StartSessionError::TransportClosed
}
Err(_) => {
// The cancel frame is already queued behind session_start. Even
// if chrome.windows.create is still blocked, the extension will
// observe the aborted signal and compensate before replying.
client.pending.lock().unwrap().cancel(rpc_id);
reason.error()
}
}
}
async fn rollback_extension_session(
client: &Arc<BrowserClient>,
session_id: &SessionId,
) -> Result<(), String> {
let rollback_id = next_rpc_id("sess-start-rollback");
let waiter = {
let mut pending = client.pending.lock().unwrap();
pending.register(rollback_id.clone())
};
let request = RequestFrame {
id: rollback_id.clone(),
method: Method::ToolSessionStop,
params: Some(
serde_json::to_value(SessionStopParams {
session_id: session_id.0.clone(),
})
.expect("serialize session_start rollback params"),
),
};
if client.sink.send(Frame::Request(request)).is_err() {
client.pending.lock().unwrap().cancel(&rollback_id);
return Err("browser disconnected before rollback was queued".into());
}
let response = match timeout(CANCEL_CLEANUP_TIMEOUT, waiter).await {
Ok(Ok(response)) => response,
Ok(Err(_)) => return Err("browser disconnected during rollback".into()),
Err(_) => {
client.pending.lock().unwrap().cancel(&rollback_id);
return Err(format!(
"extension did not finish rollback within {CANCEL_CLEANUP_TIMEOUT:?}"
));
}
};
match response.body {
ResponseBody::Ok(value) => {
let result: SessionStopResult = serde_json::from_value(value)
.map_err(|err| format!("invalid rollback response: {err}"))?;
if result.return_failures.is_empty() {
Ok(())
} else {
Err("rollback reported borrowed-tab return failures".into())
}
}
ResponseBody::Err(err) if err.code == ErrorCode::NotFound => Ok(()),
ResponseBody::Err(err) => Err(format!("rollback failed: {} ({:?})", err.message, err.code)),
}
}
/// Tear down a single session id (round-trips `tool.session_stop` to the
/// owning extension and unregisters on success). Also tears down the
/// per-session dispatch queue so the worker task exits.
@@ -701,7 +514,6 @@ pub async fn stop_session(
interrupts: &Arc<SessionInterruptRegistry>,
session_id: &SessionId,
timeout_dur: Duration,
cancel: Option<AbortToken>,
) -> Result<SessionStopResult, StopSessionError> {
let session = sessions.get(session_id).ok_or(StopSessionError::NotFound)?;
if registry.get(&session.browser_id).is_none() {
@@ -712,17 +524,12 @@ pub async fn stop_session(
session_id: session_id.0.clone(),
})
.unwrap();
// Internal callers (idle reaping and shutdown) do not have a CLI-owned
// token, but still need timeout-driven WS cancellation so a late window
// close cannot leave daemon state pointing at a vanished session.
let lifecycle_cancel = Some(cancel.unwrap_or_default());
match queues
.dispatch_after_closing(
session_id,
bsk_protocol::Method::ToolSessionStop,
params,
timeout_dur,
lifecycle_cancel,
)
.await
{
@@ -760,10 +567,6 @@ pub async fn stop_session(
drop_session_local(queues, interrupts, session_id);
return Ok(SessionStopResult::default());
}
if err.code == bsk_protocol::ErrorCode::Timeout {
queues.reopen(session_id);
return Err(StopSessionError::Timeout);
}
queues.reopen(session_id);
Err(StopSessionError::ExtensionError(err))
}
+2 -3
View File
@@ -27,7 +27,7 @@ use crate::daemon::{
browsers::{BROWSER_LIVENESS_TICK, BROWSER_LIVENESS_TIMEOUT, EXTENSION_CONNECT_WAIT},
info as daemon_info, ipc, lockfile, paths,
sessions::{StopSessionError, forget_session, stop_session},
state::DaemonState,
state::{DaemonState, PROTOCOL_VERSION},
ws,
};
@@ -299,7 +299,7 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
ws_port,
sock_path: sock_path.clone(),
daemon_version: env!("CARGO_PKG_VERSION"),
protocol_version: "1.0",
protocol_version: PROTOCOL_VERSION,
};
let handler = ipc::full_handler(status, Arc::clone(&state));
@@ -491,7 +491,6 @@ pub(crate) fn spawn_session_idle_reaper(state: Arc<DaemonState>) -> tokio::task:
&state.session_interrupts,
&session_id,
Duration::from_secs(10),
None,
)
.await
{
+5 -5
View File
@@ -16,9 +16,9 @@ use super::start::DaemonConfig;
use super::ws::WsHandle;
pub const DAEMON_VERSION: &str = env!("CARGO_PKG_VERSION");
pub const PROTOCOL_VERSION: &str = "1.0";
/// Lowest **protocol** version peers must speak (e.g. `"1.0"`).
