Files
BrowserSkill/packages/dsh-plugin-browserskill/tests/dispose.test.ts
BB-fat ce11325de2 feat(dsh-plugin-browserskill): reveal tool schemas only after skill invocation
Progressive disclosure, final stage: with lazyTools on (default), the eleven
browser_* tool schemas stay out of the system prompt — the browser-skill
catalog entry is the sole advertisement. One successful skill invocation
(model tool call via tools/result, or a /browser-skill user gesture via the
session append feed) registers the suite for the process lifetime, idempotent
on repeats; session resume is covered by scanning durable events for a past
tool/call+tool/result pair (or skill-invocation message) on session entry and
at apply time. lazyTools: false keeps the legacy always-on registration.

Investigation (recorded in the ticket): tools/result carries normalized
name/arguments with an isError discriminant; dsh 0.1 has no official global
visibility switch (ctx.tools.restrict is agent-scoped and throws from a plain
context), making conditional registration the intended pattern; per-step
assembly picks the new suite up on the next step, and visibility changes are
first-class inputs of tool-skill's catalog digest. registerTools now returns
the combined disposer via a prototype-preserving tracking overlay; the skill
prelude tells the model the browser_* tools unlock on skill load.

Tests: +8 (hidden-before/reveal/idempotent/gesture/history hit+miss/dispose/
scanner pairing/two-state apply wiring). 124/124 green.
2026-08-14 16:43:51 +00:00

95 lines
3.4 KiB
TypeScript

// Unload-cleanup ownership discipline: dispose must stop exactly the sessions
// this plugin created — never referenced/foreign sessions on the shared
// daemon — and a stale handle (already stopped elsewhere, daemon restarted)
// must not abort the remaining stops.
import type { ToolDefinition, ToolRunContext } from "@deepseek-ai/dsh-tools";
import { describe, expect, it } from "vitest";
import { apply } from "../src/index";
import type { BskRunOptions, BskRunResult } from "../src/runner";
interface FakeCall {
args: string[];
}
function ok(payload: unknown): BskRunResult {
return { code: 0, stdout: JSON.stringify(payload), stderr: "", timedOut: false, aborted: false };
}
function fail(message: string): BskRunResult {
return {
code: 4,
stdout: JSON.stringify({ code: "session_not_found", message }),
stderr: "",
timedOut: false,
aborted: false,
};
}
function makeExec(): ToolRunContext {
return {
callId: "t",
name: "t",
signal: new AbortController().signal,
} as unknown as ToolRunContext;
}
describe("dispose cleanup ownership", () => {
it("stops only plugin-owned sessions and tolerates stale handles", async () => {
const calls: FakeCall[] = [];
const startReplies = [
{ session_id: "own1", browser_instance_id: "b1" },
{ session_id: "own2", browser_instance_id: "b1" },
];
const runner = {
async run(args: string[], _options: BskRunOptions = {}): Promise<BskRunResult> {
calls.push({ args });
const joined = args.join(" ");
if (joined.startsWith("status")) return ok({});
if (joined.startsWith("session start")) return ok(startReplies.shift());
if (joined.startsWith("snapshot")) {
return ok({ text: "x", ref_count: 1, tab_id: 7, truncated: false });
}
if (joined === "session stop own1") return ok({});
if (joined === "session stop own2") {
// Stale handle: already stopped by someone else / daemon restarted.
return fail("session not found");
}
return fail(`unexpected command: ${joined}`);
},
killAll() {},
killFor: () => 0,
};
const tools = new Map<string, ToolDefinition>();
const disposers: Array<() => Promise<void>> = [];
const ctx = {
tools: { register: (def: ToolDefinition) => tools.set(def.name, def) },
get: () => undefined,
// No webServer in this harness: the inject callback never runs.
inject: () => {},
effect: (fn: () => () => Promise<void>) => {
disposers.push(fn());
},
};
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());
// 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 Promise.all(disposers.map((dispose) => dispose()));
const stops = calls
.map((call) => call.args.join(" "))
.filter((joined) => joined.startsWith("session stop"));
expect(stops.sort()).toEqual(["session stop own1", "session stop own2"]);
expect(stops.some((joined) => joined.includes("ext9"))).toBe(false);
});
});