perf(vom): optimize rendering algorithms

fix：将协议版本差异改为升级提醒

.github/workflows/release-extension.yml

@@ -325,3 +325,180 @@ jobs:
 
           echo "Chrome Web Store submission created successfully:"
           jq '{itemId, state, warningInfo}' "$PUBLISH_RESPONSE"
+
+  publish-edge-addons:
+    needs: [resolve, build-extension]
+    runs-on: ubuntu-latest
+    environment: edge-addons
+    permissions:
+      contents: read
+    steps:
+      - name: Download build artifact
+        uses: actions/download-artifact@v8
+        with:
+          name: extension-zip
+          path: dist
+
+      - name: Validate Microsoft Edge Add-ons configuration
+        shell: bash
+        env:
+          API_KEY: ${{ secrets.EDGE_ADDONS_API_KEY }}
+          CLIENT_ID: ${{ vars.EDGE_ADDONS_CLIENT_ID }}
+          PRODUCT_ID: ${{ vars.EDGE_ADDONS_PRODUCT_ID }}
+        run: |
+          set -euo pipefail
+          if [ -z "$API_KEY" ]; then
+            echo "::error::Environment secret EDGE_ADDONS_API_KEY is not configured"
+            exit 1
+          fi
+          if [ -z "$CLIENT_ID" ]; then
+            echo "::error::Environment variable EDGE_ADDONS_CLIENT_ID is not configured"
+            exit 1
+          fi
+          if [ -z "$PRODUCT_ID" ]; then
+            echo "::error::Environment variable EDGE_ADDONS_PRODUCT_ID is not configured"
+            exit 1
+          fi
+          if ! [[ "$PRODUCT_ID" =~ ^[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}$ ]]; then
+            echo "::error::EDGE_ADDONS_PRODUCT_ID must be the Partner Center product GUID, not the public Edge Add-ons extension ID"
+            exit 1
+          fi
+
+          ARCHIVE="$(find dist -maxdepth 1 -name '*.zip' -type f -print -quit)"
+          if [ -z "$ARCHIVE" ]; then
+            echo "::error::No extension zip found in the downloaded artifact"
+            exit 1
+          fi
+          echo "ARCHIVE=$ARCHIVE" >> "$GITHUB_ENV"
+
+      - name: Upload and publish Microsoft Edge Add-ons extension
+        shell: bash
+        env:
+          API_KEY: ${{ secrets.EDGE_ADDONS_API_KEY }}
+          CLIENT_ID: ${{ vars.EDGE_ADDONS_CLIENT_ID }}
+          PRODUCT_ID: ${{ vars.EDGE_ADDONS_PRODUCT_ID }}
+          RELEASE_VERSION: ${{ needs.resolve.outputs.version }}
+        run: |
+          set -euo pipefail
+          API_ROOT="https://api.addons.microsoftedge.microsoft.com/v1"
+          UPLOAD_HEADERS="$RUNNER_TEMP/edge-upload-headers.txt"
+          UPLOAD_RESPONSE="$RUNNER_TEMP/edge-upload-response.json"
+          STATUS_RESPONSE="$RUNNER_TEMP/edge-status-response.json"
+          PUBLISH_HEADERS="$RUNNER_TEMP/edge-publish-headers.txt"
+          PUBLISH_RESPONSE="$RUNNER_TEMP/edge-publish-response.json"
+
+          operation_id_from_headers() {
+            local headers="$1"
+            local location
+            location="$(awk '
+              tolower($1) == "location:" {
+                sub(/\r$/, "")
+                sub(/^[^:]*:[[:space:]]*/, "")
+                print
+                exit
+              }
+            ' "$headers")"
+            location="${location%%\?*}"
+            printf '%s' "${location##*/}"
+          }
+
+          poll_operation() {
+            local operation_name="$1"
+            local status_url="$2"
+            local status="InProgress"
+            local http_status
+
+            for attempt in $(seq 1 30); do
+              sleep 5
+              http_status="$(curl --silent --show-error \
+                --output "$STATUS_RESPONSE" \
+                --write-out '%{http_code}' \
+                --header "Authorization: ApiKey $API_KEY" \
+                --header "X-ClientID: $CLIENT_ID" \
+                "$status_url")"
+              if [ "$http_status" -lt 200 ] || [ "$http_status" -ge 300 ]; then
+                echo "::error::Microsoft Edge Add-ons $operation_name status check failed with HTTP $http_status"
+                jq . "$STATUS_RESPONSE" || cat "$STATUS_RESPONSE"
+                exit 1
+              fi
+
+              status="$(jq -r '.status // empty' "$STATUS_RESPONSE")"
+              echo "$operation_name status check $attempt/30: $status"
+              case "$status" in
+                Succeeded)
+                  return 0
+                  ;;
+                InProgress)
+                  ;;
+                Failed)
+                  echo "::error::Microsoft Edge Add-ons $operation_name failed"
+                  jq . "$STATUS_RESPONSE"
+                  exit 1
+                  ;;
+                *)
+                  echo "::error::Microsoft Edge Add-ons returned an unknown $operation_name status: $status"
+                  jq . "$STATUS_RESPONSE"
+                  exit 1
+                  ;;
+              esac
+            done
+
+            echo "::error::Microsoft Edge Add-ons $operation_name did not complete after 30 checks"
+            jq . "$STATUS_RESPONSE"
+            exit 1
+          }
+
+          HTTP_STATUS="$(curl --silent --show-error \
+            --output "$UPLOAD_RESPONSE" \
+            --dump-header "$UPLOAD_HEADERS" \
+            --write-out '%{http_code}' \
+            --request POST \
+            --header "Authorization: ApiKey $API_KEY" \
+            --header "X-ClientID: $CLIENT_ID" \
+            --header "Content-Type: application/zip" \
+            --data-binary "@$ARCHIVE" \
+            "$API_ROOT/products/$PRODUCT_ID/submissions/draft/package")"
+          if [ "$HTTP_STATUS" != "202" ]; then
+            echo "::error::Microsoft Edge Add-ons upload failed with HTTP $HTTP_STATUS"
+            jq . "$UPLOAD_RESPONSE" || cat "$UPLOAD_RESPONSE"
+            exit 1
+          fi
+
+          UPLOAD_OPERATION_ID="$(operation_id_from_headers "$UPLOAD_HEADERS")"
+          if [ -z "$UPLOAD_OPERATION_ID" ]; then
+            echo "::error::Microsoft Edge Add-ons upload response has no Location operation ID"
+            cat "$UPLOAD_HEADERS"
+            exit 1
+          fi
+          poll_operation \
+            "package upload" \
+            "$API_ROOT/products/$PRODUCT_ID/submissions/draft/package/operations/$UPLOAD_OPERATION_ID"
+
+          HTTP_STATUS="$(curl --silent --show-error \
+            --output "$PUBLISH_RESPONSE" \
+            --dump-header "$PUBLISH_HEADERS" \
+            --write-out '%{http_code}' \
+            --request POST \
+            --header "Authorization: ApiKey $API_KEY" \
+            --header "X-ClientID: $CLIENT_ID" \
+            --header "Content-Type: application/json" \
+            --data "{\"notes\":\"Automated BrowserSkill extension v${RELEASE_VERSION} release from GitHub Actions.\"}" \
+            "$API_ROOT/products/$PRODUCT_ID/submissions")"
+          if [ "$HTTP_STATUS" != "202" ]; then
+            echo "::error::Microsoft Edge Add-ons publish failed with HTTP $HTTP_STATUS"
+            jq . "$PUBLISH_RESPONSE" || cat "$PUBLISH_RESPONSE"
+            exit 1
+          fi
+
+          PUBLISH_OPERATION_ID="$(operation_id_from_headers "$PUBLISH_HEADERS")"
+          if [ -z "$PUBLISH_OPERATION_ID" ]; then
+            echo "::error::Microsoft Edge Add-ons publish response has no Location operation ID"
+            cat "$PUBLISH_HEADERS"
+            exit 1
+          fi
+          poll_operation \
+            "submission" \
+            "$API_ROOT/products/$PRODUCT_ID/submissions/operations/$PUBLISH_OPERATION_ID"
+
+          echo "Microsoft Edge Add-ons submission created successfully:"
+          jq '{status, message, errorCode, errors}' "$STATUS_RESPONSE"

apps/extension/src/entrypoints/popup/App.test.tsx

@@ -247,6 +247,41 @@ describe("App", () => {
     expect(copyButton.getAttribute("disabled")).not.toBeNull();
     expect(screen.getByText("连接后可用")).toBeTruthy();
   });
+
+  it("treats protocol drift as still connected and prompts an upgrade", async () => {
+    mockUseConnectionState.mockReturnValue({
+      snapshot: {
+        ...baseSnapshot,
+        state: "version_skew",
+        instanceId: "03c3e47f",
+        handshake: {
+          server: "bh",
+          version: mockDaemonVersion,
+          protocol_version: "1.0",
+        },
+      },
+      statusState: "version_skew",
+      setLabel,
+      setConnectionEnabled,
+    });
+
+    render(<App />);
+
+    expect(screen.getByText("已连接")).toBeTruthy();
+    expect(screen.getByText("可升级")).toBeTruthy();
+    expect(screen.queryByText("协议不一致")).toBeNull();
+    expect(screen.queryByText("Action needed")).toBeNull();
+    expect(screen.queryByText("兼容")).toBeNull();
+    const warning = screen.getByText(/协议版本不同，请及时升级/);
+    expect(warning.textContent).toContain("CLI 协议");
+    expect(warning.textContent).toContain("扩展协议");
+
+    openRecordView();
+    const copyButton = screen.getByRole("button", { name: "复制录制指令" });
+    expect(copyButton.getAttribute("disabled")).toBeNull();
+    fireEvent.click(copyButton);
+    expect(navigator.clipboard.writeText).toHaveBeenCalledTimes(1);
+  });
 });
 
 describe("control hints toggle", () => {

apps/extension/src/entrypoints/popup/App.tsx

@@ -79,6 +79,7 @@ export function App() {
   }, [copiedTick]);
 
   const isSkewed = statusState === "version_skew";
+  const connectionLive = statusState === "connected" || isSkewed;
   const daemonVersion = snapshot.handshake?.version ?? "—";
   const daemonProtocol = snapshot.handshake?.protocol_version ?? "—";
   const extensionVersion = snapshot.extensionVersion || "—";
@@ -90,7 +91,7 @@ export function App() {
     setCopiedInstanceId(true);
   };
 
-  const recordReady = statusState === "connected" && Boolean(snapshot.instanceId);
+  const recordReady = connectionLive && Boolean(snapshot.instanceId);
   const recordPurpose = purposeDraft.trim();
   const recordStartUrl = startUrlDraft.trim();
   const recordCommand = snapshot.instanceId
@@ -206,12 +207,12 @@ export function App() {
             </div>
             {isSkewed && (
               <p
-                className="mt-2 text-xs leading-snug text-amber-600 dark:text-amber-400"
+                className="mt-2 text-xs leading-snug text-muted-foreground"
                 data-slot="popup-version-skew-warning"
               >
                 {t("popup.versionSkewWarning", {
                   extensionProtocol: PROTOCOL_VERSION,
-                  daemonProtocol,
+                  cliProtocol: daemonProtocol,
                 })}
               </p>
             )}

crates/bsk-cli/skill/SKILL.md

@@ -29,7 +29,7 @@ Drive the user's **real Chromium browser** (with their logins and cookies) throu
 ## Prerequisites
 
 1. `bsk` on `PATH` (Rust CLI from browser-skill)
-2. browser-skill **extension** loaded in Chromium and connected (popup shows green)
+2. browser-skill **extension** loaded in Chromium and connected (popup shows green, or an upgrade reminder — still connected; continue)
 3. Any `bsk` command auto-starts background services as needed; use `bsk doctor` if anything fails
 
 ## Mandatory workflow
@@ -210,7 +210,7 @@ Both capture from the moment the tab is attached and read a bounded per-tab buff
 
 | Command | Summary |
 |---------|---------|
-| `bsk evaluate <expression>` | Run JS in agent tab (see red lines); JS throw → stderr, **exit 0** |
+| `bsk evaluate <expression>` | Run JS in agent tab (see red lines). JS throw → stderr, **exit 0** (RPC success); use `--json` and check `.ok` to detect JS errors |
 | `bsk wait-for-navigation` | Block until load/DOM idle/etc. (`--wait-until`, `--timeout`) |
 | `bsk wait-ms <duration>` | Sleep (`500ms`, `2s`, `1m`; **no** `--session`) |
 
@@ -294,12 +294,12 @@ bsk record stop [--output trace]   # terminal fallback if the browser panel is u
 
 | Code | Meaning | What to do |
 |------|---------|------------|
-| `0` | Success (including `evaluate` where JS threw but RPC succeeded) | Continue |
+| `0` | Success (including `evaluate` where JS threw but RPC succeeded) | Continue; for `evaluate`, check `--json` `.ok` to distinguish JS success from RPC success |
 | `1` | User error — bad args, unknown session, tab not in Agent Window, stale ref | Fix args; `bsk session list`; re-snapshot |
 | `2` | Protocol / transport — service unreachable, IPC failure | `bsk doctor`; check extension connected; retry the command |
 | `3` | Browser / CDP execution failed | Retry; simplify selector; check tab still open |
 | `4` | Timeout | Increase `--timeout`; try `--wait-until domcontentloaded` |
-| `5` | Version skew (CLI vs extension) | Upgrade/reinstall matching versions |
+| `5` | Unknown RPC method | This command is not implemented in the current build. Continue the task with other commands. If a newly added command is missing, suggest `bsk update`. |
 
 Human errors print `error:` + `hint:` on stderr; `--json` includes `code`, `message`, `hint`, `exit_code`.
 

crates/bsk-cli/src/cli/doctor.rs

@@ -351,10 +351,10 @@ fn check_version_compatible(status: Option<&StatusResult>) -> CheckResult {
     }
 }
 
-/// `bsk doctor` check: every connected browser should use the same
-/// protocol version as the daemon. Minor protocol drift is accepted by
-/// the daemon but flagged here so the user can see who needs updating
-/// (M10.4).
+/// `bsk doctor` check: connected browsers should speak a protocol the
+/// daemon accepts. A different protocol string is still a live
+/// connection — report Ok so agents keep working. The detail is an
+/// upgrade reminder, not a blocker.
 ///
 /// Review M2 (round-1 minor): when no browsers are connected, the
 /// check has nothing to compare against, so it now reports
@@ -396,13 +396,12 @@ fn check_browsers_protocol_compatible(status: Option<&StatusResult>) -> CheckRes
         })
         .collect::<Vec<_>>()
         .join(", ");
-    CheckResult::fail(
+    CheckResult::ok(
         name,
         format!(
-            "{} browser(s) have protocol minor drift from the daemon: {stale}",
+            "{} browser(s) report a different protocol version (still usable — continue, and upgrade soon): {stale}",
             status.version_skew_browsers.len()
         ),
-        "upgrade the browser-skill extension or bsk CLI so both sides use the same protocol version",
     )
 }
 
@@ -581,7 +580,7 @@ mod m2_tests {
     }
 
     #[test]
-    fn browsers_check_reports_fail_when_skew_present() {
+    fn browsers_check_reports_ok_when_skew_present() {
         let status = fake_status(
             vec![BrowserStatusEntry {
                 instance_id: "alpha".into(),
@@ -605,9 +604,17 @@ mod m2_tests {
             }],
         );
         let check = check_browsers_protocol_compatible(Some(&status));
-        assert_eq!(check.status, CheckStatus::Fail);
+        assert_eq!(check.status, CheckStatus::Ok);
         assert!(check.detail.contains("alpha"));
-        assert!(check.hint.is_some());
+        assert!(
+            check
+                .detail
+                .contains("still usable — continue, and upgrade soon")
+        );
+        assert!(
+            check.hint.is_none(),
+            "a protocol-version note must not hint to stop"
+        );
     }
 
     #[test]
@@ -635,7 +642,7 @@ mod m2_tests {
             }],
         );
         let check = check_browsers_protocol_compatible(Some(&status));
-        assert_eq!(check.status, CheckStatus::Fail);
+        assert_eq!(check.status, CheckStatus::Ok);
         assert!(
             check
                 .detail

crates/bsk-cli/src/cli/status.rs

@@ -65,7 +65,7 @@ fn render_human(s: &StatusResult) {
         let bold = "\x1b[1m";
         let reset = "\x1b[0m";
         eprintln!(
-            "{bold}{yellow}warning:{reset} {} browser(s) have protocol version drift from the daemon{reset}",
+            "{bold}{yellow}note:{reset} {} browser(s) report a different protocol version; still usable — continue, and run `bsk update` soon{reset}",
             s.version_skew_browsers.len()
         );
         for skew in &s.version_skew_browsers {
@@ -75,7 +75,7 @@ fn render_human(s: &StatusResult) {
                 skew.label.clone()
             };
             eprintln!(
-                "  {} ({}) — protocol ext {} vs daemon {} (app ext v{}, daemon v{}) — please align protocol versions",
+                "  {} ({}) — protocol ext {} vs CLI {} (app ext v{}, CLI v{}) — still usable; please upgrade",
                 skew.instance_id,
                 label,
                 display_protocol(&skew.client_protocol_version),

packages/dsh-plugin-browserskill/README.md

@@ -5,41 +5,20 @@ npm: [`@wxg-prc-cpg/browser-skill-dsh-plugin`](https://www.npmjs.com/package/@wx
 A [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (dsh) tool plugin that exposes
 [BrowserSkill](https://github.com/Tencent/BrowserSkill) (`bsk`) browser automation to the model.
 
-Each tool maps to one `bsk <cmd> --json` invocation: the plugin spawns the bsk CLI, parses its
+Each action maps to one `bsk <cmd> --json` invocation: the plugin spawns the bsk CLI, parses its
 structured JSON output, and returns a canonical typed value. The bsk daemon, browser, and browser
 extension keep owning the actual browser control — this package is a thin, well-typed bridge.
 
 ## Tools
 
-| Tool | bsk command | Purpose |
+| Tool | Actions | Purpose |
 | --- | --- | --- |
-| `browser_session_start` | `bsk session start` | Open an Agent Window session; optional initial URL, window size, and device emulation preset. Returns the session id and makes it the current session. |
-| `browser_session_stop` | `bsk session stop` | Stop a session (the current one by default) and close its Agent Window. Only plugin-created sessions can be stopped. |
-| `browser_session_list` | — (registry only) | List the sessions this plugin created, marking the current one. Foreign daemon sessions are never shown. |
-| `browser_navigate` | `bsk navigate` | Navigate the active tab, with `waitUntil` / timeout control. |
-| `browser_snapshot` | `bsk snapshot` | Indented aria-tree snapshot with `@eN` refs for interaction tools. |
-| `browser_observe` | `bsk observe` | Semantic VOM observation (read-only) with `@eN` refs. |
-| `browser_click` | `bsk click` | Click a snapshot ref or CSS selector (button / click-count options). |
-| `browser_hover` | `bsk hover` | Hover a ref or selector and allow hover-triggered UI to settle. |
-| `browser_fill` | `bsk fill` | Fill an input / textarea / contenteditable (clears first by default). |
-| `browser_select` | `bsk select` | Select one or more option values on a `<select>`. |
-| `browser_press` | `bsk press` | Dispatch a key or combo, optionally focusing a target first. |
-| `browser_screenshot` | `bsk screenshot` | PNG capture of the tab or a ref-cropped element; inlines the image when the deployment supports image input, otherwise returns a file path. |
-| `browser_emulate` | `bsk emulate` | Apply or clear mobile device emulation on the active tab. |
-| `browser_tab_list` | `bsk tab list` | List user and/or Agent Window tabs visible to an owned session. |
-| `browser_tab_create` | `bsk tab create` | Create a foreground or background tab in the Agent Window. |
-| `browser_tab_select` | `bsk tab select` | Focus an Agent Window tab. |
-| `browser_tab_close` | `bsk tab close` | Close an Agent Window tab. |
-| `browser_tab_borrow` | `bsk tab borrow` | Move a user tab into the Agent Window for controlled interaction. |
-| `browser_tab_return` | `bsk tab return` | Return a borrowed tab to its original user window. |
-| `browser_navigate_back` / `browser_navigate_forward` | `bsk navigate-back` / `bsk navigate-forward` | Traverse tab history and wait for a lifecycle phase. |
-| `browser_reload` | `bsk reload` | Reload the active tab, optionally bypassing cache. |
-| `browser_wait_for_navigation` | `bsk wait-for-navigation` | Wait explicitly for a page lifecycle event. |
-| `browser_request_help` | `bsk request-help` | Pause for login, captcha, OTP, confirmation, or another human browser step. |
-| `browser_get_html` | `bsk get-html` | Read raw document or ref-scoped HTML after semantic observation is insufficient. |
-| `browser_console` | `bsk console` | Read buffered console/log/exception entries without evaluating JavaScript. |
-| `browser_network` | `bsk network` | Read buffered network response/failure metadata. |
-| `browser_window_resize` | `bsk window resize` | Resize the owned session's Agent Window. |
+| `browser_session` | `start`, `stop`, `list` | Manage plugin-owned Agent Window sessions. |
+| `browser_page` | `navigate`, `back`, `forward`, `reload`, `wait` | Navigate the active tab and wait for page lifecycle events. |
+| `browser_inspect` | `observe`, `snapshot`, `html`, `screenshot`, `console`, `network` | Read semantic or diagnostic page state and capture screenshots. |
+| `browser_interact` | `click`, `hover`, `fill`, `select`, `press` | Interact with controls using fresh refs or selectors. |
+| `browser_tabs` | `list`, `create`, `select`, `close`, `borrow`, `return` | Manage Agent Window tabs and temporarily borrow user tabs. |
+| `browser_assist` | `resize`, `emulate`, `request-help` | Resize or emulate the browser and pause for human-only steps. |
 
 `bsk evaluate` and `bsk record` are intentionally not exposed by this plugin: arbitrary page
 evaluation is a higher-risk capability, while recording is long-running and needs a dedicated
@@ -63,7 +42,7 @@ seam degrade silently.
 