pub const MIN_COMPATIBLE_PROTOCOL: &str = "1.0";
pub const PROTOCOL_VERSION: &str = "1.1";
/// Lowest **protocol** version peers must speak (e.g. `"1.1"`).
pub const MIN_COMPATIBLE_PROTOCOL: &str = "1.1";
/// Legacy app-semver floor used only when `HandshakeResult.min_compatible_peer`
/// is emitted for old extensions. New code ignores this on read.
pub const LEGACY_MIN_COMPATIBLE_PEER: &str = "0.0.0";
@@ -34,8 +34,8 @@ pub struct DaemonState {
/// `stop_session` / browser disconnect.
pub tool_queues: Arc<ToolQueueRegistry>,
/// Per-rpc-id cancellation tokens for daemon-side long-runners
/// (`tool.wait_ms` plus `session.*` lifecycle calls). The CLI's
/// `cancel { rpc_id }` consults this registry first.
/// (M9.3 — currently only `tool.wait_ms`). The CLI's `cancel
/// { rpc_id }` consults this registry first.
pub abort_registry: Arc<AbortRegistry>,
/// Tracks `tool.*` RPCs that have been forwarded to an extension
/// over WS but have not yet received a response. Indexed by the
+2 -2
View File
@@ -112,14 +112,14 @@ async fn handshake_as_ext(
let params = HandshakeParams {
client: "browser-skill-extension".into(),
version: "0.1.0-dev.0".parse().unwrap(),
protocol_version: "1.0".into(),
protocol_version: "1.1".into(),
instance_id: TEST_EXT_ID.into(),
browser: BrowserPeerInfo {
name: "chrome".into(),
version: "131.0".into(),
},
min_compatible_peer: Some("0.1.0-dev.0".parse().unwrap()),
min_compatible_protocol: Some("1.0".into()),
min_compatible_protocol: Some("1.1".into()),
label: "Test".into(),
};
let req = RequestFrame {
+7 -62
View File
@@ -75,7 +75,7 @@ async fn handshake_as_ext(
let params = HandshakeParams {
client: "browser-skill-extension".into(),
version: "0.1.0-dev.0".parse().unwrap(),
protocol_version: "1.0".into(),
protocol_version: "1.1".into(),
instance_id: TEST_EXT_ID.into(),
browser: BrowserPeerInfo {
name: "chrome".into(),
@@ -83,7 +83,7 @@ async fn handshake_as_ext(
},
label: "Test".into(),
min_compatible_peer: Some("0.1.0-dev.0".parse().unwrap()),
min_compatible_protocol: Some("1.0".into()),
min_compatible_protocol: Some("1.1".into()),
};
let req = RequestFrame {
id: "hs".into(),
@@ -583,11 +583,11 @@ async fn cancel_for_unknown_rpc_id_returns_false() {
}
#[tokio::test]
async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
// A forwarded cancel is acknowledged immediately, but the extension
// deliberately delays the original RPC's final response to model
// handler-side compensation. The daemon must keep the session busy
// throughout that delay and release it only after cleanup settles.