 One agent conversation can drive several browser sessions at once:
 
-- `browser_session_start` returns the session id and makes it the **current session**.
+- `browser_session` with `action: start` returns the session id and makes it the **current session**.
 - Every operation tool accepts an optional `session` argument. When omitted, the call acts on the
   current session (the one most recently started or used); when given, that session becomes current.
 - Every tool result echoes the session it actually acted on, so the model never has to guess.
@@ -72,9 +51,9 @@ One agent conversation can drive several browser sessions at once:
 
 **Ownership boundary**: the bsk daemon may be shared with other agents, terminals, or dsh
 instances. The plugin therefore only ever sees and operates on sessions it created itself —
-an explicit `session` argument naming a foreign or unknown id is rejected, `browser_session_list`
-shows plugin-created sessions only (no daemon-wide view), and stop/unload cleanup can never touch
-a session owned by another program.
+an explicit `session` argument naming a foreign or unknown id is rejected, the `list` action on
+`browser_session` shows plugin-created sessions only (no daemon-wide view), and stop/unload cleanup
+can never touch a session owned by another program.
 
 ## Installation
 
@@ -109,7 +88,7 @@ All fields are optional and validated through the plugin's Schemastery `Config`:
         # lazyTools: true              # reveal browser_* tools only after the skill is invoked
 ```
 
-- **`lazyTools` (default `true`)** — the final progressive-disclosure stage: the 28
+- **`lazyTools` (default `true`)** — the final progressive-disclosure stage: the six
   `browser_*` tool schemas stay OUT of the system prompt (zero schema tokens) until the
   `browser-skill` skill is actually invoked — the skill catalog entry is the only
   advertisement. One successful invocation (model tool call, or a `/browser-skill` user
@@ -131,7 +110,7 @@ like the BrowserSkill extension without leaking a single selector into the host
 shell's theme cannot bleed back in.
 
 - **Lifecycle**: hidden while the plugin owns no sessions; appears on the first
-  `browser_session_start`; disappears when all sessions stop (or the plugin unloads).
+  `browser_session` start action; disappears when all sessions stop (or the plugin unloads).
 - **Focus view**: status row (green/idle/red dot + session + action + mm:ss), the latest page
   frame (refreshes every ~1.5s while active, ~8s when idle; the last good frame stays on
   stage while the next one loads, and is kept on errors so the card does not flash),
@@ -173,11 +152,11 @@ shell's theme cannot bleed back in.
   card). Screenshots additionally attach the image itself when the host mounts an attachment store
   and the active model route declares image input; otherwise the PNG path is returned.
 - **Web UI toolview (browser half)**: the package is dual-face. `dsh.client` (platform `web`) ships
-  `lib/client.cjs`, which registers a keyed `tool.call.toolview` view for `browser_screenshot`. The
-  custom view keeps the terminal block (command + output) and, when the settled result carries an
-  image block, resolves the durable attachment through the client session's authorized
+  `lib/client.cjs`, which registers a keyed `tool.call.toolview` view for `browser_inspect`. The
+  custom view keeps a terminal block for every inspect action and, when a screenshot result carries
+  an image block, resolves the durable attachment through the client session's authorized
   `readAttachment` RPC and renders it with the shared `MessageImage` thumbnail/lightbox atoms.
-  Every other `browser_*` tool keeps the stock terminal card. The bundle follows the dsh client
+  The other five tools keep the stock terminal card. The bundle follows the dsh client
   contract: a CJS closure factory handed to `window.__ModuleLoader__.load`, platform modules
   (`react`, `dsh-client-ui-*`) external, everything else inlined, CSS Modules compiled by
   lightningcss.
@@ -198,8 +177,9 @@ pnpm --filter @wxg-prc-cpg/browser-skill-dsh-plugin build    # tsdown -> lib/
 ## Publishing
 
 The GitHub Actions workflow **Release dsh plugin** publishes this package to npm
-as `@wxg-prc-cpg/browser-skill-dsh-plugin`. The Cordis plugin id stays
-`dsh-plugin-browserskill`.
+as `@wxg-prc-cpg/browser-skill-dsh-plugin`. The Cordis plugin id, client
+bundle registration, and `cordis.patch.yml` name are the same specifier so
+dsh can import and materialize the plugin without a ModuleLoader id mismatch.
 
 Trigger it by pushing a tag that matches `package.json`'s `version`:
 

packages/dsh-plugin-browserskill/package.json

@@ -50,6 +50,10 @@
         "@deepseek-ai/dsh-client-runtime",
         "@deepseek-ai/dsh-client-ui-tool",
         "@deepseek-ai/dsh-client-ui-layout"
+      ],
+      "external": [
+        "@deepseek-ai/dsh-client-ui-attachment",
+        "@deepseek-ai/dsh-client-ui-primitives"
       ]
     }
   },

packages/dsh-plugin-browserskill/scripts/build-skill-content.mjs

@@ -18,37 +18,15 @@ function stripFrontmatter(markdown) {
 }
 
 const SUPPORTED_BROWSER_TOOLS = [
-  "browser_session_start",
-  "browser_session_stop",
-  "browser_session_list",
-  "browser_navigate",
-  "browser_navigate_back",
-  "browser_navigate_forward",
-  "browser_reload",
-  "browser_wait_for_navigation",
-  "browser_snapshot",
-  "browser_observe",
-  "browser_get_html",
-  "browser_screenshot",
-  "browser_console",
-  "browser_network",
-  "browser_click",
-  "browser_hover",
-  "browser_fill",
-  "browser_select",
-  "browser_press",
-  "browser_tab_list",
-  "browser_tab_create",
-  "browser_tab_select",
-  "browser_tab_close",
-  "browser_tab_borrow",
-  "browser_tab_return",
-  "browser_request_help",
-  "browser_emulate",
-  "browser_window_resize",
+  "browser_session",
+  "browser_page",
+  "browser_inspect",
+  "browser_interact",
+  "browser_tabs",
+  "browser_assist",
 ];
 
-function validateModelFacingSkill(markdown) {
+function validateModelFacingSkill(markdown, supportedTools) {
   if (/\bbsk\b/i.test(markdown)) {
     throw new Error("DSH skill must not expose the internal CLI name");
   }
@@ -57,9 +35,9 @@ function validateModelFacingSkill(markdown) {
   }
 
   const mentioned = new Set(markdown.match(/\bbrowser_[a-z][a-z_]*\b/g) ?? []);
-  const supported = new Set(SUPPORTED_BROWSER_TOOLS);
+  const supported = new Set(supportedTools);
   const unknown = [...mentioned].filter((name) => !supported.has(name));
-  const missing = SUPPORTED_BROWSER_TOOLS.filter((name) => !mentioned.has(name));
+  const missing = supportedTools.filter((name) => !mentioned.has(name));
   if (unknown.length > 0) {
     throw new Error(`DSH skill mentions unsupported browser tools: ${unknown.join(", ")}`);
   }
@@ -75,11 +53,11 @@ const DESCRIPTION =
   "debug page activity, or smoke-test UI changes in a managed Agent Window.";
 
 const content = `${stripFrontmatter(source).trim()}\n`;
-validateModelFacingSkill(content);
+validateModelFacingSkill(content, SUPPORTED_BROWSER_TOOLS);
 
 const banner =
   "// GENERATED by scripts/build-skill-content.mjs — do not edit.\n" +
-  "// Source: skill/SKILL.md (DeepSeek Harness browser_* tool instructions).\n";
+  "// Source: skill/SKILL.md (DeepSeek Harness six-tool instructions).\n";
 
 writeFileSync(
   join(pkg, "src", "skill-content.generated.ts"),

packages/dsh-plugin-browserskill/skill/SKILL.md

@@ -1,271 +1,146 @@
 ---
 name: browser-skill
-description: |
-  Use the injected browser_* tools to automate the user's logged-in Chromium
-  browser in a managed Agent Window.
+description: Browser automation through six injected domain tools.
 ---
 
 # browser-skill for DeepSeek Harness
 
-Drive the user's real Chromium browser through the structured `browser_*` tools provided by this
+Drive the user's real Chromium browser through the six structured domain tools provided by this
 plugin. The browser keeps the user's existing logins and cookies while isolating automation in an
-Agent Window. Use the tool schemas exactly as exposed by DeepSeek Harness; arguments are structured
-fields, not command-line flags.
+Agent Window. Every call must include the tool's `action` field; other fields depend on that action.
 
 ## Tool availability
 
-Loading this skill reveals the browser tools for the rest of the conversation. If the tools were
-hidden before this skill was loaded, start browser work only after the skill invocation succeeds.
+Loading this skill reveals the browser tools for the rest of the conversation. All browser work
+must use the injected tools directly. Do not invoke another process to control the browser. Direct
+tool calls preserve session ownership, the live observation window, cancellation, attachments, and
+cleanup.
 
-All browser operations must use the injected tools directly. Do not translate the examples in this
-document into terminal commands or invoke another process to control the browser. Direct tool calls
-preserve session ownership, the live observation window, tool cards, cancellation, and cleanup.
+The six tools are:
 
-## When to use
-
-- Visit and read pages that require the user's existing browser session.
-- Fill forms, select options, click through bounded flows, or smoke-test a web UI.
-- Inspect tabs, page structure, screenshots, console messages, or network metadata.
-- Work with a user tab after explicitly borrowing it into the Agent Window.
-
-Do not use browser automation for tasks that need only local files, a purpose-built API, or an
-explanation. Never extract credentials, cookies, tokens, or other secrets. Do not keep control of a
-personal tab longer than the immediate task requires.
+- `browser_session`: start, stop, or list plugin-owned sessions.
+- `browser_page`: navigate, back, forward, reload, or wait for navigation.
+- `browser_inspect`: observe, snapshot, html, screenshot, console, or network.
+- `browser_interact`: click, hover, fill, select, or press.
+- `browser_tabs`: list, create, select, close, borrow, or return tabs.
+- `browser_assist`: resize the Agent Window, emulate a device, or request help from the user.
 
 ## Mandatory workflow
 
 Every browser task follows this lifecycle:
 
-1. Call `browser_session_start` and retain its returned `sessionId`.
-2. Use that session for every browser operation. Pass `session` explicitly when more than one
-   session exists.
-3. Call `browser_session_stop` when the bounded goal is reached or an error ends the task. Treat
-   cleanup like a `finally` path unless the user explicitly asks to keep the session open.
+1. Call `browser_session` with `action: "start"` and retain its returned `sessionId`.
+2. Use that session for every later call. Pass `session` explicitly when more than one exists.
+3. Call `browser_session` with `action: "stop"` when the goal is reached or an error ends the task.
+   Treat cleanup like a finally path unless the user explicitly asks to keep the session open.
 
-Conceptual tool-call sequence:
+Conceptual sequence:
 
 ```text
-browser_session_start({ url: "https://example.com" })
-browser_observe({ session: sessionId })
-browser_click({ session: sessionId, target: "@e4" })
-browser_observe({ session: sessionId })
-browser_session_stop({ session: sessionId })
+browser_session({ action: "start", url: "https://example.com" })
+browser_inspect({ action: "observe", session: sessionId })
+browser_interact({ action: "click", session: sessionId, target: "@e4" })
+browser_inspect({ action: "observe", session: sessionId })
+browser_session({ action: "stop", session: sessionId })
 ```
 
-The plugin owns only sessions created by `browser_session_start`. `browser_session_list` reports
-only those owned sessions. Never guess or reuse a session id from another program. When `session`
-is omitted, most tools use the most recently started or used owned session; explicit ids are safer
-for multi-session work.
+The plugin owns only sessions it starts. Never guess or reuse a session id from another program.
+When `session` is omitted, most actions use the current owned session, but explicit ids are safer.
 
 ## Stop when the goal is met
 
-Browser work is a bounded task, not open-ended browsing.
+Browser work is bounded. Define an observable success condition, take the shortest path, and stop
+as soon as success is visible. Do not refresh or click after success. If an action fails twice
+without progress, pause instead of brute-forcing. If state is ambiguous, make at most one fresh
+observation and then ask the user.
 
-1. Define one observable success condition from the user's request.
-2. Take the shortest path: observe, act, then perform at most one relevant check.
-3. Stop immediately when success is visible. Do not refresh, navigate, click, or perform extra
-   verification after success.
-4. If a human-only step appears, use `browser_request_help`. If an action fails twice without
-   progress, pause instead of brute-forcing.
-5. When uncertain, make at most one fresh observation. If the state is still ambiguous, ask the
-   user rather than continuing to click.
+## Observe and interact
 
-## Core interaction loop
+Use `browser_inspect` with `action: "observe"` as the primary semantic page view. It returns roles,
+states, text, and `@eN` refs. Use `action: "snapshot"` when a static accessibility tree is more useful.
+Prefer a fresh ref over a raw CSS selector for interactions.
 
-Use `browser_navigate` to open a destination in the active Agent Window tab, then inspect the page
-before interacting:
+Refs invalidate after navigation and may become stale after large DOM changes. Observe again after
+navigation or a meaningful DOM update before using another ref. If an observation identifies a
+hover-triggered surface, hover the trigger, observe again, then interact with the revealed control.
 
-1. `browser_observe` is the primary semantic view. It returns roles, states, text, and `@eN` refs,
-   including hints about conditional hover or focus surfaces.
-2. `browser_snapshot` is the static accessibility-tree fallback when the semantic view is
-   insufficient or a stricter tree is useful.
-3. Prefer a fresh `@eN` ref over a raw CSS selector for `browser_click`, `browser_hover`,
-   `browser_fill`, `browser_select`, and targeted `browser_press` calls.
-4. Observe again after navigation or a meaningful DOM change before using another ref.
+Use `browser_interact` as follows:
 
-Refs invalidate after navigation and may also become stale after large DOM changes. Never carry a
-ref to a new page. If an observation marks a control as hover-triggered, call `browser_hover` on the
-trigger, observe again, and then act on the newly visible ref.
+- `action: "click"`: requires `target`; optionally accepts `button` and `clickCount`.
+- `action: "hover"`: requires `target`; optionally accepts `modifiers`, `settleMs`, and `timeoutMs`.
+- `action: "fill"`: requires `target` and `value`; `noClear` appends instead of replacing.
+- `action: "select"`: requires `target` and a non-empty `values` array. Values are option value
+  attributes, not necessarily the visible labels.
+- `action: "press"`: requires `key`; optionally focus `target` first and set `holdMs`.
 
-Use interaction tools according to their schemas:
+## Navigate and wait
 
-- `browser_click`: click a ref or selector; optionally choose mouse button and click count.
-- `browser_hover`: move to a ref or selector and allow hover UI to settle; then observe again.
-- `browser_fill`: clear and fill a text-capable element; set `noClear` only when appending is
-  intentional.
-- `browser_select`: pass `values` as an array, including for a single option.
-- `browser_press`: send a key or combo, optionally focusing `target` first.
+Use `browser_page` with these actions:
 
-Actions may trigger a navigation after the action itself returns. Use
-`browser_wait_for_navigation` only when that lifecycle wait is actually needed, then observe the
-new page.
+- `navigate` requires `url` and optionally accepts `waitUntil` and `timeoutMs`.
+- `back` and `forward` move one history entry and optionally accept lifecycle wait fields.
+- `reload` refreshes the current tab; set `hard: true` only when bypassing cache is necessary.
+- `wait` waits for an expected navigation after an interaction that did not already wait.
 
-## Observation and page reading priority
+Avoid speculative waits when no navigation is expected. After any page change, discard old refs and
+observe again.
+
+## Reading priority
 
 Escalate page reading only as needed:
 
-1. `browser_observe` for normal semantic understanding and interaction refs.
-2. `browser_snapshot` for a stricter static accessibility tree.
-3. `browser_get_html` for hidden DOM, metadata, or exact markup that the semantic views cannot
-   provide. A `ref` must come from a fresh observation; keep `maxBytes` bounded.
-4. `browser_screenshot` for layout, styling, canvas, image content, or another genuinely visual
-   question. Pass a fresh `ref` to crop to one element, or omit it for the visible tab.
+1. `browser_inspect` with `action: "observe"` for normal understanding and refs.
+2. The same tool with `action: "snapshot"` for a stricter static tree.
+3. `action: "html"` for exact markup that semantic views cannot provide; keep `maxBytes` bounded and
+   use only a fresh ref when scoping a subtree.
+4. `action: "screenshot"` for layout, styling, canvas, or genuinely visual evidence. A fresh `ref`
+   crops to an element; omit it for the visible tab.
 
-Do not start with raw HTML or a screenshot merely to discover ordinary controls. Screenshots may
-return an inline image or a saved path depending on the deployment.
+Do not start with raw HTML or a screenshot merely to discover ordinary controls.
 
 ## Tabs and the Agent Window boundary
 
-Write actions normally affect only Agent Window tabs.
+Use `browser_tabs` with these actions:
 
-- `browser_tab_list` lists `agent`, `user`, or `all` tabs visible to the owned session. Use
-  `scope: "user"` to locate a user tab before borrowing it.
-- `browser_tab_create` creates an Agent Window tab. It is active by default; set `active: false`
-  for a background tab.
-- `browser_tab_select` focuses an Agent Window tab.
-- `browser_tab_close` closes an Agent Window tab.
-- `browser_tab_borrow` moves a chosen user tab into the Agent Window before any write action.
-- `browser_tab_return` returns a borrowed tab to its original window and position. Return it as
-  soon as the immediate task is finished; stopping the session is only the fallback cleanup.
+- `list` returns visible tabs; `scope: "user"` finds a user tab before borrowing.
+- `create` makes an Agent Window tab and optionally accepts `url`, `active`, and `index`.
+- `select` focuses an Agent Window tab and requires a returned `tabId`.
+- `close` closes an Agent Window tab and requires a returned `tabId`.
+- `borrow` moves a user tab into the Agent Window and requires a listed user `tabId`.
+- `return` restores a borrowed tab and requires that borrowed `tabId`.
 
-Use only a `tabId` returned by `browser_tab_list` or `browser_tab_create`. Do not assume a tab id
-from page position or title.
+Return borrowed tabs as soon as the immediate task finishes. Stopping the session is only fallback
+cleanup. Never invent a tab id.
 
-## Navigation and lifecycle
+## Human help and display controls
 
-- `browser_navigate` opens a URL and can wait for a requested lifecycle phase.
-- `browser_navigate_back` and `browser_navigate_forward` move one history entry.
-- `browser_reload` reloads the current tab; use `hard: true` only when bypassing cache is necessary.
-- `browser_wait_for_navigation` waits for a lifecycle event after an action that did not already
-  wait. Avoid speculative waits when no navigation is expected.
+Use `browser_assist` with `action: "request-help"` for login, captcha, OTP, payment confirmation,
+consent, or another human-only step. `prompt` is required. Optional `targets` highlight fresh refs or
+selectors. `completionCriteria` can detect explicit stable success through URL, selector, or text
+conditions. After the user continues or criteria complete, observe again before reasoning about the
+new state.
 
-After navigation, history traversal, or reload, discard old refs and call `browser_observe` or
-`browser_snapshot` again.
+Use `action: "resize"` with `width` and `height` to resize the outer Agent Window. Use
+`action: "emulate"` with a device preset or explicit width and height; `mobile: true` also requires
+dimensions. Use `off: true` alone to clear emulation. Emulation is per tab.
 
-## Ask the user for browser help
+## Read-only diagnostics
 
-Call `browser_request_help` for login, captcha, OTP, payment confirmation, consent, or another step
-that requires the user. Give a clear `prompt`; when the prompt refers to a concrete control, pass
-its fresh ref or selector in `targets` so the overlay can highlight it. Use `title` only when it
-adds clarity.
+Use `browser_inspect` with `action: "console"` for buffered logs and exceptions, and
+`action: "network"` for response and failure metadata. Both accept `tabId`, `since`, `limit`, and
+`maxTextChars`; continue with the returned sequence cursor rather than rereading the same entries.
+Console additionally accepts `includeStack`.
 
-Optional `completionCriteria` can finish the pause automatically when the page has an explicit,
-stable success signal. It supports `any` or `all` conditions using `urlContains`, `urlMatches`,
-`selectorExists`, `selectorMissing`, `textExists`, or `textMissing`, plus `stableForMs`.
-
-```text
-browser_request_help({
-  session: sessionId,
-  title: "Sign in required",
-  prompt: "Complete sign-in, then return control after the account page appears.",
-  targets: ["@e7"],
-  timeoutMs: 300000,
-  completionCriteria: {
-    any: [{ urlContains: "/account" }, { selectorExists: "[data-testid='account-menu']" }],
-    stableForMs: 1000
-  }
-})
-```
-
-Interpret outcomes deliberately:
-
-- `continued` or `completed`: the user returned control or the explicit detector matched. Observe
-  again before reasoning about the new state or using refs.
-- `cancelled`: stop the requested flow.
-- `timed_out`: report the timeout; do not repeatedly reopen the prompt.
-- `disabled`: unattended mode does not allow help; proceed only if safe or stop gracefully.
-- `navigated`: legacy outcome; do not treat navigation alone as proof of success.
-
-## Debugging without page scripting
-
-Use the read-only debugging tools only when relevant to the user's task:
-
-- `browser_console` reads buffered console messages, browser logs, and JavaScript exceptions.
-  Set `includeStack: true` when stack frames are useful.
-- `browser_network` reads buffered response and failure metadata. It does not return request or
-  response headers or bodies and does not modify traffic.
-
-Both tools accept `tabId`, `since`, `limit`, and `maxTextChars`. Their result returns `nextSince`;
-pass that value as the next `since` cursor instead of repeatedly requesting the same entries.
-
-Arbitrary page script evaluation and interaction recording are intentionally unsupported in this
-DeepSeek Harness plugin. Do not invent tool names or route around this limitation. If the task
-strictly requires either capability, explain that it is not available in this phase.
-
-## Window and device controls
-
-- `browser_window_resize` changes the Agent Window's outer `width` and `height`; each dimension
-  must be from 100 through 7680 CSS pixels.
-- `browser_emulate` applies a built-in device preset or explicit viewport settings to the active
-  tab. Explicit `width` and `height` must be supplied together. `mobile: true` also requires both
-  dimensions. Use `off: true` alone to clear emulation. Emulation is per-tab and new tabs do not
-  inherit it.
-
-## Tool reference
-
-### Sessions
-
-| Tool | Purpose |
-| --- | --- |
-| `browser_session_start` | Start an owned Agent Window session, optionally with URL, size, browser instance, focus behavior, or device preset. |
-| `browser_session_stop` | Stop an owned session and close its Agent Window. |
-| `browser_session_list` | List owned sessions and identify the current one. |
-
-### Navigation and reading
-
-| Tool | Purpose |
-| --- | --- |
-| `browser_navigate` | Navigate the active tab to a URL. |
-| `browser_navigate_back` | Go back one history entry. |
-| `browser_navigate_forward` | Go forward one history entry. |
-| `browser_reload` | Reload the active tab, optionally bypassing cache. |
-| `browser_wait_for_navigation` | Wait for an expected page lifecycle event. |
-| `browser_observe` | Read the primary semantic page representation with refs. |
-| `browser_snapshot` | Read the static accessibility-tree fallback with refs. |
-| `browser_get_html` | Read raw document or ref-scoped HTML as a last resort. |
-| `browser_screenshot` | Capture the visible tab or a ref-cropped element. |
-
-### Interaction
-
-| Tool | Purpose |
-| --- | --- |
-| `browser_click` | Click a ref or CSS selector. |
-| `browser_hover` | Hover a ref or selector to reveal conditional UI. |
-| `browser_fill` | Fill a text-capable element. |
-| `browser_select` | Select one or more option values. |
-| `browser_press` | Send a keyboard key or combo. |
-
-### Tabs
-
-| Tool | Purpose |
-| --- | --- |
-| `browser_tab_list` | List visible user and Agent Window tabs. |
-| `browser_tab_create` | Create an Agent Window tab. |
-| `browser_tab_select` | Focus an Agent Window tab. |
-| `browser_tab_close` | Close an Agent Window tab. |
-| `browser_tab_borrow` | Move a user tab into the Agent Window. |
-| `browser_tab_return` | Return a borrowed tab to its original window. |
-
-### Assistance, diagnostics, and display
-
-| Tool | Purpose |
-| --- | --- |
-| `browser_request_help` | Pause for an explicit user browser step. |
-| `browser_console` | Read buffered console and exception entries. |
-| `browser_network` | Read buffered response and failure metadata. |
-| `browser_window_resize` | Resize the Agent Window. |
-| `browser_emulate` | Apply or clear per-tab device emulation. |
+Arbitrary page script evaluation and interaction recording are intentionally unsupported. Do not
+invent tool names or route around that limitation.
 
 ## Error handling and final checklist
 
-On a stale ref, observe again and retry once with a fresh ref. On a closed or unknown tab, list
-tabs instead of guessing. On an unknown session, list owned sessions or start a new one; never try
-foreign ids. On timeout, decide whether one longer, purposeful wait is justified. Avoid repeated
-actions with unchanged state.
+On a stale ref, observe again and retry once. On an unknown tab, list tabs instead of guessing. On
+an unknown session, list owned sessions or start one; never try foreign ids. On timeout, decide
+whether one longer purposeful wait is justified, and never immediately repeat a failed history
+action without first checking page state.
 
-Before finishing a browser task, verify that:
-
-- The user's observable goal is met, or the blocker is clearly reported.
-- Any borrowed tab has been returned when practical.
-- The owned session has been stopped unless the user explicitly asked to keep it open.
-- No further page action is taken after success.
+Before finishing, verify that the observable goal is met or the blocker is clear, borrowed tabs are
+returned when practical, the owned session is stopped, and no extra page action occurs after
+success.

packages/dsh-plugin-browserskill/src/archive-cleanup.ts

@@ -7,7 +7,7 @@
  * are hidden from every surface, so their Agent Windows would otherwise
  * linger with no way back to them.
  *
- * Ownership lineage: browser_session_start records the calling agent's
+ * Ownership lineage: browser_session action=start records the calling agent's
  * session id PLUS every ancestor along `header.parentSession` (a subagent's
  * browsers are reaped when any ancestor is archived, the root conversation
  * included). Lineage is resolved through the host session store; an

packages/dsh-plugin-browserskill/src/browser-tools.ts

@@ -0,0 +1,334 @@
+/**
+ * The six model-facing BrowserSkill tools. Each action dispatches to a private
+ * operation handler, so the model receives only six schemas while every action
+ * retains the existing ownership, cancellation, queuing, observation,
+ * screenshot, and presentation behavior.
+ */
+
+import { defineTool, type ParameterSchemaSpec, type ToolDefinition } from "@deepseek-ai/dsh-tools";
+import { SESSION_PARAM, TAB_ID_PARAM, TIMEOUT_MS_PARAM, WAIT_UNTIL_PARAM } from "./tool-params";
+import { createBrowserOperationDefinitions, type ToolDeps } from "./tools";
+
+const DEVICE_PRESETS = [
+  "iphone-14",
+  "iphone-14-pro-max",
+  "iphone-se",
+  "pixel-7",
+  "galaxy-s23",
+  "ipad-mini",
+  "galaxy-tab-s8",
+] as const;
+
+const TARGET_PARAM = {
+  type: "string",
+  description: "Snapshot ref such as @e3, or a CSS selector.",
+} as const;
+
+const DOMAIN_RESULT = { type: "json" } as const;
+
+interface BrowserToolSpec {
+  name: string;
+  description: string;
+  actions: Record<string, string>;
+  parameters: ParameterSchemaSpec;
+}
+
+function operationArgs(args: unknown): Record<string, unknown> {
+  const { action: _action, ...rest } = args as Record<string, unknown>;
+  return rest;
+}
+
+function definitionFor(
+  definitions: ReadonlyMap<string, ToolDefinition>,
+  actions: Record<string, string>,
+  action: unknown,
+): ToolDefinition {
+  const operationName = typeof action === "string" ? actions[action] : undefined;
+  const definition = operationName === undefined ? undefined : definitions.get(operationName);
+  if (definition === undefined) throw new Error(`unsupported browser action: ${String(action)}`);
+  return definition;
+}
+
+function defineBrowserTool(
+  spec: BrowserToolSpec,
+  definitions: ReadonlyMap<string, ToolDefinition>,
+): ToolDefinition {
+  const actionNames = Object.keys(spec.actions);
+  return defineTool({
+    name: spec.name,
+    description: spec.description,
+    parameters: {
+      action: {
+        type: "string",
+        required: true,
+        enum: actionNames,
+        description:
+          "Operation to perform. Other arguments are action-dependent and are validated by that operation.",
+      },
+      ...spec.parameters,
+    },
+    output: {
+      schema: DOMAIN_RESULT,
+      render(args, value) {
+        const definition = definitionFor(definitions, spec.actions, args.action);
+        return definition.output.render(operationArgs(args), value);
+      },
+    },
+    isConcurrencySafe(args) {
+      const definition = definitionFor(definitions, spec.actions, args.action);
+      return definition.isConcurrencySafe?.(operationArgs(args)) === true;
+    },
+    async execute(args, exec) {
+      const definition = definitionFor(definitions, spec.actions, args.action);
+      return (await definition.execute(operationArgs(args), exec)) as never;
+    },
+    presentCall(args) {
+      const definition = definitionFor(definitions, spec.actions, args.action);
+      return definition.presentCall?.(operationArgs(args));
+    },
+    presentResult(args, result) {
+      const definition = definitionFor(definitions, spec.actions, args.action);
+      return definition.presentResult?.(operationArgs(args), result);
+    },
+  });
+}
+
+const BROWSER_TOOL_SPECS: BrowserToolSpec[] = [
+  {
+    name: "browser_session",
+    description:
+      "Manage plugin-owned browser sessions. Actions: start opens an Agent Window; stop closes an " +
+      "owned session; list returns owned sessions. For start, url/device/width/height/noFocus/browser " +
+      "are optional. For stop, session is optional and defaults to the current owned session.",
+    actions: {
+      start: "session.start",
+      stop: "session.stop",
+      list: "session.list",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      url: { type: "string", description: "Initial URL for start." },
+      width: { type: "integer", description: "Agent Window width; start requires height too." },
+      height: { type: "integer", description: "Agent Window height; start requires width too." },
+      noFocus: { type: "boolean", description: "Start the Agent Window in the background." },
+      browser: { type: "string", description: "Browser instance id for start." },
+      device: { type: "string", enum: DEVICE_PRESETS, description: "Device preset for start." },
+    },
+  },
+  {
+    name: "browser_page",
+    description:
+      "Navigate and wait on the active Agent Window tab. Actions: navigate, back, forward, reload, " +
+      "wait. navigate requires url; reload optionally accepts hard; navigation actions accept " +
+      "waitUntil/timeoutMs. Observe again after a meaningful page change before reusing refs.",
+    actions: {
+      navigate: "page.navigate",
+      back: "page.back",
+      forward: "page.forward",
+      reload: "page.reload",
+      wait: "page.wait",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      tabId: TAB_ID_PARAM,
+      url: { type: "string", description: "Destination URL; required for navigate." },
+      waitUntil: WAIT_UNTIL_PARAM,
+      timeoutMs: TIMEOUT_MS_PARAM,
+      hard: { type: "boolean", description: "Bypass cache for reload." },
+    },
+  },
+  {
+    name: "browser_inspect",
+    description:
+      "Read page state without arbitrary script execution. Actions: observe, snapshot, html, " +
+      "screenshot, console, network. Prefer observe, then snapshot, then bounded html; use screenshot " +
+      "for visual evidence. console/network support cursor fields since/limit/maxTextChars.",
+    actions: {
+      observe: "inspect.observe",
+      snapshot: "inspect.snapshot",
+      html: "inspect.html",
+      screenshot: "inspect.screenshot",
+      console: "inspect.console",
+      network: "inspect.network",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      tabId: TAB_ID_PARAM,
+      maxDepth: { type: "integer", description: "Tree depth cap for observe/snapshot." },
+      maxTokens: { type: "integer", description: "Token cap for observe/snapshot." },
+      ref: { type: "string", description: "Fresh ref for scoped html or cropped screenshot." },
+      maxBytes: { type: "integer", description: "HTML byte cap." },
+      since: { type: "integer", description: "Console/network sequence cursor." },
+      limit: { type: "integer", description: "Console/network entry cap." },
+      maxTextChars: { type: "integer", description: "Console/network per-entry text cap." },
+      includeStack: { type: "boolean", description: "Include console stack frames." },
+    },
+  },
+  {
+    name: "browser_interact",
+    description:
+      "Interact with an element in the active Agent Window tab. Actions: click, hover, fill, select, " +
+      "press. click/hover/fill/select require target; fill also requires value; select requires " +
+      "values; press requires key and may optionally focus target first.",
+    actions: {
+      click: "interact.click",
+      hover: "interact.hover",
+      fill: "interact.fill",
+      select: "interact.select",