async fn concurrent_tool_call_returns_session_busy_while_slow_rpc_inflight() {
// With session busy fast-fail, a second tool RPC submitted while
// the worker is occupied returns `session_busy` immediately
// instead of queuing behind the slow call. Cancel still unblocks
// the in-flight RPC through the existing forwarded-cancel path.
use std::sync::atomic::{AtomicUsize, Ordering as AOrdering};
let (handle, sock) = spawn_daemon().await;
let mut ws = connect_ext(handle.ws_addr()).await;
@@ -596,10 +596,6 @@ async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
let ws = Arc::new(tokio::sync::Mutex::new(ws));
let snapshots_seen = Arc::new(AtomicUsize::new(0));
let snapshots_seen_clone = Arc::clone(&snapshots_seen);
let cleanup_started = Arc::new(tokio::sync::Notify::new());
let cleanup_started_clone = Arc::clone(&cleanup_started);
let cleanup_release = Arc::new(tokio::sync::Notify::new());
let cleanup_release_clone = Arc::clone(&cleanup_release);
let ws_clone = Arc::clone(&ws);
let responder = tokio::spawn(async move {
let mut pending_snapshot: Option<String> = None;
@@ -640,16 +636,6 @@ async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
snapshots_seen_clone.fetch_add(1, AOrdering::SeqCst);
pending_snapshot = Some(req.id);
}
Method::ToolTabList => {
let reply = ResponseFrame {
id: req.id,
body: ResponseBody::Ok(json!({ "tabs": [] })),
};
let mut g = ws_clone.lock().await;
g.send(Message::Text(serde_json::to_string(&reply).unwrap()))
.await
.unwrap();
}
Method::Cancel => {
let target = req
.params
@@ -661,8 +647,6 @@ async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
if let (Some(target), Some(snap)) = (target, snapshot_id)
&& target == snap
{
cleanup_started_clone.notify_one();
cleanup_release_clone.notified().await;
let reply = ResponseFrame {
id: snap,
body: ResponseBody::Err(RpcError {
@@ -769,33 +753,6 @@ async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
.expect("cancel slow rpc succeeds");
assert!(cancel_slow.cancelled);
cleanup_started.notified().await;
let busy_after_cancel = tokio::time::timeout(
Duration::from_millis(200),
busy_ipc.call_with_id::<_, serde_json::Value>(
"snap-busy-after-cancel".into(),
Method::ToolSnapshot,
Some(json!({"session_id": session_id.clone()})),
Duration::from_secs(10),
),
)
.await
.expect("post-cancel RPC should fast-fail while cleanup is pending")
.unwrap();
let busy_after_cancel_err =
busy_after_cancel.expect_err("session must remain busy during delayed compensation");
assert_eq!(busy_after_cancel_err.code, ErrorCode::Timeout);
assert_eq!(
busy_after_cancel_err
.data
.as_ref()
.and_then(|d| d.get("reason"))
.and_then(|v| v.as_str()),
Some(bsk::rpc_reason::SESSION_BUSY)
);
cleanup_release.notify_one();
let slow_outcome = tokio::time::timeout(Duration::from_secs(5), slow_handle)
.await
.expect("slow snapshot did not resolve")
@@ -804,18 +761,6 @@ async fn cancel_keeps_session_busy_until_delayed_extension_cleanup_finishes() {
let slow_err = slow_outcome.expect_err("slow snapshot must surface cancelled");
assert_eq!(slow_err.code, ErrorCode::Cancelled);
let after_cleanup: serde_json::Value = busy_ipc
.call(
"tabs-after-cleanup",
Method::ToolTabList,
Some(json!({"session_id": session_id})),
Duration::from_secs(3),
)
.await
.unwrap()
.expect("session queue should reopen after extension cleanup");
assert_eq!(after_cleanup, json!({ "tabs": [] }));
assert_eq!(
snapshots_seen.load(AOrdering::SeqCst),
1,

Some files were not shown because too many files have changed in this diff Show More