+      press: "interact.press",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      tabId: TAB_ID_PARAM,
+      target: TARGET_PARAM,
+      button: { type: "string", enum: ["left", "middle", "right"], description: "Click button." },
+      clickCount: { type: "integer", description: "Click count." },
+      value: { type: "string", description: "Text for fill." },
+      noClear: { type: "boolean", description: "Append instead of clearing for fill." },
+      modifiers: {
+        type: "array",
+        items: { type: "string", enum: ["alt", "ctrl", "meta", "shift"] },
+        description: "Modifiers held during hover.",
+      },
+      settleMs: { type: "integer", description: "Hover settle delay." },
+      timeoutMs: TIMEOUT_MS_PARAM,
+      values: {
+        type: "array",
+        items: { type: "string" },
+        description: "Option values for select; at least one is required.",
+      },
+      key: { type: "string", description: "Key or combo for press." },
+      holdMs: { type: "integer", description: "Key hold duration for press." },
+    },
+  },
+  {
+    name: "browser_tabs",
+    description:
+      "Manage tabs visible to an owned session. Actions: list, create, select, close, borrow, return. " +
+      "select/close/borrow/return require tabId from list/create. Borrow moves a user tab into the " +
+      "Agent Window; return it as soon as the task finishes.",
+    actions: {
+      list: "tabs.list",
+      create: "tabs.create",
+      select: "tabs.select",
+      close: "tabs.close",
+      borrow: "tabs.borrow",
+      return: "tabs.return",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      tabId: TAB_ID_PARAM,
+      scope: {
+        type: "string",
+        enum: ["user", "agent", "all"],
+        description: "Tab scope for list.",
+      },
+      url: { type: "string", description: "Initial URL for create." },
+      active: { type: "boolean", description: "Focus the created tab (default true)." },
+      index: { type: "integer", description: "Insertion index for create." },
+    },
+  },
+  {
+    name: "browser_assist",
+    description:
+      "Display and human-assistance operations. Actions: resize, emulate, request-help. resize " +
+      "requires width/height; emulate accepts device or width/height/mobile, or off alone; " +
+      "request-help requires prompt and can wait for explicit completion criteria.",
+    actions: {
+      resize: "assist.resize",
+      emulate: "assist.emulate",
+      "request-help": "assist.request-help",
+    },
+    parameters: {
+      session: SESSION_PARAM,
+      tabId: TAB_ID_PARAM,
+      width: { type: "integer", description: "Window/viewport width." },
+      height: { type: "integer", description: "Window/viewport height." },
+      device: { type: "string", enum: DEVICE_PRESETS, description: "Emulation device preset." },
+      mobile: { type: "boolean", description: "Enable a mobile viewport with width/height." },
+      off: { type: "boolean", description: "Clear emulation; use alone." },
+      prompt: { type: "string", description: "Instructions shown to the user for request-help." },
+      title: { type: "string", description: "Optional request-help title." },
+      targets: {
+        type: "array",
+        items: { type: "string" },
+        description: "Refs/selectors highlighted for request-help.",
+      },
+      timeoutMs: TIMEOUT_MS_PARAM,
+      completionCriteria: {
+        type: "object",
+        additionalProperties: false,
+        description: "Automatic completion detector for request-help.",
+        properties: {
+          any: {
+            type: "array",
+            items: {
+              type: "object",
+              additionalProperties: false,
+              properties: {
+                urlContains: { type: "string" },
+                urlMatches: { type: "string" },
+                selectorExists: { type: "string" },
+                selectorMissing: { type: "string" },
+                textExists: { type: "string" },
+                textMissing: { type: "string" },
+              },
+            },
+          },
+          all: {
+            type: "array",
+            items: {
+              type: "object",
+              additionalProperties: false,
+              properties: {
+                urlContains: { type: "string" },
+                urlMatches: { type: "string" },
+                selectorExists: { type: "string" },
+                selectorMissing: { type: "string" },
+                textExists: { type: "string" },
+                textMissing: { type: "string" },
+              },
+            },
+          },
+          stableForMs: { type: "integer" },
+        },
+      },
+    },
+  },
+];
+
+function indexOperations(definitions: ToolDefinition[]): ReadonlyMap<string, ToolDefinition> {
+  const indexed = new Map<string, ToolDefinition>();
+  for (const definition of definitions) {
+    if (indexed.has(definition.name)) {
+      throw new Error(`duplicate browser operation definition: ${definition.name}`);
+    }
+    indexed.set(definition.name, definition);
+  }
+
+  const routed = new Set(BROWSER_TOOL_SPECS.flatMap((spec) => Object.values(spec.actions)));
+  const missing = [...routed].filter((name) => !indexed.has(name));
+  const unreachable = [...indexed.keys()].filter((name) => !routed.has(name));
+  if (missing.length > 0 || unreachable.length > 0) {
+    throw new Error(
+      [
+        missing.length > 0 ? `missing operations: ${missing.join(", ")}` : undefined,
+        unreachable.length > 0 ? `unreachable operations: ${unreachable.join(", ")}` : undefined,
+      ]
+        .filter(Boolean)
+        .join("; "),
+    );
+  }
+  return indexed;
+}
+
+/** Register the complete six-tool browser suite; returns its combined disposer. */
+export function registerBrowserTools(deps: ToolDeps): () => void {
+  const definitions = indexOperations(createBrowserOperationDefinitions(deps));
+  const disposers = BROWSER_TOOL_SPECS.map((spec) =>
+    deps.ctx.tools.register(defineBrowserTool(spec, definitions)),
+  ).filter((dispose): dispose is () => void => typeof dispose === "function");
+  return () => {
+    for (const dispose of disposers.splice(0)) dispose();
+  };
+}

packages/dsh-plugin-browserskill/src/client/BrowserInspectToolView.module.css

@@ -1,4 +1,4 @@
-/* browser_screenshot toolview: disclosure row over a terminal block plus the
+/* browser_inspect screenshot toolview: disclosure row over a terminal block plus the
    screenshot image. Semantic tokens only (--dsw-alias-*), no literal colors. */
 
 .card {

packages/dsh-plugin-browserskill/src/client/BrowserInspectToolView.tsx

@@ -1,8 +1,6 @@
-// browser_screenshot keyed toolview: the terminal card's command line and
-// output stay exactly as the host renders them, and a result carrying an
-// image block additionally renders the screenshot itself (thumbnail with the
-// shared click-to-open lightbox). Pure function of the frozen call/result
-// slice — live and replay paths render identically.
+// browser_inspect keyed toolview: all inspect actions retain a terminal card;
+// screenshot results additionally render the image attachment. Pure function
+// of the frozen call/result slice, so live and replay paths stay identical.
 
 import type { ImageAttachmentRef } from "@deepseek-ai/dsh-attachment";
 import {
@@ -19,13 +17,13 @@ import {
 } from "@deepseek-ai/dsh-client-ui-primitives";
 import type { ToolCallViewProps } from "@deepseek-ai/dsh-client-ui-tool/client";
 import { useState } from "react";
-import css from "./ScreenshotToolView.module.css";
+import css from "./BrowserInspectToolView.module.css";
 
 /** Resolved loader bound to the owning session at registration time. */
-export type ScreenshotImageLoader = (attachment: ImageAttachmentRef) => Promise<string>;
+export type BrowserInspectImageLoader = (attachment: ImageAttachmentRef) => Promise<string>;
 
-export type ScreenshotToolViewProps = ToolCallViewProps & {
-  loadImage: ScreenshotImageLoader;
+export type BrowserInspectToolViewProps = ToolCallViewProps & {
+  loadImage: BrowserInspectImageLoader;
 };
 
 type ViewState = "running" | "ok" | "error";
@@ -36,6 +34,7 @@ interface ViewModel {
   readonly output: string | null;
   readonly image: ImageAttachmentRef | null;
   readonly summary: string;
+  readonly title: string;
 }
 
 const TERMINAL_LABELS: Partial<TerminalBlockLabels> = {
@@ -66,18 +65,71 @@ function firstLine(text: string): string {
 }
 
 /** Rebuild the command line from the logged arguments (mirrors the host presenter). */
-function commandOf(argsRaw: string, callId: string): string {
+const INSPECT_COMMANDS: Record<string, string> = {
+  observe: "observe",
+  snapshot: "snapshot",
+  html: "get-html",
+  screenshot: "screenshot",
+  console: "console",
+  network: "network",
+};
+
+function titleOf(action: string): string {
+  return action === "html" ? "HTML" : `${action.slice(0, 1).toUpperCase()}${action.slice(1)}`;
+}
+
+interface InspectArgs {
+  action?: unknown;
+  session?: unknown;
+  tabId?: unknown;
+  maxDepth?: unknown;
+  maxTokens?: unknown;
+  ref?: unknown;
+  maxBytes?: unknown;
+  since?: unknown;
+  limit?: unknown;
+  maxTextChars?: unknown;
+  includeStack?: unknown;
+}
+
+function appendNumber(parts: string[], flag: string, value: unknown): void {
+  if (typeof value === "number" && Number.isFinite(value)) parts.push(flag, String(value));
+}
+
+function commandOf(argsRaw: string, callId: string): { command: string; title: string } {
   try {
-    const args = JSON.parse(argsRaw) as { session?: unknown; ref?: unknown };
-    const parts = ["bsk", "screenshot", "--session"];
+    const args = JSON.parse(argsRaw) as InspectArgs;
+    const action = typeof args.action === "string" ? args.action : "inspect";
+    const command = INSPECT_COMMANDS[action] ?? action;
+    const parts = ["bsk", command, "--session"];
     parts.push(
       typeof args.session === "string" && args.session !== "" ? args.session : "(current)",
     );
-    if (typeof args.ref === "string" && args.ref !== "") parts.push("--ref", args.ref);
-    return parts.join(" ");
+    if (action === "observe" || action === "snapshot") {
+      appendNumber(parts, "--max-depth", args.maxDepth);
+      appendNumber(parts, "--max-tokens", args.maxTokens);
+    }
+    if (action === "html" || action === "console" || action === "network") {
+      appendNumber(parts, "--tab-id", args.tabId);
+    }
+    if ((action === "html" || action === "screenshot") && typeof args.ref === "string") {
+      if (args.ref !== "") parts.push("--ref", args.ref);
+    }
+    if (action === "html") appendNumber(parts, "--max-bytes", args.maxBytes);
+    if (action === "console" || action === "network") {
+      appendNumber(parts, "--since", args.since);
+      appendNumber(parts, "--limit", args.limit);
+      appendNumber(parts, "--max-text-chars", args.maxTextChars);
+    }
+    if (action === "console" && args.includeStack === true) parts.push("--include-stack");
+    return { command: parts.join(" "), title: titleOf(action) };
   } catch {
     // Streaming can expose a truncated JSON prefix; it is still the best label.
-    return argsRaw === "" ? `bsk screenshot (${callId})` : `bsk screenshot ${firstLine(argsRaw)}`;
+    return {
+      command:
+        argsRaw === "" ? `browser_inspect (${callId})` : `browser_inspect ${firstLine(argsRaw)}`,
+      title: "Inspect",
+    };
   }
 }
 
@@ -85,9 +137,9 @@ function commandOf(argsRaw: string, callId: string): string {
 export function viewModelOf(block: ToolCallViewProps["block"]): ViewModel {
   const settled = "kind" in block;
   const argsRaw = (settled ? block.call?.argsRaw : block.argsRaw) ?? "";
-  const command = commandOf(argsRaw, block.callId);
+  const { command, title } = commandOf(argsRaw, block.callId);
   if (!settled) {
-    return { state: "running", command, output: null, image: null, summary: command };
+    return { state: "running", command, output: null, image: null, summary: command, title };
   }
   const textBlock = block.content.find((item) => item.type === "text");
   const imageBlock = block.content.find((item) => item.type === "image");
@@ -96,7 +148,7 @@ export function viewModelOf(block: ToolCallViewProps["block"]): ViewModel {
     imageBlock !== undefined && imageBlock.type === "image" ? imageBlock.attachment : null;
   const state: ViewState = block.isError ? "error" : "ok";
   const summary = output !== null ? firstLine(output) : command;
-  return { state, command, output, image, summary };
+  return { state, command, output, image, summary, title };
 }
 
 function dotState(state: ViewState): StateDotState {
@@ -111,19 +163,24 @@ function dotState(state: ViewState): StateDotState {
 }
 
 /**
- * Render one browser_screenshot call: a disclosure row over a terminal block
- * (command + output) plus the screenshot image when the result carries one.
+ * Render one browser_inspect call: a disclosure row over a terminal block,
+ * plus the screenshot image when that action returns an image attachment.
  */
-export function ScreenshotToolView({ block, cwd, inspect, loadImage }: ScreenshotToolViewProps) {
+export function BrowserInspectToolView({
+  block,
+  cwd,
+  inspect,
+  loadImage,
+}: BrowserInspectToolViewProps) {
   const model = viewModelOf(block);
   const [expanded, setExpanded] = useState(false);
   const expandable = model.state === "running" || model.output !== null || model.image !== null;
   const open = expanded && expandable;
   return (
-    <div className={css.card} data-tool="browser_screenshot" data-state={model.state}>
+    <div className={css.card} data-tool="browser_inspect" data-state={model.state}>
       <DisclosureRow
         icon={<StateDot state={dotState(model.state)} />}
-        title="Screenshot"
+        title={model.title}
         open={open}
         expandable={expandable}
         onToggle={() => setExpanded((value) => !value)}

packages/dsh-plugin-browserskill/src/client/index.ts

@@ -1,6 +1,6 @@
 /**
- * dsh-plugin-browserskill browser half: the `browser_screenshot` keyed
- * toolview plus the observation overlay (live thumbnails + interrupt). The
+ * dsh-plugin-browserskill browser half: the `browser_inspect` keyed toolview
+ * plus the observation overlay (live thumbnails + interrupt). The
  * default carrier is a floating card on the `shell.overlay` seat; when the
  * dsh-better-sidebar plugin provides its `betterSidebar` service, the view
  * moves into a sidebar tab instead (see observation-sidebar.tsx).
@@ -16,10 +16,10 @@ import { createElement } from "react";
 // <style> tags; selectors stay unhashed so `cn(..., "bsk-obs")` roots match).
 import "./bsk-tokens.nomodule.css";
 import "./bsk-ui.nomodule.css";
+import { BrowserInspectToolView } from "./BrowserInspectToolView";
 import { ObservationOverlay } from "./ObservationOverlay";
 import { type BetterSidebarLike, registerObservationSidebar } from "./observation-sidebar";
 import { type EventSourceLike, ObservationClientStore } from "./observation-store";
-import { ScreenshotToolView } from "./ScreenshotToolView";
 
 /** Required services: slots, session-scoped attachment reads, and the overlay seat. */
 export const inject = ["slots", "sessions"];
@@ -65,9 +65,9 @@ export function apply(ctx: ClientContext): void {
   const sessions = ctx.get("sessions") as unknown as ISessions;
   ctx.slots.inject("tool.call.toolview", () =>
     ctx.slots.register(
-      { name: "tool.call.toolview", key: "browser_screenshot" },
+      { name: "tool.call.toolview", key: "browser_inspect" },
       (props: ToolCallViewProps) =>
-        createElement(ScreenshotToolView, {
+        createElement(BrowserInspectToolView, {
           ...props,
           loadImage: (attachment: ImageAttachmentRef) =>
             loadSessionImage(sessions, props.sessionId, attachment),

packages/dsh-plugin-browserskill/src/image.ts

@@ -1,10 +1,11 @@
 /**
- * Screenshot delivery. The canonical tool value stays plain JSON (a PNG file
- * path plus pixel metadata); when the host mounts a durable attachment store
- * AND the calling route declares image input, the PNG bytes are additionally
- * committed through `ctx.attachments` so the render step can attach the image
- * itself to the tool result. Any uncertainty (no store, unknown route,
- * text-only model) falls back to the path-only form.
+ * Screenshot delivery. The canonical tool value stays plain JSON (a screenshot
+ * file path plus pixel metadata); when the host mounts a durable attachment
+ * store AND the calling route declares image input, the capture bytes are
+ * additionally committed through `ctx.attachments` so the render step can
+ * attach the image itself to the tool result. Any uncertainty (no store,
+ * unknown route, text-only model, unrecognized bytes) falls back to the
+ * path-only form.
  */
 
 import type { Context } from "@deepseek-ai/cordis";
@@ -33,6 +34,54 @@ interface LlmLike {
   ): Promise<{ inputModalities?: readonly string[] }>;
 }
 
+/**
+ * Sniff the image media type from magic bytes. The capture path can hand back
+ * JPEG (e.g. `chrome.tabs.captureVisibleTab` on some Chromium/Edge builds
+ * returning JPEG regardless of the requested format), so the declared type can
+ * never be trusted from the file extension alone.
+ * @param data - first bytes of the capture.
+ * @returns the detected media type, or undefined when unrecognized.
+ */
+function sniffImageMediaType(data: Uint8Array): string | undefined {
+  if (data.length >= 4) {
+    // PNG signature: 89 50 4E 47 (0D 0A 1A 0A)
+    if (data[0] === 0x89 && data[1] === 0x50 && data[2] === 0x4e && data[3] === 0x47) {
+      return "image/png";
+    }
+    // JPEG SOI: FF D8 FF
+    if (data[0] === 0xff && data[1] === 0xd8 && data[2] === 0xff) {
+      return "image/jpeg";
+    }
+    // WebP: "RIFF" (52 49 46 46) + "WEBP" at offset 8
+    if (
+      data[0] === 0x52 &&
+      data[1] === 0x49 &&
+      data[2] === 0x46 &&
+      data[3] === 0x46 &&
+      data.length >= 12 &&
+      data[8] === 0x57 &&
+      data[9] === 0x45 &&
+      data[10] === 0x42 &&
+      data[11] === 0x50
+    ) {
+      return "image/webp";
+    }
+    // GIF: "GIF87a" or "GIF89a"
+    if (
+      data[0] === 0x47 &&
+      data[1] === 0x49 &&
+      data[2] === 0x46 &&
+      data[3] === 0x38 &&
+      data.length >= 6 &&
+      (data[4] === 0x37 || data[4] === 0x39) &&
+      data[5] === 0x61
+    ) {
+      return "image/gif";
+    }
+  }
+  return undefined;
+}
+
 /**
  * Commit screenshot bytes to the host attachment store when the composition
  * supports durable images on the current model route.
@@ -46,7 +95,14 @@ export async function trySaveScreenshot(
 ): Promise<ImageAttachmentRef | undefined> {
   const attachments = ctx.get("attachments") as AttachmentLike | undefined;
   if (attachments === undefined) return undefined;
-  if (!attachments.imageLimits.mediaTypes.includes("image/png")) return undefined;
+  // Declare the real media type from the bytes, never a hard-coded "image/png".
+  // Declaring the wrong type trips the store's IMAGE_TYPE_MISMATCH check and
+  // silently degrades to path-only; sniffing first lets JPEG captures (returned
+  // by some Chromium builds for `captureVisibleTab({ format: "png" })`) still
+  // inline correctly.
+  const mediaType = sniffImageMediaType(data);
+  if (mediaType === undefined) return undefined;
+  if (!attachments.imageLimits.mediaTypes.includes(mediaType)) return undefined;
   if (
     data.byteLength >
     Math.min(attachments.imageLimits.maxImageBytes, attachments.imageLimits.maxMessageImageBytes)
@@ -55,9 +111,9 @@ export async function trySaveScreenshot(
   }
   if (!(await isImageCapableRoute(ctx, exec))) return undefined;
   try {
-    return await attachments.saveImage({ data, mediaType: "image/png", name });
+    return await attachments.saveImage({ data, mediaType, name });
   } catch {
-    // A store hiccup must not fail the tool call; the PNG path still works.
+    // A store hiccup must not fail the tool call; the path-only form still works.
     return undefined;
   }
 }

packages/dsh-plugin-browserskill/src/index.ts

@@ -13,6 +13,7 @@
 import type { Context } from "@deepseek-ai/cordis";
 import Schema from "@deepseek-ai/schemastery";
 import { armArchiveCleanup } from "./archive-cleanup";
+import { registerBrowserTools } from "./browser-tools";
 import { armLazyTools } from "./lazy-tools";
 import { ObservationService } from "./observation";
 import { registerObservationRoutes } from "./observation-http";
@@ -20,9 +21,9 @@ import { KeyedExecutor } from "./queue";
 import { type BskRunner, createBskRunner } from "./runner";
 import { SessionRegistry } from "./sessions";
 import { armAgentScopedBskSkill, registerBskSkill } from "./skill";
-import { type PluginConfig, registerTools } from "./tools";
+import type { PluginConfig } from "./tools";
 
-export const name = "dsh-plugin-browserskill";
+export const name = "@wxg-prc-cpg/browser-skill-dsh-plugin";
 export const inject = ["tools"];
 
 /** Runtime configuration schema (validated and defaulted by Cordis). */
@@ -100,11 +101,9 @@ export function apply(
   // exact agent context at startup so DSH always loads the browser_* protocol
   // instructions; the shared CLI skill remains untouched for other agents.
   const disarmAgentSkill = armAgentScopedBskSkill(ctx);
-  const removeSuite = resolved.lazyTools
-    ? armLazyTools(ctx, () =>
-        registerTools({ ctx, runner, registry, config: resolved, observation, queue }),
-      )
-    : registerTools({ ctx, runner, registry, config: resolved, observation, queue });
+  const registerSuite = () =>
+    registerBrowserTools({ ctx, runner, registry, config: resolved, observation, queue });
+  const removeSuite = resolved.lazyTools ? armLazyTools(ctx, registerSuite) : registerSuite();
   // Route registration rides ctx.inject: the webServer service may be provided
   // AFTER this plugin loads, and in headless compositions it never appears (the
   // callback simply never runs, leaving the rest of the plugin unaffected).
@@ -133,7 +132,7 @@ export function apply(
   );
 
   // Unload cleanup: kill in-flight children, then stop every session this
-  // plugin OWNS (created via browser_session_start). Referenced or unknown
+  // plugin OWNS (created via browser_session action=start). Referenced or unknown
   // sessions belonging to other programs on the shared daemon are never
   // touched; per-stop failures (already stopped externally, daemon restart)
   // are swallowed so one stale handle cannot abort the rest.
@@ -154,6 +153,7 @@ export function apply(
 }
 
 export { armArchiveCleanup, ownerSessionIds } from "./archive-cleanup";
+export { registerBrowserTools } from "./browser-tools";
 export type { ObservationEvent, ObservationOptions, SessionObservation } from "./observation";
 export { ObservationService } from "./observation";
 export { registerObservationRoutes } from "./observation-http";
@@ -162,4 +162,3 @@ export type { BskRunner, BskRunResult, SpawnImpl } from "./runner";
 export { BskError, createBskRunner } from "./runner";
 export { SessionRegistry } from "./sessions";
 export type { PluginConfig, ToolDeps } from "./tools";
-export { registerTools } from "./tools";

packages/dsh-plugin-browserskill/src/lazy-tools.ts

@@ -1,6 +1,6 @@
 /**
  * Lazy tool-schema injection ("progressive disclosure", final stage): with
- * `lazyTools` on (the default), the 28 browser_* tool schemas stay OUT of the
+ * `lazyTools` on (the default), the six browser tool schemas stay OUT of the
  * system prompt until the `browser-skill` skill has actually been invoked —
  * the skill catalog entry is the only advertisement. One successful
  * invocation (model tool call, or a user's `/browser-skill` gesture) reveals
@@ -96,7 +96,7 @@ export function hasSuccessfulSkillInvocation(events: readonly SessionEventLike[]
 /**
  * Arm the lazy reveal. Returns a disposer tearing down listeners and — when
  * the reveal already happened — the tool suite itself.
- * @param registerSuite - registers the 28 browser_* tools, returns their disposer.
+ * @param registerSuite - registers the six browser tools and returns their disposer.
  */
 export function armLazyTools(ctx: Context, registerSuite: () => () => void): () => void {
   let suiteDisposer: (() => void) | undefined;

packages/dsh-plugin-browserskill/src/observation.ts

@@ -161,7 +161,7 @@ export class ObservationService {
     this.emit({ type: "upsert", session: { ...entry } });
   }
 
-  /** Register a fresh owned session (called from browser_session_start). */
+  /** Register a fresh owned session (called from browser_session action=start). */
   addSession(sessionId: string, url?: string): void {
     if (!this.deps.options.enabled || this.disposed) return;
     const dshSessionIds = this.deps.registry.dshOwnersOf(sessionId);
@@ -176,7 +176,7 @@ export class ObservationService {
     this.scheduleCapture(sessionId, 0);
   }
 
-  /** Drop a session (called from browser_session_stop). */
+  /** Drop a session (called from browser_session action=stop). */
   removeSession(sessionId: string): void {
     this.cancelCapture(sessionId);
     this.foregroundDepth.delete(sessionId);
@@ -276,7 +276,7 @@ export class ObservationService {
 
   /**
    * Stop one owned session and close its Agent Window (the overlay's stop
-   * button — same end state as `browser_session_stop`). Never waits behind a
+   * button — same end state as `browser_session` action=stop). Never waits behind a
    * hung in-flight command: tool children are killed first so the session's
    * keyed queue drains immediately, and no further captures queue up. A
    * session the daemon already forgot stops idempotently — the goal state

packages/dsh-plugin-browserskill/src/phase-one-runtime.ts

@@ -1,7 +1,7 @@
 import type { ToolDefinition, ToolResult, ToolRunContext } from "@deepseek-ai/dsh-tools";
 import type { ToolDeps } from "./tools";
 
-export type ToolRegistrar = (definition: ToolDefinition) => () => void;
+export type ToolRegistrar = (definition: ToolDefinition) => void;
 
 type TerminalPresentation = { card: "terminal"; output: string; exitCode: number } | undefined;
 

packages/dsh-plugin-browserskill/src/phase-one-tools-interaction.ts

@@ -23,10 +23,10 @@ export function registerPhaseOneInteractionTools(
 
   register(
     defineTool({
-      name: "browser_hover",
+      name: "interact.hover",
       description:
         "Move the mouse over a snapshot ref or CSS selector to reveal hover-triggered UI. " +
-        "Run browser_observe or browser_snapshot again afterwards to discover newly visible refs.",
+        "Run browser_inspect action=observe or snapshot afterwards to discover newly visible refs.",
       parameters: {
         target: {
           type: "string",
@@ -68,7 +68,7 @@ export function registerPhaseOneInteractionTools(
         requireNonEmpty(args.target, "target");
         requirePositive(args.timeoutMs, "timeoutMs");
         requirePositive(args.settleMs, "settleMs");
-        const sessionId = registry.resolve(args.session, "browser_hover");
+        const sessionId = registry.resolve(args.session, "browser_interact(action=hover)");
         const cmdArgs = ["hover", "--session", sessionId];
         appendTabId(cmdArgs, args.tabId);
         if (args.modifiers !== undefined && args.modifiers.length > 0) {
@@ -102,7 +102,7 @@ export function registerPhaseOneInteractionTools(
 
   register(
     defineTool({
-      name: "browser_select",
+      name: "interact.select",
       description:
         "Set a select element's option values. Pass one value for a normal select or multiple " +
         "values for a multi-select; values replace the current selection.",
@@ -155,7 +155,7 @@ export function registerPhaseOneInteractionTools(
         requireNonEmpty(args.target, "target");
         requirePositive(args.timeoutMs, "timeoutMs");
         if (args.values.length === 0) throw new Error("values must contain at least one option");
-        const sessionId = registry.resolve(args.session, "browser_select");
+        const sessionId = registry.resolve(args.session, "browser_interact(action=select)");
         const cmdArgs = ["select", "--session", sessionId];
         appendTabId(cmdArgs, args.tabId);
         for (const value of args.values) cmdArgs.push("--value", value);

packages/dsh-plugin-browserskill/src/phase-one-tools-navigation.ts

@@ -51,7 +51,7 @@ export function registerPhaseOneNavigationTools(
   const { registry } = deps;
 
   const registerHistory = (direction: "back" | "forward") => {
-    const toolName = `browser_navigate_${direction}` as const;
+    const toolName = `page.${direction}` as const;
     const command = `navigate-${direction}`;
     register(
       defineTool({
@@ -106,7 +106,7 @@ export function registerPhaseOneNavigationTools(
 
   register(
     defineTool({
-      name: "browser_reload",
+      name: "page.reload",
       description:
         "Reload the active Agent Window tab and wait for a lifecycle phase. Set hard=true to " +
         "bypass the HTTP cache.",
@@ -132,7 +132,7 @@ export function registerPhaseOneNavigationTools(
       },
       async execute(args, exec) {
         requirePositive(args.timeoutMs, "timeoutMs");
-        const sessionId = registry.resolve(args.session, "browser_reload");
+        const sessionId = registry.resolve(args.session, "browser_page(action=reload)");
         const cmdArgs = ["reload", "--session", sessionId];
         appendTabId(cmdArgs, args.tabId);
         appendWaitOptions(cmdArgs, args.waitUntil, args.timeoutMs);
@@ -163,7 +163,7 @@ export function registerPhaseOneNavigationTools(
 
   register(
     defineTool({
-      name: "browser_wait_for_navigation",
+      name: "page.wait",
       description:
         "Wait for a page lifecycle event on the active Agent Window tab. Use after an action " +
         "that may navigate when the action result itself does not wait for navigation.",
@@ -195,7 +195,7 @@ export function registerPhaseOneNavigationTools(
       },
       async execute(args, exec) {
         requirePositive(args.timeoutMs, "timeoutMs");
-        const sessionId = registry.resolve(args.session, "browser_wait_for_navigation");
+        const sessionId = registry.resolve(args.session, "browser_page(action=wait)");
         const timeoutMs = args.timeoutMs ?? 30_000;
         const cmdArgs = ["wait-for-navigation", "--session", sessionId];
         appendTabId(cmdArgs, args.tabId);

packages/dsh-plugin-browserskill/src/phase-one-tools-support.ts

@@ -143,7 +143,7 @@ export function registerPhaseOneSupportTools(
 
   register(
     defineTool({
-      name: "browser_request_help",
+      name: "assist.request-help",
       description:
         "Pause browser automation and ask the user to complete an in-page step such as login, " +
         "captcha, OTP, or confirmation. After continued/completed, observe again before using refs.",
@@ -233,7 +233,7 @@ export function registerPhaseOneSupportTools(
         if (criteria?.stableForMs !== undefined && criteria.stableForMs < 0) {
           throw new Error("completionCriteria.stableForMs must be zero or greater");
         }
-        const sessionId = registry.resolve(args.session, "browser_request_help");
+        const sessionId = registry.resolve(args.session, "browser_assist(action=request-help)");
         const timeoutMs = args.timeoutMs ?? 300_000;
         const cmdArgs = [
           "request-help",
@@ -285,10 +285,10 @@ export function registerPhaseOneSupportTools(
 
   register(
     defineTool({
-      name: "browser_get_html",
+      name: "inspect.html",
       description:
         "Read raw DOM HTML from the active Agent Window tab, optionally scoped to a fresh " +
-        "snapshot ref. Use only when browser_observe/browser_snapshot cannot answer the question.",
+        "snapshot ref. Use only when browser_inspect observe/snapshot cannot answer the question.",
       parameters: {
         session: SESSION_PARAM,
         tabId: TAB_ID_PARAM,
@@ -326,7 +326,7 @@ export function registerPhaseOneSupportTools(
           if (!isSnapshotRef(args.ref)) throw new Error("ref must be a snapshot ref such as @e3");
         }
         requirePositive(args.maxBytes, "maxBytes");
-        const sessionId = registry.resolve(args.session, "browser_get_html");
+        const sessionId = registry.resolve(args.session, "browser_inspect(action=html)");
         const cmdArgs = ["get-html", "--session", sessionId];
         appendTabId(cmdArgs, args.tabId);
         if (args.ref !== undefined) cmdArgs.push("--ref", args.ref);
@@ -361,7 +361,7 @@ export function registerPhaseOneSupportTools(
 
   register(
     defineTool({
-      name: "browser_console",
+      name: "inspect.console",
       description:
         "Read buffered console messages, browser log entries, and JavaScript exceptions from a tab. " +
         "This is read-only and does not evaluate JavaScript.",
@@ -448,7 +448,7 @@ export function registerPhaseOneSupportTools(
       isConcurrencySafe: () => true,
       async execute(args, exec) {
         validateDebugArgs(args);
-        const sessionId = registry.resolve(args.session, "browser_console");
+        const sessionId = registry.resolve(args.session, "browser_inspect(action=console)");
         const cmdArgs = ["console", "--session", sessionId];
         appendDebugOptions(cmdArgs, args);
         if (args.includeStack === true) cmdArgs.push("--include-stack");
@@ -493,7 +493,7 @@ export function registerPhaseOneSupportTools(
 
   register(
     defineTool({
-      name: "browser_network",
+      name: "inspect.network",
       description:
         "Read buffered network responses and failures from a tab. Returns metadata only; request " +
         "and response headers and bodies are not captured.",
@@ -552,7 +552,7 @@ export function registerPhaseOneSupportTools(
       isConcurrencySafe: () => true,
       async execute(args, exec) {
         validateDebugArgs(args);
-        const sessionId = registry.resolve(args.session, "browser_network");
+        const sessionId = registry.resolve(args.session, "browser_inspect(action=network)");
         const cmdArgs = ["network", "--session", sessionId];
         appendDebugOptions(cmdArgs, args);
         const reply = (await runtime.run(exec, cmdArgs, "network", sessionId)) as {
@@ -592,7 +592,7 @@ export function registerPhaseOneSupportTools(
 
   register(
     defineTool({
-      name: "browser_window_resize",
+      name: "assist.resize",
       description: "Resize the owned session's Agent Window using outer CSS-pixel dimensions.",
       parameters: {
         width: {
@@ -629,7 +629,7 @@ export function registerPhaseOneSupportTools(
         if (args.width < 100 || args.width > 7680 || args.height < 100 || args.height > 7680) {
           throw new Error("width and height must each be in the range 100..=7680");
         }
-        const sessionId = registry.resolve(args.session, "browser_window_resize");
+        const sessionId = registry.resolve(args.session, "browser_assist(action=resize)");
         const reply = (await runtime.run(
           exec,
           [

packages/dsh-plugin-browserskill/src/phase-one-tools-tabs.ts

@@ -7,7 +7,7 @@ import type { ToolDeps } from "./tools";
 const TAB_ID_REQUIRED = {
   type: "integer",
   required: true,
-  description: "Chrome tab id returned by browser_tab_list or browser_tab_create.",
+  description: "Chrome tab id returned by browser_tabs list or create.",
 } as const;
 
 interface RawTabInfo {
@@ -29,7 +29,7 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_list",
+      name: "tabs.list",
       description:
         "List tabs visible to an owned browser session. Other sessions' Agent Windows remain " +
         "hidden; use scope=user to find a user tab before borrowing it.",
@@ -83,7 +83,7 @@ export function registerPhaseOneTabTools(
       },
       isConcurrencySafe: () => true,
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_tab_list");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=list)");
         const cmdArgs = ["tab", "list", "--session", sessionId];
         if (args.scope !== undefined) cmdArgs.push("--scope", args.scope);
         const reply = (await runtime.run(exec, cmdArgs, "tab list", sessionId)) as {
@@ -118,7 +118,7 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_create",
+      name: "tabs.create",
       description:
         "Create a tab inside the owned session's Agent Window. The tab is focused by default; " +
         "set active=false to open it in the background.",
@@ -148,7 +148,7 @@ export function registerPhaseOneTabTools(
       },
       async execute(args, exec) {
         if (args.url !== undefined) requireNonEmpty(args.url, "url");
-        const sessionId = registry.resolve(args.session, "browser_tab_create");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=create)");
         const cmdArgs = ["tab", "create", "--session", sessionId];
         if (args.url !== undefined) cmdArgs.push("--url", args.url);
         if (args.active === false) cmdArgs.push("--no-active");
@@ -185,7 +185,7 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_close",
+      name: "tabs.close",
       description: "Close a tab in the owned session's Agent Window.",
       parameters: { tabId: TAB_ID_REQUIRED, session: SESSION_PARAM },
       output: {
@@ -202,7 +202,7 @@ export function registerPhaseOneTabTools(
         ],
       },
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_tab_close");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=close)");
         const reply = (await runtime.run(
           exec,
           ["tab", "close", String(args.tabId), "--session", sessionId],
@@ -228,7 +228,7 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_select",
+      name: "tabs.select",
       description: "Focus a tab in the owned session's Agent Window.",
       parameters: { tabId: TAB_ID_REQUIRED, session: SESSION_PARAM },
       output: {
@@ -249,7 +249,7 @@ export function registerPhaseOneTabTools(
         ],
       },
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_tab_select");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=select)");
         const reply = (await runtime.run(
           exec,
           ["tab", "select", String(args.tabId), "--session", sessionId],
@@ -275,10 +275,10 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_borrow",
+      name: "tabs.borrow",
       description:
         "Move a user-window tab into the owned session's Agent Window for controlled interaction. " +
-        "Return it with browser_tab_return when finished; stopping the session also auto-returns it.",
+        "Return it with browser_tabs action=return; stopping the session also auto-returns it.",
       parameters: { tabId: TAB_ID_REQUIRED, session: SESSION_PARAM },
       output: {
         schema: {
@@ -302,7 +302,7 @@ export function registerPhaseOneTabTools(
         ],
       },
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_tab_borrow");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=borrow)");
         const reply = (await runtime.run(
           exec,
           ["tab", "borrow", String(args.tabId), "--session", sessionId],
@@ -339,7 +339,7 @@ export function registerPhaseOneTabTools(
 
   register(
     defineTool({
-      name: "browser_tab_return",
+      name: "tabs.return",
       description: "Return a borrowed tab to its original user window and tab-strip position.",
       parameters: { tabId: TAB_ID_REQUIRED, session: SESSION_PARAM },
       output: {
@@ -365,7 +365,7 @@ export function registerPhaseOneTabTools(
         ],
       },
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_tab_return");
+        const sessionId = registry.resolve(args.session, "browser_tabs(action=return)");
         const reply = (await runtime.run(
           exec,
           ["tab", "return", String(args.tabId), "--session", sessionId],

packages/dsh-plugin-browserskill/src/sessions.ts

@@ -7,7 +7,7 @@
  *
  * Strict ownership boundary (the bsk daemon may be shared with other agents,
  * terminals, or dsh instances): the registry ONLY ever holds sessions created
- * by this plugin's browser_session_start. Tools cannot see or act on any
+ * by this plugin's browser_session start action. Tools cannot see or act on any
  * other session — an explicit `session` argument naming a foreign id is an
  * error, the list tool shows owned sessions only, and stop/unload cleanup
  * can never touch a session this plugin did not create.
@@ -48,7 +48,7 @@ export class SessionRegistry {
     if (this.sessions.size + this.pendingStarts >= this.maxSessions) {
       throw new Error(
         `session limit reached (${this.maxSessions} concurrent sessions); ` +
-          "stop one with browser_session_stop before starting another",
+          "stop one with browser_session action=stop before starting another",
       );
     }
     this.pendingStarts += 1;
@@ -69,7 +69,7 @@ export class SessionRegistry {
     if (!this.sessions.has(session.sessionId) && this.sessions.size >= this.maxSessions) {
       throw new Error(
         `session limit reached (${this.maxSessions} concurrent sessions); ` +
-          "stop one with browser_session_stop before starting another",
+          "stop one with browser_session action=stop before starting another",
       );
     }
     this.sessions.set(session.sessionId, { ...session, owned: true });
@@ -149,7 +149,7 @@ export class SessionRegistry {
   private foreignError(sessionId: string, toolName: string): Error {
     return new Error(
       `${toolName}: session "${sessionId}" does not belong to this plugin — only sessions ` +
-        "created by browser_session_start are visible and operable here",
+        "created by browser_session action=start are visible and operable here",
     );
   }
 
@@ -168,7 +168,7 @@ export class SessionRegistry {
     const current = this.current();
     if (current === undefined) {
       throw new Error(
-        `${toolName} needs a session but none is active — start one with browser_session_start`,
+        `${toolName} needs a session but none is active — use browser_session action=start`,
       );
     }
     this.touch(current);
@@ -184,10 +184,11 @@ export class SessionRegistry {
       explicit !== undefined && explicit.trim().length > 0 ? explicit : this.current();
     if (candidate === undefined) {
       throw new Error(
-        "browser_session_stop needs a session but none is active — start one with browser_session_start",
+        "browser_session action=stop needs a session but none is active — use action=start first",
       );
     }
-    if (!this.isOwned(candidate)) throw this.foreignError(candidate, "browser_session_stop");
+    if (!this.isOwned(candidate))
+      throw this.foreignError(candidate, "browser_session(action=stop)");
     return candidate;
   }
 }

packages/dsh-plugin-browserskill/src/tool-params.ts

@@ -3,7 +3,7 @@
 export const SESSION_PARAM = {
   type: "string",
   description:
-    "bsk session id to act on; must be one created by this plugin's browser_session_start. " +
+    "bsk session id to act on; must be one created by browser_session with action=start. " +
     "Omit to use the current session (the one most recently started or used).",
 } as const;
 

packages/dsh-plugin-browserskill/src/tools.ts

@@ -1,9 +1,8 @@
 /**
- * Model-facing tool definitions. Each tool maps to one `bsk <cmd> --json`
- * invocation: spawn via the runner, parse the JSON envelope, and return a
- * canonical value whose `output.render` produces the model-facing text.
- * UI cards reuse the terminal style: the pending card is the command line,
- * the completed card carries the output.
+ * Internal browser operation definitions shared by the six model-facing tools.
+ * These definitions are never registered with `ctx.tools`: they provide the
+ * action-level validation, execution, rendering, and UI presentation reused by
+ * the domain tools without exposing one schema per operation to the model.
  */
 
 import { readFile, unlink } from "node:fs/promises";
@@ -158,22 +157,15 @@ function abortError(): Error {
   return error;
 }
 
-/** Register the full browser tool suite; returns the combined unregister disposer. */
-export function registerTools(deps: ToolDeps): () => void {
-  const { ctx, registry } = deps;
-  // Track every registration without touching the cordis context itself
-  // (contexts are fiber-owned; the lazy path registers from a listener fiber).
-  const disposers: (() => void)[] = [];
-  const register = (definition: ToolDefinition): (() => void) => {
-    const dispose = ctx.tools.register(definition);
-    // Test doubles occasionally return the registry map instead of a disposer.
-    if (typeof dispose === "function") disposers.push(dispose);
-    return dispose;
-  };
+type DefinitionRegistrar = (definition: ToolDefinition) => void;
+
+/** Define the private action handlers through an injected collector. */
+function defineBrowserOperations(deps: ToolDeps, register: DefinitionRegistrar): void {
+  const { registry } = deps;
 
   register(
     defineTool({
-      name: "browser_session_start",
+      name: "session.start",
       description:
         "Start a new browser session: opens an Agent Window in the connected browser and returns " +
         "its session id. The new session becomes the current session for subsequent browser_* calls. " +
@@ -314,10 +306,10 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_session_stop",
+      name: "session.stop",
       description:
         "Stop a browser session and close its Agent Window. Stops the given session, or the " +
-        "current session when `session` is omitted. Only sessions created by browser_session_start " +
+        "current session when `session` is omitted. Only plugin-created sessions " +
         "can be stopped — sessions owned by other programs sharing the bsk daemon are refused.",
       parameters: { session: SESSION_PARAM },
       output: {
@@ -354,7 +346,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_session_list",
+      name: "session.list",
       description:
         "List the browser sessions created by this plugin. Sessions owned by other programs on the " +
         "shared bsk daemon are not visible here. The session marked `current` is the one browser_* " +
@@ -419,7 +411,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_navigate",
+      name: "page.navigate",
       description:
         "Navigate the session's active tab to a URL and wait for a page lifecycle phase " +
         "(default: load). Returns the final URL after redirects.",
@@ -461,7 +453,7 @@ export function registerTools(deps: ToolDeps): () => void {
       },
       async execute(args, exec) {
         if (args.url.trim().length === 0) throw new Error("url must be a non-empty string");
-        const sessionId = registry.resolve(args.session, "browser_navigate");
+        const sessionId = registry.resolve(args.session, "browser_page(action=navigate)");
         const cmdArgs = ["navigate", "--session", sessionId];
         if (args.waitUntil !== undefined) cmdArgs.push("--wait-until", args.waitUntil);
         if (args.timeoutMs !== undefined) cmdArgs.push("--timeout", `${args.timeoutMs}ms`);
@@ -492,12 +484,12 @@ export function registerTools(deps: ToolDeps): () => void {
 
   const registerObservationTool = (kind: "snapshot" | "observe") => {
     const isSnapshot = kind === "snapshot";
-    const name = isSnapshot ? "browser_snapshot" : "browser_observe";
+    const name = isSnapshot ? "inspect.snapshot" : "inspect.observe";
     const description = isSnapshot
       ? "Capture an indented aria-tree snapshot of the session's active tab. Interactive elements " +
-        "carry @eN refs for browser_click / browser_fill. Prefer browser_observe for a richer semantic view."
+        "carry @eN refs for browser_interact actions. Prefer browser_inspect action=observe for a richer semantic view."
       : "Produce a semantic VOM observation of the session's active tab (roles, states, perception " +
-        "probes) with @eN refs for browser_click / browser_fill. Read-only: never submits input.";
+        "probes) with @eN refs for browser_interact actions. Read-only: never submits input.";
     register(
       defineTool({
         name,
@@ -566,10 +558,10 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_click",
+      name: "interact.click",
       description:
         "Click an element in the session's active tab. Target is a snapshot ref (@e3) from the " +
-        "last browser_snapshot/browser_observe, or a CSS selector.",
+        "last browser_inspect observation, or a CSS selector.",
       parameters: {
         target: {
           type: "string",
@@ -607,7 +599,7 @@ export function registerTools(deps: ToolDeps): () => void {
       },
       async execute(args, exec) {
         if (args.target.trim().length === 0) throw new Error("target must be a non-empty string");
-        const sessionId = registry.resolve(args.session, "browser_click");
+        const sessionId = registry.resolve(args.session, "browser_interact(action=click)");
         const cmdArgs = ["click", "--session", sessionId];
         if (args.button !== undefined) cmdArgs.push("--button", args.button);
         if (args.clickCount !== undefined) cmdArgs.push("--click-count", String(args.clickCount));
@@ -630,7 +622,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_fill",
+      name: "interact.fill",
       description:
         "Fill an input / textarea / contenteditable element, clearing it first by default. " +
         "Target is a snapshot ref (@e3) or a CSS selector.",
@@ -666,7 +658,7 @@ export function registerTools(deps: ToolDeps): () => void {
       },
       async execute(args, exec) {
         if (args.target.trim().length === 0) throw new Error("target must be a non-empty string");
-        const sessionId = registry.resolve(args.session, "browser_fill");
+        const sessionId = registry.resolve(args.session, "browser_interact(action=fill)");
         const cmdArgs = ["fill", "--session", sessionId, "--value", args.value];
         if (args.noClear === true) cmdArgs.push("--no-clear");
         cmdArgs.push(args.target);
@@ -687,7 +679,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_press",
+      name: "interact.press",
       description:
         "Dispatch a keyboard key or combo (Enter, Escape, ArrowDown, Ctrl+A, …) to the session's " +
         "active tab, optionally focusing a target element first.",
@@ -726,7 +718,7 @@ export function registerTools(deps: ToolDeps): () => void {
       },
       async execute(args, exec) {
         if (args.key.trim().length === 0) throw new Error("key must be a non-empty string");
-        const sessionId = registry.resolve(args.session, "browser_press");
+        const sessionId = registry.resolve(args.session, "browser_interact(action=press)");
         const cmdArgs = ["press", "--session", sessionId];
         if (args.target !== undefined) {
           if (args.target.trim().length === 0) throw new Error("target must be a non-empty string");
@@ -757,7 +749,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_screenshot",
+      name: "inspect.screenshot",
       description:
         "Capture a PNG screenshot of the session's active tab, or crop to a snapshot ref element. " +
         "Returns the image itself when the deployment supports image input, otherwise a file path.",
@@ -784,7 +776,11 @@ export function registerTools(deps: ToolDeps): () => void {
               additionalProperties: false,
               properties: {
                 attachmentId: { type: "string", required: true },
-                mediaType: { type: "string", required: true, const: "image/png" },
+                mediaType: {
+                  type: "string",
+                  required: true,
+                  enum: ["image/png", "image/jpeg", "image/webp", "image/gif"],
+                },
                 bytes: { type: "integer", required: true },
                 width: { type: "integer", required: true },
                 height: { type: "integer", required: true },
@@ -818,7 +814,7 @@ export function registerTools(deps: ToolDeps): () => void {
       },
       isConcurrencySafe: () => true,
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_screenshot");
+        const sessionId = registry.resolve(args.session, "browser_inspect(action=screenshot)");
         const outPath = join(
           tmpdir(),
           `bsk-screenshot-${Date.now()}-${Math.random().toString(36).slice(2, 8)}.png`,
@@ -854,7 +850,7 @@ export function registerTools(deps: ToolDeps): () => void {
               ? {
                   image: {
                     attachmentId: String(ref.attachmentId),
-                    mediaType: "image/png" as const,
+                    mediaType: ref.mediaType,
                     bytes: ref.bytes,
                     width: ref.width,
                     height: ref.height,
@@ -883,7 +879,7 @@ export function registerTools(deps: ToolDeps): () => void {
 
   register(
     defineTool({
-      name: "browser_emulate",
+      name: "assist.emulate",
       description:
         "Apply (or clear) mobile device emulation on the session's active tab: viewport metrics, " +
         "user agent, and touch. Overrides are per-tab and not inherited by new tabs.",
@@ -924,7 +920,7 @@ export function registerTools(deps: ToolDeps): () => void {
         ],
       },
       async execute(args, exec) {
-        const sessionId = registry.resolve(args.session, "browser_emulate");
+        const sessionId = registry.resolve(args.session, "browser_assist(action=emulate)");
         if (args.off === true) {
           if (args.device !== undefined || args.width !== undefined || args.height !== undefined) {
             throw new Error("off is mutually exclusive with device/width/height");
@@ -973,7 +969,17 @@ export function registerTools(deps: ToolDeps): () => void {
     commandLine: (args) => cmdline(deps, args),
     presentTerminalResult,
   });
-  return () => {
-    for (const dispose of disposers.splice(0)) dispose();
-  };
+}
+
+/**
+ * Build the private action handlers without publishing them to the model. The
+ * six browser tools dispatch through these handlers, preserving validation,
+ * rendering, cancellation, ownership, UI presentation, and attachments.
+ */
+export function createBrowserOperationDefinitions(deps: ToolDeps): ToolDefinition[] {
+  const definitions: ToolDefinition[] = [];
+  defineBrowserOperations(deps, (definition) => {
+    definitions.push(definition);
+  });
+  return definitions;
 }

packages/dsh-plugin-browserskill/tests/client/browser-inspect-toolview.test.tsx

@@ -1,17 +1,17 @@
 // @vitest-environment happy-dom
-// browser_screenshot keyed toolview: view-model derivation from frozen blocks,
+// browser_inspect keyed toolview: view-model derivation from frozen blocks,
 // image/path-only/running/error rendering, the registration key, and the
 // session-bound attachment loader.
 
 import type { RunningToolCall, ToolResultNode } from "@deepseek-ai/dsh-client-runtime/client";
 import { cleanup, render, screen, waitFor } from "@testing-library/react";
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { apply } from "../../src/client/index";
 import {
-  ScreenshotToolView,
-  type ScreenshotToolViewProps,
+  BrowserInspectToolView,
+  type BrowserInspectToolViewProps,
   viewModelOf,
-} from "../../src/client/ScreenshotToolView";
+} from "../../src/client/BrowserInspectToolView";
+import { apply } from "../../src/client/index";
 
 const ATTACHMENT = {
   attachmentId: "sha256:abc123",
@@ -25,7 +25,7 @@ const ATTACHMENT = {
 function runningBlock(argsRaw: string): RunningToolCall {
   return {
     callId: "c1",
-    name: "browser_screenshot",
+    name: "browser_inspect",
     argsRaw,
     turn: 1,
     step: 1,
@@ -41,7 +41,7 @@ function settledBlock(argsRaw: string, content: unknown[], isError = false): Too
     seq: 1,
     time: 0,
     callId: "c1",
-    call: { name: "browser_screenshot", argsRaw },
+    call: { name: "browser_inspect", argsRaw },
     callTime: 0,
     content,
     isError,
@@ -58,14 +58,14 @@ afterEach(cleanup);
 
 describe("viewModelOf", () => {
   it("derives the running model from the call frame", () => {
-    const model = viewModelOf(runningBlock('{"session":"s1"}'));
+    const model = viewModelOf(runningBlock('{"action":"screenshot","session":"s1"}'));
     expect(model.state).toBe("running");
     expect(model.command).toBe("bsk screenshot --session s1");
     expect(model.image).toBeNull();
   });
 
   it("marks settled results carrying an image block", () => {
-    const block = settledBlock("{}", [
+    const block = settledBlock('{"action":"screenshot"}', [
       { type: "text", text: IMAGE_TEXT },
       { type: "image", attachment: ATTACHMENT },
     ]);
@@ -77,30 +77,56 @@ describe("viewModelOf", () => {
   });
 
   it("keeps the path-only form image-free", () => {
-    const model = viewModelOf(settledBlock("{}", [{ type: "text", text: PATH_TEXT }]));
+    const model = viewModelOf(
+      settledBlock('{"action":"screenshot"}', [{ type: "text", text: PATH_TEXT }]),
+    );
     expect(model.state).toBe("ok");
     expect(model.image).toBeNull();
     expect(model.summary).toBe(PATH_TEXT);
   });
 
   it("marks error results", () => {
-    const model = viewModelOf(settledBlock("{}", [{ type: "text", text: "Error: boom" }], true));
+    const model = viewModelOf(
+      settledBlock('{"action":"screenshot"}', [{ type: "text", text: "Error: boom" }], true),
+    );
     expect(model.state).toBe("error");
   });
+
+  it("renders non-screenshot inspect actions as ordinary terminal calls", () => {
+    const model = viewModelOf(
+      settledBlock('{"action":"observe","session":"s1"}', [
+        { type: "text", text: "page observation" },
+      ]),
+    );
+    expect(model.title).toBe("Observe");
+    expect(model.command).toBe("bsk observe --session s1");
+    expect(model.image).toBeNull();
+  });
+
+  it("preserves action-specific arguments in diagnostic command cards", () => {
+    const model = viewModelOf(
+      runningBlock(
+        '{"action":"console","session":"s1","tabId":7,"since":4,"limit":20,"maxTextChars":500,"includeStack":true}',
+      ),
+    );
+    expect(model.command).toBe(
+      "bsk console --session s1 --tab-id 7 --since 4 --limit 20 --max-text-chars 500 --include-stack",
+    );
+  });
 });
 
-describe("ScreenshotToolView", () => {
-  const baseProps = { callId: "c1", toolName: "browser_screenshot", openFile: () => {} };
+describe("BrowserInspectToolView", () => {
+  const baseProps = { callId: "c1", toolName: "browser_inspect", openFile: () => {} };
   const renderView = (
-    block: ScreenshotToolViewProps["block"],
-    loadImage: ScreenshotToolViewProps["loadImage"],
+    block: BrowserInspectToolViewProps["block"],
+    loadImage: BrowserInspectToolViewProps["loadImage"],
   ) => {
-    const props = { ...baseProps, block, loadImage } as unknown as ScreenshotToolViewProps;
-    return render(<ScreenshotToolView {...props} />);
+    const props = { ...baseProps, block, loadImage } as unknown as BrowserInspectToolViewProps;
+    return render(<BrowserInspectToolView {...props} />);
   };
 
   it("renders the screenshot image through the loader", async () => {
-    const block = settledBlock("{}", [
+    const block = settledBlock('{"action":"screenshot"}', [
       { type: "text", text: IMAGE_TEXT },
       { type: "image", attachment: ATTACHMENT },
     ]);
@@ -115,7 +141,7 @@ describe("ScreenshotToolView", () => {
   });
 
   it("renders the path-only form without touching the loader", async () => {
-    const block = settledBlock("{}", [{ type: "text", text: PATH_TEXT }]);
+    const block = settledBlock('{"action":"screenshot"}', [{ type: "text", text: PATH_TEXT }]);
     const loadImage = vi.fn(async () => "blob:unused");
     renderView(block, loadImage);
     screen.getByRole("button", { name: /screenshot/i }).click();
@@ -125,14 +151,14 @@ describe("ScreenshotToolView", () => {
 
   it("renders the running form without output", () => {
     const loadImage = vi.fn();
-    renderView(runningBlock("{}"), loadImage);
+    renderView(runningBlock('{"action":"screenshot"}'), loadImage);
     expect(screen.getByText("bsk screenshot --session (current)")).toBeTruthy();
     expect(loadImage).not.toHaveBeenCalled();
   });
 });
 
 describe("client plugin registration", () => {
-  it("registers the browser_screenshot toolview and the shell observation overlay", () => {
+  it("registers the browser_inspect toolview and the shell observation overlay", () => {
     const registrations: { name: string; key?: string; id?: string }[] = [];
     const sessions = { binding: () => undefined };
     const ctx = {
@@ -150,7 +176,7 @@ describe("client plugin registration", () => {
     };
     apply(ctx as never);
     expect(registrations).toEqual([
-      { name: "tool.call.toolview", key: "browser_screenshot" },
+      { name: "tool.call.toolview", key: "browser_inspect" },
       { name: "shell.overlay", id: "bsk-observation" },
     ]);
   });

packages/dsh-plugin-browserskill/tests/dispose.test.ts

@@ -73,15 +73,15 @@ describe("dispose cleanup ownership", () => {
     };
     apply(ctx as never, { maxSessions: 5, lazyTools: false }, { runnerFactory: () => runner });
 
-    const start = tools.get("browser_session_start");
-    const snapshot = tools.get("browser_snapshot");
-    if (start === undefined || snapshot === undefined) throw new Error("tools not registered");
-    await start.execute({}, makeExec());
-    await start.execute({}, makeExec());
+    const session = tools.get("browser_session");
+    const inspect = tools.get("browser_inspect");
+    if (session === undefined || inspect === undefined) throw new Error("tools not registered");
+    await session.execute({ action: "start" }, makeExec());
+    await session.execute({ action: "start" }, makeExec());
     // A foreign session cannot even be referenced any more.
-    await expect(snapshot.execute({ session: "ext9" }, makeExec())).rejects.toThrow(
-      /does not belong to this plugin/,
-    );
+    await expect(
+      inspect.execute({ action: "snapshot", session: "ext9" }, makeExec()),
+    ).rejects.toThrow(/does not belong to this plugin/);
 
     await Promise.all(disposers.map((dispose) => dispose()));
 

packages/dsh-plugin-browserskill/tests/lazy-tools.test.ts

@@ -211,19 +211,25 @@ describe("lazyTools wiring in apply()", () => {
 
   it("lazyTools: false registers the suite at apply time", () => {
     const { tools } = applyHarness({ lazyTools: false });
-    expect(tools.has("browser_session_start")).toBe(true);
-    expect(tools.has("browser_emulate")).toBe(true);
+    expect([...tools.keys()].sort()).toEqual([
+      "browser_assist",
+      "browser_inspect",
+      "browser_interact",
+      "browser_page",
+      "browser_session",
+      "browser_tabs",
+    ]);
   });
 
   it("lazyTools: true hides the suite until the skill fires", () => {
     const { tools, listeners } = applyHarness({ lazyTools: true });
-    expect(tools.has("browser_session_start")).toBe(false);
+    expect(tools.has("browser_session")).toBe(false);
     callListeners(
       listeners,
       "tools/result",
       { name: "skill", arguments: { name: "browser-skill" } },
       { isError: false },
     );
-    expect(tools.has("browser_session_start")).toBe(true);
+    expect(tools.has("browser_session")).toBe(true);
   });
 });

packages/dsh-plugin-browserskill/tests/plugin-id.test.ts

@@ -0,0 +1,32 @@
+// dsh 0.1 keys the client graph and Node import by `export const name`.
+// That string must stay identical to the published package name (and therefore
+// the client.cjs __ModuleLoader__ id, which tsdown reads from package.json).
+
+import { readFileSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { describe, expect, it } from "vitest";
+import { name } from "../src/index";
+
+const pkg = JSON.parse(
+  readFileSync(join(dirname(fileURLToPath(import.meta.url)), "..", "package.json"), "utf8"),
+) as {
+  name: string;
+  dsh: { client: { external?: string[] } };
+};
+
+describe("plugin identity", () => {
+  it("uses the published package name as the Cordis plugin id", () => {
+    expect(name).toBe("@wxg-prc-cpg/browser-skill-dsh-plugin");
+    expect(name).toBe(pkg.name);
+  });
+
+  it("declares client require()s so dsh arrives them before materialize", () => {
+    expect(pkg.dsh.client.external).toEqual(
+      expect.arrayContaining([
+        "@deepseek-ai/dsh-client-ui-attachment",
+        "@deepseek-ai/dsh-client-ui-primitives",
+      ]),
+    );
+  });
+});

packages/dsh-plugin-browserskill/tests/sessions.test.ts

@@ -71,7 +71,9 @@ describe("SessionRegistry", () => {
 
   it("resolve throws actionable guidance when no session exists", () => {
     const registry = new SessionRegistry(5);
-    expect(() => registry.resolve(undefined, "browser_click")).toThrow(/browser_session_start/);
+    expect(() => registry.resolve(undefined, "browser_interact(action=click)")).toThrow(
+      /browser_session action=start/,
+    );
   });
 
   it("recency order survives remove of the current session", () => {

packages/dsh-plugin-browserskill/tests/skill.test.ts

@@ -32,44 +32,22 @@ describe("registerBskSkill", () => {
     const content = String(skill.content);
     const mentionedTools = [...new Set(content.match(/\bbrowser_[a-z][a-z_]*\b/g) ?? [])].sort();
     expect(mentionedTools).toEqual([
-      "browser_click",
-      "browser_console",
-      "browser_emulate",
-      "browser_fill",
-      "browser_get_html",
-      "browser_hover",
-      "browser_navigate",
-      "browser_navigate_back",
-      "browser_navigate_forward",
-      "browser_network",
-      "browser_observe",
-      "browser_press",
-      "browser_reload",
-      "browser_request_help",
-      "browser_screenshot",
-      "browser_select",
-      "browser_session_list",
-      "browser_session_start",
-      "browser_session_stop",
-      "browser_snapshot",
-      "browser_tab_borrow",
-      "browser_tab_close",
-      "browser_tab_create",
-      "browser_tab_list",
-      "browser_tab_return",
-      "browser_tab_select",
-      "browser_wait_for_navigation",
-      "browser_window_resize",
+      "browser_assist",
+      "browser_inspect",
+      "browser_interact",
+      "browser_page",
+      "browser_session",
+      "browser_tabs",
     ]);
     expect(String(skill.description)).not.toMatch(/\bbsk\b/i);
     expect(content).not.toMatch(/\bbsk\b/i);
     expect(content).not.toMatch(/```(?:bash|sh|shell)\b/i);
     expect(content).not.toMatch(/--[a-z]/);
-    expect(content).toContain("All browser operations must use the injected tools directly");
+    expect(content).toMatch(/All browser work\s+must use the injected tools directly/);
     expect(content).toContain("Mandatory workflow");
     expect(content).toContain("Refs invalidate after navigation");
     expect(content).toContain("evaluation and interaction recording are intentionally unsupported");
-    expect(content.length).toBeGreaterThan(10_000);
+    expect(content.length).toBeGreaterThan(6_000);
     expect(skill.source).toBe("bundled");
     // Source frontmatter is registration metadata and must not leak into the body.
     expect(content.startsWith("---")).toBe(false);
@@ -144,7 +122,7 @@ describe("armAgentScopedBskSkill", () => {
     });
 
     const after = await skillFiber.ctx.skills.get("browser-skill", { scope: agentKey });
-    expect(after?.content).toContain("All browser operations must use the injected tools directly");
+    expect(after?.content).toMatch(/All browser work\s+must use the injected tools directly/);
     expect(after?.content).not.toMatch(/\bbsk\b/i);
     expect(after?.source).toBe("bundled");
 

packages/dsh-plugin-browserskill/tests/tools.test.ts

@@ -3,11 +3,12 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 import type { ToolDefinition, ToolRunContext } from "@deepseek-ai/dsh-tools";
 import { describe, expect, it, vi } from "vitest";
+import { registerBrowserTools } from "../src/browser-tools";
 import { ObservationService } from "../src/observation";
 import { KeyedExecutor } from "../src/queue";
 import type { BskRunner, BskRunOptions, BskRunResult } from "../src/runner";
 import { SessionRegistry } from "../src/sessions";
-import { type PluginConfig, registerTools } from "../src/tools";
+import type { PluginConfig } from "../src/tools";
 
 const CONFIG: PluginConfig = {
   bskPath: "bsk",
@@ -84,8 +85,66 @@ function fakeRunner(responses: Record<string, unknown>) {
   };
 }
 
+const ACTION_ROUTES: Record<string, readonly [string, string]> = {
+  "session.start": ["browser_session", "start"],
+  "session.stop": ["browser_session", "stop"],
+  "session.list": ["browser_session", "list"],
+  "page.navigate": ["browser_page", "navigate"],
+  "page.back": ["browser_page", "back"],
+  "page.forward": ["browser_page", "forward"],
+  "page.reload": ["browser_page", "reload"],
+  "page.wait": ["browser_page", "wait"],
+  "inspect.observe": ["browser_inspect", "observe"],
+  "inspect.snapshot": ["browser_inspect", "snapshot"],
+  "inspect.html": ["browser_inspect", "html"],
+  "inspect.screenshot": ["browser_inspect", "screenshot"],
+  "inspect.console": ["browser_inspect", "console"],
+  "inspect.network": ["browser_inspect", "network"],
+  "interact.click": ["browser_interact", "click"],
+  "interact.hover": ["browser_interact", "hover"],
+  "interact.fill": ["browser_interact", "fill"],
+  "interact.select": ["browser_interact", "select"],
+  "interact.press": ["browser_interact", "press"],
+  "tabs.list": ["browser_tabs", "list"],
+  "tabs.create": ["browser_tabs", "create"],
+  "tabs.select": ["browser_tabs", "select"],
+  "tabs.close": ["browser_tabs", "close"],
+  "tabs.borrow": ["browser_tabs", "borrow"],
+  "tabs.return": ["browser_tabs", "return"],
+  "assist.resize": ["browser_assist", "resize"],
+  "assist.emulate": ["browser_assist", "emulate"],
+  "assist.request-help": ["browser_assist", "request-help"],
+};
+
+class TestToolRegistry extends Map<string, ToolDefinition> {
+  override get(name: string): ToolDefinition | undefined {
+    const direct = super.get(name);
+    if (direct !== undefined) return direct;
+    const route = ACTION_ROUTES[name];
+    if (route === undefined) return undefined;
+    const [toolName, action] = route;
+    const tool = super.get(toolName);
+    if (tool === undefined) return undefined;
+    const withAction = (args: unknown) => ({
+      action,
+      ...(args as Record<string, unknown>),
+    });
+    return {
+      ...tool,
+      output: {
+        ...tool.output,
+        render: (args, value) => tool.output.render(withAction(args) as never, value),
+      },
+      execute: (args, exec) => tool.execute(withAction(args) as never, exec),
+      isConcurrencySafe: (args) => tool.isConcurrencySafe?.(withAction(args) as never) === true,
+      presentCall: (args) => tool.presentCall?.(withAction(args) as never),
+      presentResult: (args, result) => tool.presentResult?.(withAction(args) as never, result),
+    } as ToolDefinition;
+  }
+}
+
 function makeCtx(services: Record<string, unknown> = {}) {
-  const tools = new Map<string, ToolDefinition>();
+  const tools = new TestToolRegistry();
   const ctx = {
     tools: { register: (def: ToolDefinition) => tools.set(def.name, def) },
     get: (key: string) => services[key],
@@ -118,7 +177,7 @@ function setup(
   const { runner, calls } = fakeRunner(responses);
   const registry = new SessionRegistry(maxSessions);
   const config = { ...CONFIG, maxSessions };
-  registerTools({
+  registerBrowserTools({
     ctx: ctx as never,
     runner: runner as BskRunner,
     registry,
@@ -130,56 +189,98 @@ function setup(
 }
 
 async function startSession(tools: Map<string, ToolDefinition>, sessionId = "s1") {
-  const tool = tools.get("browser_session_start");
-  if (tool === undefined) throw new Error("browser_session_start not registered");
+  const tool = tools.get("session.start");
+  if (tool === undefined) throw new Error("session.start not registered");
   return tool.execute({}, makeExec()) as Promise<{ sessionId: string }>;
 }
 
 const START_REPLY = (id: string) => ({ session_id: id, browser_instance_id: "chrome-1" });
 
+const EXPECTED_ACTIONS = {
+  browser_session: ["start", "stop", "list"],
+  browser_page: ["navigate", "back", "forward", "reload", "wait"],
+  browser_inspect: ["observe", "snapshot", "html", "screenshot", "console", "network"],
+  browser_interact: ["click", "hover", "fill", "select", "press"],
+  browser_tabs: ["list", "create", "select", "close", "borrow", "return"],
+  browser_assist: ["resize", "emulate", "request-help"],
+} as const;
+
 describe("tool registration", () => {
-  it("registers the full browser tool suite", () => {
+  it("registers exactly six public schemas with the complete action contract", () => {
     const { tools } = setup({});
-    expect([...tools.keys()].sort()).toEqual([
-      "browser_click",
-      "browser_console",
-      "browser_emulate",
-      "browser_fill",
-      "browser_get_html",
-      "browser_hover",
-      "browser_navigate",
-      "browser_navigate_back",
-      "browser_navigate_forward",
-      "browser_network",
-      "browser_observe",
-      "browser_press",
-      "browser_reload",
-      "browser_request_help",
-      "browser_screenshot",
-      "browser_select",
-      "browser_session_list",
-      "browser_session_start",
-      "browser_session_stop",
-      "browser_snapshot",
-      "browser_tab_borrow",
-      "browser_tab_close",
-      "browser_tab_create",
-      "browser_tab_list",
-      "browser_tab_return",
-      "browser_tab_select",
-      "browser_wait_for_navigation",
-      "browser_window_resize",
-    ]);
+    expect([...tools.keys()].sort()).toEqual(Object.keys(EXPECTED_ACTIONS).sort());
+    for (const [name, actions] of Object.entries(EXPECTED_ACTIONS)) {
+      const properties = tools.get(name)?.parameters.properties as
+        | Record<string, unknown>
+        | undefined;
+      expect(properties?.action).toMatchObject({ enum: [...actions] });
+    }
+  });
+
+  it("requires an action on every public tool", async () => {
+    const { tools } = setup({});
+    for (const name of Object.keys(EXPECTED_ACTIONS)) {
+      await expect(tools.get(name)?.execute({}, makeExec())).rejects.toThrow(/invalid arguments/i);
+    }
   });
 
   it("validates model args through defineTool (missing required param)", async () => {
     const { tools } = setup({});
-    const navigate = tools.get("browser_navigate");
+    const navigate = tools.get("page.navigate");
     await expect(navigate?.execute({}, makeExec())).rejects.toThrow(/invalid arguments/);
   });
 });
 
-describe("browser_session_start", () => {
+describe("action dispatch", () => {
+  it("rejects actions outside the public contract", async () => {
+    const { tools } = setup({});
+    await expect(
+      tools.get("browser_page")?.execute({ action: "unsupported" }, makeExec()),
+    ).rejects.toThrow(/invalid arguments|unsupported browser action/i);
+  });
+
+  it("dispatches through the existing operation definitions", async () => {
+    const { tools, calls, registry } = setup({
+      "session start": START_REPLY("s1"),
+      navigate: { tab_id: 7, url: "https://example.test/", reached: "load" },
+      observe: SNAPSHOT_REPLY,
+      fill: { tab_id: 7, value_length: 5 },
+      "tab list": { tabs: [] },
+      "window resize": { window_id: 9, width: 900, height: 700 },
+    });
+    await tools.get("browser_session")?.execute({ action: "start" }, makeExec());
+    await tools
+      .get("browser_page")
+      ?.execute({ action: "navigate", url: "https://example.test/" }, makeExec());
+    await tools.get("browser_inspect")?.execute({ action: "observe" }, makeExec());
+    await tools
+      .get("browser_interact")
+      ?.execute({ action: "fill", target: "@e1", value: "hello" }, makeExec());
+    await tools.get("browser_tabs")?.execute({ action: "list" }, makeExec());
+    await tools
+      .get("browser_assist")
+      ?.execute({ action: "resize", width: 900, height: 700 }, makeExec());
+
+    expect(registry.current()).toBe("s1");
+    expect(calls.map(({ args }) => args)).toEqual([
+      ["session", "start"],
+      ["navigate", "--session", "s1", "https://example.test/"],
+      ["observe", "--session", "s1"],
+      ["fill", "--session", "s1", "--value", "hello", "@e1"],
+      ["tab", "list", "--session", "s1"],
+      ["window", "resize", "--session", "s1", "--width", "900", "--height", "700"],
+    ]);
+  });
+
+  it("keeps action-specific validation in the private operation", async () => {
+    const { tools } = setup({});
+    await expect(
+      tools.get("browser_page")?.execute({ action: "navigate" }, makeExec()),
+    ).rejects.toThrow(/url.*required|invalid arguments/i);
+  });
+});
+
+describe("session.start", () => {
   it("maps the start reply and tracks the session as current", async () => {
     const { tools, registry } = setup({ "session start": START_REPLY("s1") });
     const value = await startSession(tools);
@@ -193,7 +294,7 @@ describe("browser_session_start", () => {
       emulate: { tab_id: 7, cleared: false },
       navigate: { tab_id: 7, url: "https://example.com", reached: "load" },
     });
-    const tool = tools.get("browser_session_start");
+    const tool = tools.get("session.start");
     const value = (await tool?.execute(
       { width: 1280, height: 800, url: "https://example.com", device: "iphone-14", noFocus: true },
       makeExec(),
@@ -218,7 +319,7 @@ describe("browser_session_start", () => {
 
   it("requires width and height together", async () => {
     const { tools } = setup({ "session start": START_REPLY("s1") });
-    const tool = tools.get("browser_session_start");
+    const tool = tools.get("session.start");
     await expect(tool?.execute({ width: 1280 }, makeExec())).rejects.toThrow(/together/);
   });
 
@@ -228,7 +329,7 @@ describe("browser_session_start", () => {
       "session stop": {},
     });
     // navigate has no canned response -> non-zero exit -> error.
-    const tool = tools.get("browser_session_start");
+    const tool = tools.get("session.start");
     await expect(tool?.execute({ url: "https://example.com" }, makeExec())).rejects.toThrow(
       /navigate/,
     );
@@ -246,7 +347,7 @@ describe("multi-session behavior", () => {
   it("operation tools default to the current session and echo it back", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const snapshot = tools.get("browser_snapshot");
+    const snapshot = tools.get("inspect.snapshot");
     const value = (await snapshot?.execute({}, makeExec())) as {
       session: string;
       refCount: number;
@@ -262,7 +363,7 @@ describe("multi-session behavior", () => {
     });
     await startSession(tools, "s1");
     await startSession(tools, "s2");
-    const snapshot = tools.get("browser_snapshot");
+    const snapshot = tools.get("inspect.snapshot");
     const value = (await snapshot?.execute({ session: "s1" }, makeExec())) as {
       session: string;
     };
@@ -275,7 +376,7 @@ describe("multi-session behavior", () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
     const callsBefore = calls.length;
-    const snapshot = tools.get("browser_snapshot");
+    const snapshot = tools.get("inspect.snapshot");
     await expect(snapshot?.execute({ session: "other" }, makeExec())).rejects.toThrow(
       /does not belong to this plugin/,
     );
@@ -284,8 +385,8 @@ describe("multi-session behavior", () => {
 
   it("fails with guidance when no session exists", async () => {
     const { tools } = setup(responses);
-    const snapshot = tools.get("browser_snapshot");
-    await expect(snapshot?.execute({}, makeExec())).rejects.toThrow(/browser_session_start/);
+    const snapshot = tools.get("inspect.snapshot");
+    await expect(snapshot?.execute({}, makeExec())).rejects.toThrow(/browser_session action=start/);
   });
 
   it("enforces the configured session cap before spawning", async () => {
@@ -295,7 +396,7 @@ describe("multi-session behavior", () => {
       1,
     );
     await startSession(tools);
-    const start = tools.get("browser_session_start");
+    const start = tools.get("session.start");
     await expect(start?.execute({}, makeExec())).rejects.toThrow(/session limit/);
     // A rejected start must not spawn another bsk session (no leaked session).
     expect(calls.filter((c) => c.args.join(" ").startsWith("session start"))).toHaveLength(1);
@@ -307,7 +408,7 @@ describe("multi-session behavior", () => {
       {},
       1,
     );
-    const start = tools.get("browser_session_start");
+    const start = tools.get("session.start");
     if (start === undefined) throw new Error("not registered");
     // Two truly concurrent starts against a cap of 1: exactly one may spawn.
     const outcomes = await Promise.allSettled([
@@ -325,7 +426,7 @@ describe("multi-session behavior", () => {
   it("refuses to stop a session the plugin did not create", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const stop = tools.get("browser_session_stop");
+    const stop = tools.get("session.stop");
     await expect(stop?.execute({ session: "ext7" }, makeExec())).rejects.toThrow(
       /does not belong to this plugin/,
     );
@@ -334,14 +435,14 @@ describe("multi-session behavior", () => {
   });
 });
 
-describe("browser_session_stop / list", () => {
+describe("session.stop / list", () => {
   it("stops the current session and clears the registry", async () => {
     const { tools, calls, registry } = setup({
       "session start": START_REPLY("s1"),
       "session stop": {},
     });
     await startSession(tools);
-    const stop = tools.get("browser_session_stop");
+    const stop = tools.get("session.stop");
     const value = (await stop?.execute({}, makeExec())) as { stopped: string };
     expect(value.stopped).toBe("s1");
     expect(calls[1].args).toEqual(["session", "stop", "s1"]);
@@ -354,7 +455,7 @@ describe("browser_session_stop / list", () => {
     });
     await startSession(tools);
     await startSession(tools);
-    const list = tools.get("browser_session_list");
+    const list = tools.get("session.list");
     const value = (await list?.execute({}, makeExec())) as {
       sessions: { sessionId: string; current: boolean }[];
     };
@@ -377,10 +478,10 @@ describe("interaction tools", () => {
     emulate: { tab_id: 7, cleared: false },
   };
 
-  it("browser_click maps target, button, and click count", async () => {
+  it("interact.click maps target, button, and click count", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const click = tools.get("browser_click");
+    const click = tools.get("interact.click");
     const value = (await click?.execute(
       { target: "@e1", button: "right", clickCount: 2 },
       makeExec(),
@@ -398,10 +499,10 @@ describe("interaction tools", () => {
     ]);
   });
 
-  it("browser_fill passes value and --no-clear", async () => {
+  it("interact.fill passes value and --no-clear", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const fill = tools.get("browser_fill");
+    const fill = tools.get("interact.fill");
     await fill?.execute({ target: "e2", value: "hello", noClear: true }, makeExec());
     expect(calls[1].args).toEqual([
       "fill",
@@ -414,20 +515,20 @@ describe("interaction tools", () => {
     ]);
   });
 
-  it("browser_press picks --ref for snapshot refs and --selector otherwise", async () => {
+  it("interact.press picks --ref for snapshot refs and --selector otherwise", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const press = tools.get("browser_press");
+    const press = tools.get("interact.press");
     await press?.execute({ key: "Enter", target: "@e3" }, makeExec());
     expect(calls[1].args).toEqual(["press", "--session", "s1", "--ref", "@e3", "Enter"]);
     await press?.execute({ key: "Ctrl+A", target: "#field" }, makeExec());
     expect(calls[2].args).toEqual(["press", "--session", "s1", "--selector", "#field", "Ctrl+A"]);
   });
 
-  it("browser_navigate converts waitUntil and timeoutMs", async () => {
+  it("page.navigate converts waitUntil and timeoutMs", async () => {
     const { tools, calls } = setup(responses);
     await startSession(tools);
-    const navigate = tools.get("browser_navigate");
+    const navigate = tools.get("page.navigate");
     const value = (await navigate?.execute(
       { url: "https://a.test", waitUntil: "networkidle", timeoutMs: 1500 },
       makeExec(),
@@ -445,16 +546,16 @@ describe("interaction tools", () => {
     ]);
   });
 
-  it("browser_emulate rejects off combined with overrides", async () => {
+  it("assist.emulate rejects off combined with overrides", async () => {
     const { tools } = setup(responses);
     await startSession(tools);
-    const emulate = tools.get("browser_emulate");
+    const emulate = tools.get("assist.emulate");
     await expect(emulate?.execute({ off: true, device: "iphone-14" }, makeExec())).rejects.toThrow(
       /mutually exclusive/,
     );
   });
 
-  it("browser_hover and browser_select map their structured arguments", async () => {
+  it("interact.hover and interact.select map their structured arguments", async () => {
     const { tools, calls } = setup({
       "session start": START_REPLY("s1"),
       hover: { tab_id: 7, used_ref: "@e3", x: 12, y: 24 },
@@ -468,7 +569,7 @@ describe("interaction tools", () => {
     });
     await startSession(tools);
     const hovered = (await tools
-      .get("browser_hover")
+      .get("interact.hover")
       ?.execute(
         { target: "@e3", modifiers: ["shift"], settleMs: 250, timeoutMs: 2000 },
         makeExec(),
@@ -488,7 +589,7 @@ describe("interaction tools", () => {
     ]);
 
     const selected = (await tools
-      .get("browser_select")
+      .get("interact.select")
       ?.execute({ target: "@e4", values: ["a", "b"], tabId: 7 }, makeExec())) as {
       selectedValues: string[];
       selectedLabels: string[];
@@ -546,25 +647,25 @@ describe("phase-one tab tools", () => {
     });
     await startSession(tools);
 
-    const listed = (await tools
-      .get("browser_tab_list")
-      ?.execute({ scope: "agent" }, makeExec())) as { tabs: { tabId: number }[] };
+    const listed = (await tools.get("tabs.list")?.execute({ scope: "agent" }, makeExec())) as {
+      tabs: { tabId: number }[];
+    };
     expect(listed.tabs[0].tabId).toBe(7);
     const created = (await tools
-      .get("browser_tab_create")
+      .get("tabs.create")
       ?.execute({ url: "https://new.test", active: false, index: 1 }, makeExec())) as {
       tabId: number;
     };
     expect(created.tabId).toBe(8);
-    await tools.get("browser_tab_close")?.execute({ tabId: 8 }, makeExec());
-    await tools.get("browser_tab_select")?.execute({ tabId: 7 }, makeExec());
-    const borrowed = (await tools
-      .get("browser_tab_borrow")
-      ?.execute({ tabId: 20 }, makeExec())) as { originalWindowId: number };
+    await tools.get("tabs.close")?.execute({ tabId: 8 }, makeExec());
+    await tools.get("tabs.select")?.execute({ tabId: 7 }, makeExec());
+    const borrowed = (await tools.get("tabs.borrow")?.execute({ tabId: 20 }, makeExec())) as {
+      originalWindowId: number;
+    };
     expect(borrowed.originalWindowId).toBe(9);
-    const returned = (await tools
-      .get("browser_tab_return")
-      ?.execute({ tabId: 20 }, makeExec())) as { fallback: boolean };
+    const returned = (await tools.get("tabs.return")?.execute({ tabId: 20 }, makeExec())) as {
+      fallback: boolean;
+    };
     expect(returned.fallback).toBe(false);
 
     expect(calls.slice(1).map((call) => call.args)).toEqual([
@@ -614,15 +715,14 @@ describe("phase-one navigation tools", () => {
     });
     await startSession(tools);
     await tools
-      .get("browser_navigate_back")
+      .get("page.back")
       ?.execute({ tabId: 7, waitUntil: "commit", timeoutMs: 1000 }, makeExec());
-    await tools.get("browser_navigate_forward")?.execute({}, makeExec());
-    await tools
-      .get("browser_reload")
-      ?.execute({ hard: true, waitUntil: "networkidle" }, makeExec());
-    const waited = (await tools
-      .get("browser_wait_for_navigation")
-      ?.execute({ timeoutMs: 2500 }, makeExec())) as { reached: string; errorText: string };
+    await tools.get("page.forward")?.execute({}, makeExec());
+    await tools.get("page.reload")?.execute({ hard: true, waitUntil: "networkidle" }, makeExec());
+    const waited = (await tools.get("page.wait")?.execute({ timeoutMs: 2500 }, makeExec())) as {
+      reached: string;
+      errorText: string;
+    };
     expect(waited).toMatchObject({ reached: "timeout", errorText: "no navigation" });
     expect(calls[1].args).toEqual([
       "navigate-back",
@@ -668,7 +768,7 @@ describe("phase-one support tools", () => {
       },
     });
     await startSession(tools);
-    const value = (await tools.get("browser_request_help")?.execute(
+    const value = (await tools.get("assist.request-help")?.execute(
       {
         prompt: "Please sign in",
         title: "Login required",
@@ -736,17 +836,17 @@ describe("phase-one support tools", () => {
     });
     await startSession(tools);
     const html = (await tools
-      .get("browser_get_html")
+      .get("inspect.html")
       ?.execute({ ref: "@e2", maxBytes: 2048 }, makeExec())) as { html: string };
     expect(html.html).toBe("<main>Hi</main>");
     const consoleValue = (await tools
-      .get("browser_console")
+      .get("inspect.console")
       ?.execute({ since: 1, limit: 10, includeStack: true }, makeExec())) as {
       entries: { stackTrace: { functionName: string }[] }[];
     };
     expect(consoleValue.entries[0].stackTrace[0].functionName).toBe("run");
     const networkValue = (await tools
-      .get("browser_network")
+      .get("inspect.network")
       ?.execute({ tabId: 7, maxTextChars: 512 }, makeExec())) as {
       entries: { mimeType: string }[];
     };
@@ -764,7 +864,7 @@ describe("phase-one support tools", () => {
       "window resize": { window_id: 2, width: 1280, height: 800 },
     });
     await startSession(tools);
-    const resize = tools.get("browser_window_resize");
+    const resize = tools.get("assist.resize");
     const value = (await resize?.execute({ width: 1280, height: 800 }, makeExec())) as {
       windowId: number;
     };
@@ -785,7 +885,7 @@ describe("phase-one support tools", () => {
   });
 });
 
-describe("browser_screenshot", () => {
+describe("inspect.screenshot", () => {
   function pngFile(): string {
     const dir = mkdtempSync(join(tmpdir(), "bsk-test-"));
     const path = join(dir, "shot.png");
@@ -794,6 +894,15 @@ describe("browser_screenshot", () => {
     return path;
   }
 
+  function jpegFile(): string {
+    const dir = mkdtempSync(join(tmpdir(), "bsk-test-"));
+    const path = join(dir, "shot.png");
+    // A JPEG SOI marker (FF D8 FF) even though the file is named ".png" —
+    // mirrors captureVisibleTab returning JPEG on some Chromium builds.
+    writeFileSync(path, Buffer.from([0xff, 0xd8, 0xff, 0xdb, 0, 1, 2, 3, 4]));
+    return path;
+  }
+
   it("returns the PNG path when no attachment store is mounted", async () => {
     const path = pngFile();
     const { tools, calls } = setup({
@@ -801,7 +910,7 @@ describe("browser_screenshot", () => {
       screenshot: { tab_id: 7, width: 800, height: 600, format: "png", path, byte_size: 8 },
     });
     await startSession(tools);
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const value = (await screenshot?.execute({ ref: "@e1" }, makeExec())) as {
       path: string;
       image?: unknown;
@@ -850,7 +959,7 @@ describe("browser_screenshot", () => {
         options: {},
       },
     });
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const value = (await screenshot?.execute({}, exec)) as {
       image?: { attachmentId: string; mediaType: string };
     };
@@ -859,6 +968,50 @@ describe("browser_screenshot", () => {
     expect(rendered.map((block) => block.type)).toEqual(["text", "image"]);
   });
 
+  it("declares the real media type when capture bytes are JPEG, not a hard-coded PNG", async () => {
+    const path = jpegFile();
+    const services = {
+      attachments: {
+        imageLimits: {
+          mediaTypes: ["image/png", "image/jpeg", "image/webp", "image/gif"],
+          maxImageBytes: 10_000_000,
+          maxMessageImageBytes: 10_000_000,
+        },
+        saveImage: async (input: { data: Uint8Array; mediaType: string; name?: string }) => ({
+          attachmentId: "att-jpeg",
+          mediaType: input.mediaType,
+          bytes: input.data.byteLength,
+          width: 800,
+          height: 600,
+          name: input.name,
+        }),
+      },
+      llm: {
+        resolveModelInfo: async () => ({ inputModalities: ["text", "image"] }),
+      },
+    };
+    const { tools } = setup(
+      {
+        "session start": START_REPLY("s1"),
+        screenshot: { tab_id: 7, width: 800, height: 600, format: "png", path, byte_size: 9 },
+      },
+      services,
+    );
+    await startSession(tools);
+    const exec = makeExec({
+      agent: {
+        session: { requestHeader: () => ({ config: { provider: "deepseek", model: "vl" } }) },
+        options: {},
+      },
+    });
+    const screenshot = tools.get("inspect.screenshot");
+    const value = (await screenshot?.execute({}, exec)) as {
+      image?: { attachmentId: string; mediaType: string };
+    };
+    // The reference media type must reflect the sniffed JPEG bytes, not "image/png".
+    expect(value.image).toMatchObject({ attachmentId: "att-jpeg", mediaType: "image/jpeg" });
+  });
+
   it("stays path-only when the model route is text-only", async () => {
     const path = pngFile();
     const services = {
@@ -884,7 +1037,7 @@ describe("browser_screenshot", () => {
         options: {},
       },
     });
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const value = (await screenshot?.execute({}, exec)) as { image?: unknown; path: string };
     expect(value.image).toBeUndefined();
     expect(value.path).toBe(path);
@@ -897,14 +1050,14 @@ describe("error and cancellation semantics", () => {
       "session start": START_REPLY("s1"),
     });
     await startSession(tools);
-    const snapshot = tools.get("browser_snapshot");
+    const snapshot = tools.get("inspect.snapshot");
     await expect(snapshot?.execute({}, makeExec())).rejects.toThrow(/no canned response/);
   });
 
   it("turns an aborted child into an AbortError", async () => {
     const { tools } = setup({ "session start": START_REPLY("s1") });
     await startSession(tools);
-    const snapshot = tools.get("browser_snapshot");
+    const snapshot = tools.get("inspect.snapshot");
     const controller = new AbortController();
     controller.abort();
     const exec = makeExec({ signal: controller.signal });
@@ -921,7 +1074,7 @@ describe("error and cancellation semantics", () => {
       killAll() {},
       killFor: () => 0,
     };
-    registerTools({
+    registerBrowserTools({
       ctx: ctx as never,
       runner,
       registry,
@@ -929,7 +1082,7 @@ describe("error and cancellation semantics", () => {
       observation: disabledObservation({ ctx, runner, registry }),
       queue: new KeyedExecutor(),
     });
-    const start = tools.get("browser_session_start");
+    const start = tools.get("session.start");
     await expect(start?.execute({}, makeExec())).rejects.toThrow(/BrowserSkill must be installed/);
   });
 });
@@ -937,7 +1090,7 @@ describe("error and cancellation semantics", () => {
 describe("presentation", () => {
   it("presents calls as terminal cards with the bsk command line", () => {
     const { tools } = setup({});
-    const click = tools.get("browser_click");
+    const click = tools.get("interact.click");
     const view = click?.presentCall?.({ target: "@e1" }) as {
       card: string;
       title: string;
@@ -948,7 +1101,7 @@ describe("presentation", () => {
 
   it("presents results as terminal cards carrying the rendered text", () => {
     const { tools } = setup({});
-    const click = tools.get("browser_click");
+    const click = tools.get("interact.click");
     const view = click?.presentResult?.({ target: "@e1" }, {
       content: [{ type: "text", text: "clicked" }],
       isError: false,
@@ -958,7 +1111,7 @@ describe("presentation", () => {
 
   it("projects the text block of multi-block results (screenshot text + image)", () => {
     const { tools } = setup({});
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const view = screenshot?.presentResult?.({}, {
       content: [
         { type: "text", text: "[session s1] screenshot of tab 7" },
@@ -1017,7 +1170,7 @@ describe("screenshot scratch file lifecycle", () => {
   function setupWithRunner(runner: BskRunner, services: Record<string, unknown> = {}) {
     const { ctx, tools } = makeCtx(services);
     const registry = new SessionRegistry(5);
-    registerTools({
+    registerBrowserTools({
       ctx: ctx as never,
       runner,
       registry,
@@ -1052,7 +1205,7 @@ describe("screenshot scratch file lifecycle", () => {
         options: {},
       },
     });
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const value = (await screenshot?.execute({}, exec)) as { image?: unknown; path: string };
     expect(value.image).toBeDefined();
     const out = calls.find((c) => c.args[0] === "screenshot")?.args.at(-1);
@@ -1064,7 +1217,7 @@ describe("screenshot scratch file lifecycle", () => {
     const { runner, calls } = outWritingRunner();
     const { tools } = setupWithRunner(runner);
     await startSession(tools);
-    const screenshot = tools.get("browser_screenshot");
+    const screenshot = tools.get("inspect.screenshot");
     const value = (await screenshot?.execute({}, makeExec())) as { image?: unknown; path: string };
     expect(value.image).toBeUndefined();
     const out = calls.find((c) => c.args[0] === "screenshot")?.args.at(-1);
@@ -1139,11 +1292,18 @@ describe("observation action instrumentation timing", () => {
       queue,
       options: { enabled: true, thumbnailIntervalMs: 1500, idleIntervalMs: 8000 },
     });
-    registerTools({ ctx: ctx as never, runner, registry, config: CONFIG, observation, queue });
+    registerBrowserTools({
+      ctx: ctx as never,
+      runner,
+      registry,
+      config: CONFIG,
+      observation,
+      queue,
+    });
 
     await startSession(tools);
-    const navigate = tools.get("browser_navigate");
-    const snapshot = tools.get("browser_snapshot");
+    const navigate = tools.get("page.navigate");
+    const snapshot = tools.get("inspect.snapshot");
     const a = navigate?.execute({ url: "http://x" }, makeExec()) as Promise<unknown>;
     // A starts immediately (empty queue) and blocks on the gate.
     await vi.waitFor(() =>

packages/i18n/src/locales/en-US/extension.json

@@ -6,27 +6,27 @@
       "disconnected": "Disconnected",
       "connecting": "Connecting…",
       "connected": "Connected",
-      "version_skew": "Protocol mismatch",
+      "version_skew": "Connected",
       "disabled": "Connection off"
     },
     "stateDetail": {
       "disconnected": "Open BrowserSkill to connect.",
       "connecting": "Connecting to BrowserSkill…",
       "connected": "Ready for automation tasks.",
-      "version_skew": "Some features may be unavailable."
+      "version_skew": "Protocol versions differ. Please upgrade."
     },
     "stateBadge": {
       "disconnected": "Standby",
       "connecting": "Igniting",
       "connected": "Ready",
-      "version_skew": "Action needed",
+      "version_skew": "Upgradable",
       "disabled": "Off"
     },
     "connectionToggleTitle": "BrowserSkill connection",
     "controlHintsToggleTitle": "Control hints",
     "controlHintsToggleHint": "Show the status pill and orange glow while the Agent controls a page.",
     "controlHintsInfoLabel": "About control hints",
-    "versionSkewWarning": "Extension protocol v{{extensionProtocol}}, daemon v{{daemonProtocol}}.",
+    "versionSkewWarning": "Extension protocol v{{extensionProtocol}}, CLI protocol v{{cliProtocol}}. Protocol versions differ — please upgrade.",
     "upgradeAvailable": "Upgradable",
     "launcher": {
       "title": "Quick actions"

packages/i18n/src/locales/zh-CN/extension.json

@@ -6,27 +6,27 @@
       "disconnected": "未连接",
       "connecting": "连接中…",
       "connected": "已连接",
-      "version_skew": "协议不一致",
+      "version_skew": "已连接",
       "disabled": "连接已关闭"
     },
     "stateDetail": {
       "disconnected": "请先打开 BrowserSkill。",
       "connecting": "正在连接 BrowserSkill…",
       "connected": "可接收自动化任务。",
-      "version_skew": "部分能力可能不可用。"
+      "version_skew": "协议版本不同，请及时升级。"
     },
     "stateBadge": {
       "disconnected": "Standby",
       "connecting": "Igniting",
       "connected": "Ready",
-      "version_skew": "Action needed",
+      "version_skew": "可升级",
       "disabled": "Off"
     },
     "connectionToggleTitle": "BrowserSkill 连接",
     "controlHintsToggleTitle": "控制提示",
     "controlHintsToggleHint": "Agent 控制页面时显示提示条和橙色闪光。",
     "controlHintsInfoLabel": "控制提示说明",
-    "versionSkewWarning": "扩展协议 v{{extensionProtocol}}，daemon 协议 v{{daemonProtocol}}。",
+    "versionSkewWarning": "扩展协议 v{{extensionProtocol}}，CLI 协议 v{{cliProtocol}}。协议版本不同，请及时升级。",
     "upgradeAvailable": "可升级",
     "launcher": {
       "title": "快捷功能"

packages/vom/src/__tests__/render.test.ts

@@ -298,6 +298,34 @@ describe("renderVom single-layer page", () => {
     expect(coreRefs(out.refs)).toEqual([{ ref: "e1", backendNodeId: 4 }]);
   });
 
+  it("renders through deeply nested transparent wrappers without exhausting the call stack", () => {
+    const wrapperCount = 10_000;
+    const nodes = [node({ id: 1, role: "RootWebArea", name: "Doc" })];
+    for (let index = 0; index < wrapperCount; index += 1) {
+      nodes.push(
+        node({
+          id: index + 2,
+          parentId: index + 1,
+          role: "generic",
+        }),
+      );
+    }
+    nodes.push(
+      node({
+        id: wrapperCount + 2,
+        parentId: wrapperCount + 1,
+        role: "button",
+        name: "Deep action",
+        tag: "button",
+      }),
+    );
+
+    const out = renderVom(scene(nodes));
+
+    expect(out.text).toContain('@e1 button "Deep action"');
+    expect(coreRefs(out.refs)).toEqual([{ ref: "e1", backendNodeId: wrapperCount + 2 }]);
+  });
+
   it("truncates when maxTokens is exceeded while keeping refs in sync with rendered text", () => {
     const out = renderVom(
       scene([
@@ -785,6 +813,55 @@ describe("renderVom single-layer page", () => {
     expect(out.text).not.toContain("北京市小客车指标调控管理信息系统]");
   });
 
+  it("derives duplicate-label context from indexed DOM ancestry without semantic proximity", () => {
+    const out = renderVom(
+      scene([
+        node({ id: 1, role: "RootWebArea", name: "Doc", contextScopeId: "root" }),
+        node({
+          id: 3,
+          parentId: 1,
+          role: "button",
+          name: "View",
+          tag: "button",
+          domParentId: 2,
+          domAncestorIds: [2, 1],
+          contextScopeId: "root",
+        }),
+        node({
+          id: 5,
+          parentId: 1,
+          role: "button",
+          name: "View",
+          tag: "button",
+          domParentId: 4,
+          domAncestorIds: [4, 1],
+          contextScopeId: "root",
+        }),
+        node({
+          id: 2,
+          parentId: 1,
+          role: "group",
+          name: "Project Alpha",
+          domParentId: 1,
+          domAncestorIds: [1],
+          contextScopeId: "root",
+        }),
+        node({
+          id: 4,
+          parentId: 1,
+          role: "group",
+          name: "Project Beta",
+          domParentId: 1,
+          domAncestorIds: [1],
+          contextScopeId: "root",
+        }),
+      ]),
+    );
+
+    expect(out.text).toContain('@e1 button "View" [ctx: Project Alpha]');
+    expect(out.text).toContain('@e2 button "View" [ctx: Project Beta]');
+  });
+
   it("does not add low-value root context to duplicated brand links", () => {
     const out = renderVom(
       scene([
@@ -833,6 +910,34 @@ describe("renderVom single-layer page", () => {
 });
 
 describe("renderVom double-layer page", () => {
+  it("collects blocking-layer descendants even when children precede parents in source order", () => {
+    const out = renderVom(
+      scene([
+        node({ id: 1, role: "RootWebArea", name: "Doc" }),
+        node({
+          id: 3,
+          parentId: 2,
+          role: "button",
+          name: "Close",
+          tag: "button",
+          paintOrder: 0,
+        }),
+        node({
+          id: 2,
+          parentId: 1,
+          role: "dialog",
+          name: "Modal",
+          rect: { x: 0, y: 0, w: 1280, h: 720 },
+          paintOrder: 10,
+          position: "fixed",
+        }),
+      ]),
+    );
+
+    expect(out.text).toContain('@e1 button "Close"');
+    expect(coreRefs(out.refs)).toEqual([{ ref: "e1", backendNodeId: 3 }]);
+  });
+
   it("renders a blocking modal first and folds the base page into an occluded summary", () => {
     const out = renderVom(
       scene([

packages/vom/src/render.ts

@@ -421,6 +421,41 @@ function buildChildren(nodes: VomNode[]): Map<number | null, VomNode[]> {
   return children;
 }
 
+const CONTEXT_LABEL_ROLES = new Set([
+  "heading",
+  "article",
+  "section",
+  "group",
+  "region",
+  "row",
+  "form",
+]);
+
+interface DomContextIndex {
+  nonBoundaryByScopeParent: Map<string | undefined, Map<number, VomNode[]>>;
+  sourceOrderById: Map<number, number>;
+}
+
+function buildDomContextIndex(nodes: VomNode[]): DomContextIndex {
+  const nonBoundaryByScopeParent = new Map<string | undefined, Map<number, VomNode[]>>();
+  const sourceOrderById = new Map<number, number>();
+  for (const [sourceOrder, node] of nodes.entries()) {
+    sourceOrderById.set(node.id, sourceOrder);
+    if (!CONTEXT_LABEL_ROLES.has(normalizedRole(node)) || isContextBoundary(node)) continue;
+    const parentId = node.domParentId;
+    if (parentId === undefined || parentId === null) continue;
+    let byParent = nonBoundaryByScopeParent.get(node.contextScopeId);
+    if (!byParent) {
+      byParent = new Map();
+      nonBoundaryByScopeParent.set(node.contextScopeId, byParent);
+    }
+    const siblings = byParent.get(parentId) ?? [];
+    siblings.push(node);
+    byParent.set(parentId, siblings);
+  }
+  return { nonBoundaryByScopeParent, sourceOrderById };
+}
+
 function duplicateReferenceNames(nodes: VomNode[]): Set<string> {
   const counts = new Map<string, number>();
   for (const node of nodes) {
@@ -440,13 +475,7 @@ function needsHandleContext(nodeName: string, duplicateRefNames: Set<string>): b
 }
 
 function contextLabel(node: VomNode, targetName: string): string | undefined {
-  if (
-    !["heading", "article", "section", "group", "region", "row", "form"].includes(
-      normalizedRole(node),
-    )
-  ) {
-    return undefined;
-  }
+  if (!CONTEXT_LABEL_ROLES.has(normalizedRole(node))) return undefined;
   return weakContextText(
     cleaned(node.name) ?? cleaned(node.text) ?? cleaned(node.nearbyText),
     targetName,
@@ -607,19 +636,45 @@ function collectDomContext(
 ): ContextCandidate[] {
   const ancestorIds = node.domAncestorIds ?? [];
   if (ancestorIds.length === 0) return [];
-  const ancestors = new Set(ancestorIds);
   const contexts: ContextCandidate[] = [];
+  const ancestorRank = new Map<number, number>();
+  for (const [rank, ancestorId] of ancestorIds.entries()) {
+    if (!ancestorRank.has(ancestorId)) ancestorRank.set(ancestorId, rank);
+  }
 
-  for (const candidate of state.nodesById.values()) {
+  const candidates = new Map<number, VomNode>();
+  const nonBoundaryByParent = state.domContextIndex.nonBoundaryByScopeParent.get(
+    node.contextScopeId,
+  );
+  for (const ancestorId of ancestorRank.keys()) {
+    const boundary = state.nodesById.get(ancestorId);
+    const boundaryParentId = boundary?.domParentId;
+    if (
+      boundary &&
+      sharesContextScope(node, boundary) &&
+      isContextBoundary(boundary) &&
+      boundaryParentId !== undefined &&
+      boundaryParentId !== null &&
+      ancestorRank.has(boundaryParentId)
+    ) {
+      candidates.set(boundary.id, boundary);
+    }
+    for (const candidate of nonBoundaryByParent?.get(ancestorId) ?? []) {
+      candidates.set(candidate.id, candidate);
+    }
+  }
+
+  const sourceOrderById = state.domContextIndex.sourceOrderById;
+  const orderedCandidates = [...candidates.values()].sort(
+    (a, b) => (sourceOrderById.get(a.id) ?? 0) - (sourceOrderById.get(b.id) ?? 0),
+  );
+  for (const candidate of orderedCandidates) {
     if (candidate.id === node.id) continue;
-    if (!sharesContextScope(node, candidate)) continue;
-    if (isContextBoundary(candidate) && !ancestorIds.includes(candidate.id)) continue;
     const parentId = candidate.domParentId;
-    if (parentId === undefined || parentId === null || !ancestors.has(parentId)) continue;
+    if (parentId === undefined || parentId === null) continue;
     const label = contextLabel(candidate, nodeName);
     if (!label) continue;
-    const rank = ancestorIds.indexOf(parentId);
-    contexts.push({ text: label, source: "dom", order: rank < 0 ? contexts.length : rank });
+    contexts.push({ text: label, source: "dom", order: ancestorRank.get(parentId) ?? 0 });
   }
 
   return contexts;
@@ -734,6 +789,7 @@ interface RenderState {
   children: Map<number | null, VomNode[]>;
   parentMap: Map<number, number | null>;
   nodesById: Map<number, VomNode>;
+  domContextIndex: DomContextIndex;
   duplicateRefNames: Set<string>;
   surfaceMap: Map<number, CondSurface>;
   scopeMap: Map<number, ActiveScopeBlock>;
@@ -768,20 +824,25 @@ function emitActiveScopeBlock(scope: ActiveScopeBlock, depth: number, state: Ren
   }
 }
 
-function renderTree(
-  children: Map<number | null, VomNode[]>,
-  parentId: number | null,
+function pushRenderChildren(
+  stack: Array<{ node: VomNode; depth: number }>,
+  children: readonly VomNode[],
   depth: number,
-  state: RenderState,
 ): void {
-  if (state.stopped) return;
+  for (let index = children.length - 1; index >= 0; index -= 1) {
+    stack.push({ node: children[index], depth });
+  }
+}
 
-  const nodes = children.get(parentId) ?? [];
-  for (const node of nodes) {
-    if (state.stopped) return;
+function renderTree(children: Map<number | null, VomNode[]>, state: RenderState): void {
+  const stack: Array<{ node: VomNode; depth: number }> = [];
+  pushRenderChildren(stack, children.get(null) ?? [], 1);
+
+  while (stack.length > 0 && !state.stopped) {
+    const { node, depth } = stack.pop() as { node: VomNode; depth: number };
 
     if (!shouldRender(node)) {
-      renderTree(children, node.id, depth, state);
+      pushRenderChildren(stack, children.get(node.id) ?? [], depth);
       continue;
     }
 
@@ -798,7 +859,7 @@ function renderTree(
     if (nextTokens > state.maxTokens) {
       state.truncated = true;
       state.stopped = true;
-      return;
+      break;
     }
 
     state.lines.push(line);
@@ -823,7 +884,7 @@ function renderTree(
     if ((ref && shouldSkipRedundantRefChildren(node)) || shouldSkipRedundantChildren(node, state)) {
       continue;
     }
-    renderTree(children, node.id, depth + 1, state);
+    pushRenderChildren(stack, children.get(node.id) ?? [], depth + 1);
   }
 }
 
@@ -848,12 +909,13 @@ function renderNodes(
     children,
     parentMap: buildParentMap(nodes),
     nodesById: new Map(nodes.map((node) => [node.id, node])),
+    domContextIndex: buildDomContextIndex(nodes),
     duplicateRefNames: duplicateReferenceNames(nodes),
     surfaceMap: new Map(surfaces.map((surface) => [surface.triggerId, surface])),
     scopeMap: new Map(activeScopeBlocks.map((scope) => [scope.triggerId, scope])),
   };
 
-  renderTree(children, null, 1, state);
+  renderTree(children, state);
 
   return state;
 }
@@ -922,23 +984,33 @@ function paintLineageByFrame(
   return lineage;
 }
 
+function cachedPaintLineageByFrame(
+  node: VomNode,
+  parentMap: Map<number, number | null>,
+  nodesById: Map<number, VomNode>,
+  cache: Map<number, Map<string | undefined, VomNode>>,
+): Map<string | undefined, VomNode> {
+  const cached = cache.get(node.id);
+  if (cached) return cached;
+  const lineage = paintLineageByFrame(node, parentMap, nodesById);
+  cache.set(node.id, lineage);
+  return lineage;
+}
+
 /** Resolve both nodes to the nearest document where their paint order is comparable. */
 function comparablePaintNodes(
   target: VomNode,
   blocker: VomNode,
   parentMap: Map<number, number | null>,
   nodesById: Map<number, VomNode>,
+  lineageCache: Map<number, Map<string | undefined, VomNode>>,
 ): { target: VomNode; blocker: VomNode } | null {
   if (target.frameId === blocker.frameId) return { target, blocker };
-  const targetLineage = paintLineageByFrame(target, parentMap, nodesById);
-  let blockerInFrame: VomNode | undefined = blocker;
-  let guard = 0;
-  while (blockerInFrame && guard <= parentMap.size) {
-    const targetInFrame = targetLineage.get(blockerInFrame.frameId);
+  const targetLineage = cachedPaintLineageByFrame(target, parentMap, nodesById, lineageCache);
+  const blockerLineage = cachedPaintLineageByFrame(blocker, parentMap, nodesById, lineageCache);
+  for (const [frameId, blockerInFrame] of blockerLineage) {
+    const targetInFrame = targetLineage.get(frameId);
     if (targetInFrame) return { target: targetInFrame, blocker: blockerInFrame };
-    const parentId: number | null = parentMap.get(blockerInFrame.id) ?? null;
-    blockerInFrame = parentId === null ? undefined : nodesById.get(parentId);
-    guard += 1;
   }
   return null;
 }
@@ -949,12 +1021,13 @@ function isBlockedByRegion(
   parentMap: Map<number, number | null>,
   nodesById: Map<number, VomNode>,
   viewportCoverage: number,
+  lineageCache: Map<number, Map<string | undefined, VomNode>>,
 ): boolean {
   if (target.id === blocker.id) return false;
   if (isAncestorOf(parentMap, blocker.id, target.id)) return false;
   if (isAncestorOf(parentMap, target.id, blocker.id)) return false;
   if (!target.rect || !blocker.rect) return false;
-  const paintNodes = comparablePaintNodes(target, blocker, parentMap, nodesById);
+  const paintNodes = comparablePaintNodes(target, blocker, parentMap, nodesById, lineageCache);
   if (!paintNodes) return false;
 
   const points = interactionPoints(target.rect);
@@ -979,17 +1052,22 @@ function isBlockedByRegion(
   return points.some(([x, y]) => rectContains(blocker.rect as Rect, x, y));
 }
 
-function collectDescendantsOfIds(nodes: VomNode[], roots: Set<number>): Set<number> {
+function collectDescendants(nodes: VomNode[], roots: Set<number>): Set<number> {
+  const children = new Map<number, number[]>();
+  for (const node of nodes) {
+    if (node.parentId === null) continue;
+    const siblings = children.get(node.parentId) ?? [];
+    siblings.push(node.id);
+    children.set(node.parentId, siblings);
+  }
+
   const included = new Set(roots);
-  let changed = true;
-  while (changed) {
-    changed = false;
-    for (const node of nodes) {
-      if (included.has(node.id)) continue;
-      if (node.parentId !== null && included.has(node.parentId)) {
-        included.add(node.id);
-        changed = true;
-      }
+  const queue = [...roots];
+  for (let index = 0; index < queue.length; index += 1) {
+    for (const childId of children.get(queue[index]) ?? []) {
+      if (included.has(childId)) continue;
+      included.add(childId);
+      queue.push(childId);
     }
   }
   return included;
@@ -998,6 +1076,7 @@ function collectDescendantsOfIds(nodes: VomNode[], roots: Set<number>): Set<numb
 function applyActiveRegionPolicy(nodes: VomNode[], scene: VomScene): VomNode[] {
   const parentMap = buildParentMap(nodes);
   const nodesById = new Map(nodes.map((node) => [node.id, node]));
+  const lineageCache = new Map<number, Map<string | undefined, VomNode>>();
   const candidates = nodes
     .filter(isPositionedRegionCandidate)
     .map((node) => {
@@ -1014,40 +1093,29 @@ function applyActiveRegionPolicy(nodes: VomNode[], scene: VomScene): VomNode[] {
   for (const target of nodes) {
     if (!isVomReferenceNode(target)) continue;
     const blocked = candidates.some((candidate) =>
-      isBlockedByRegion(target, candidate.node, parentMap, nodesById, candidate.viewportCoverage),
+      isBlockedByRegion(
+        target,
+        candidate.node,
+        parentMap,
+        nodesById,
+        candidate.viewportCoverage,
+        lineageCache,
+      ),
     );
     if (blocked) blockedRoots.add(target.id);
   }
 
   if (blockedRoots.size === 0) return nodes;
-  const blocked = collectDescendantsOfIds(nodes, blockedRoots);
+  const blocked = collectDescendants(nodes, blockedRoots);
   return nodes.filter((node) => !blocked.has(node.id));
 }
 
-function collectDescendantsOfMembers(nodes: VomNode[], members: Set<number>): Set<number> {
-  const included = new Set(members);
-  let changed = true;
-
-  while (changed) {
-    changed = false;
-    for (const node of nodes) {
-      if (included.has(node.id)) continue;
-      if (node.parentId !== null && included.has(node.parentId)) {
-        included.add(node.id);
-        changed = true;
-      }
-    }
-  }
-
-  return included;
-}
-
 function countRenderable(nodes: VomNode[]): number {
   return nodes.filter(shouldRender).length;
 }
 
 function renderDoubleLayer(scene: VomScene, layer: BlockingLayer, options: VomOptions): VomResult {
-  const included = collectDescendantsOfMembers(scene.nodes, layer.members);
+  const included = collectDescendants(scene.nodes, layer.members);
   const visibleNodes = scene.nodes.filter((node) => included.has(node.id));
   const hiddenCount = countRenderable(scene.nodes.filter((node) => !included.has(node.id)));
   const header = [

skill/SKILL.md

@@ -29,7 +29,7 @@ Drive the user's **real Chromium browser** (with their logins and cookies) throu
 ## Prerequisites
 
 1. `bsk` on `PATH` (Rust CLI from browser-skill)
-2. browser-skill **extension** loaded in Chromium and connected (popup shows green)
+2. browser-skill **extension** loaded in Chromium and connected (popup shows green, or an upgrade reminder — still connected; continue)
 3. Any `bsk` command auto-starts background services as needed; use `bsk doctor` if anything fails
 
 ## Mandatory workflow
@@ -210,7 +210,7 @@ Both capture from the moment the tab is attached and read a bounded per-tab buff
 
 | Command | Summary |
 |---------|---------|
-| `bsk evaluate <expression>` | Run JS in agent tab (see red lines); JS throw → stderr, **exit 0** |
+| `bsk evaluate <expression>` | Run JS in agent tab (see red lines). JS throw → stderr, **exit 0** (RPC success); use `--json` and check `.ok` to detect JS errors |
 | `bsk wait-for-navigation` | Block until load/DOM idle/etc. (`--wait-until`, `--timeout`) |
 | `bsk wait-ms <duration>` | Sleep (`500ms`, `2s`, `1m`; **no** `--session`) |
 
@@ -294,12 +294,12 @@ bsk record stop [--output trace]   # terminal fallback if the browser panel is u
 
 | Code | Meaning | What to do |
 |------|---------|------------|
-| `0` | Success (including `evaluate` where JS threw but RPC succeeded) | Continue |
+| `0` | Success (including `evaluate` where JS threw but RPC succeeded) | Continue; for `evaluate`, check `--json` `.ok` to distinguish JS success from RPC success |
 | `1` | User error — bad args, unknown session, tab not in Agent Window, stale ref | Fix args; `bsk session list`; re-snapshot |
 | `2` | Protocol / transport — service unreachable, IPC failure | `bsk doctor`; check extension connected; retry the command |
 | `3` | Browser / CDP execution failed | Retry; simplify selector; check tab still open |
 | `4` | Timeout | Increase `--timeout`; try `--wait-until domcontentloaded` |
-| `5` | Version skew (CLI vs extension) | Upgrade/reinstall matching versions |
+| `5` | Unknown RPC method | This command is not implemented in the current build. Continue the task with other commands. If a newly added command is missing, suggest `bsk update`. |
 
 Human errors print `error:` + `hint:` on stderr; `--json` includes `code`, `message`, `hint`, `exit_code`.