fix(dsh-plugin-browserskill): address review findings (leaks, trust fence, queue race)

Blocking: - B1: delete screenshot scratch PNGs — observation frames reuse one fixed per-session path and unlink after every read (finally); browser_screenshot unlinks once the bytes are in the attachment store (kept only when the file itself is the model-facing artifact). - B2: replicate dsh's browser-trust fence on /bsk-observation/*: loopback Host only, Origin must match Host, sec-fetch-site: cross-site refused, POST requires application/json; README documents the loopback trust premise and the 0.0.0.0 warning. Major: - M1: KeyedExecutor tail chains previous+task (allSettled) so a task aborted while queued cannot release the next one into the running session. - M2: client tracks the live frame per session — replacing/removing/reset revokes the old blob URL at once; loads settling after replacement never resurrect it. Minor: drop replaced thumbRefs (+on remove); beginAction fires inside the queue (no label overwrite while queued, no idle flash); half-initialized session cleanup stops through the queue with one retry; SSE (re)open refetches /state (+ snapshot flag renamed subscribed); install probe uses --version (no daemon spawn); emulate mobile documents the width+height requirement (the daemon refuses it alone — verified). Nit: single BskRunOptions declaration; tails map entry dropped on drain; SSE cleanup on res close; popOut catches requestWindow rejections; resize handle gains keyboard control (arrows, 16px steps, aria value attrs). Tests: +14 (fence rules, scratch lifecycle, queue race, blob revocation, instrumentation timing, SSE resync). 113/113 green.
2026-09-10 13:01:23 +08:00 · 2026-08-14 11:56:02 +00:00
parent 52aa356153
commit a4e668542d
13 changed files with 784 additions and 104 deletions
@@ -111,3 +111,87 @@ describe("ObservationClientStore", () => {
    expect(store.getSnapshot().sessions).toHaveLength(0);
  });
 });
+
+describe("thumbnail blob URL lifecycle", () => {
+  function withRevokeSpy() {
+    const revoke = vi.fn();
+    (URL as unknown as { revokeObjectURL: unknown }).revokeObjectURL = revoke;
+    return revoke;
+  }
+
+  it("replacing a frame revokes the old blob URL and drops its entry", async () => {
+    const revoke = withRevokeSpy();
+    const { store, esInstances } = harness([{ ...OBS_IDLE, thumbnailAttachmentId: "a1" }]);
+    store.start();
+    await vi.waitFor(() => expect(store.getSnapshot().sessions).toHaveLength(1));
+    store.ensureThumbnail("a1");
+    await vi.waitFor(() =>
+      expect(store.getSnapshot().thumbnails["a1"]).toEqual({ status: "ready", url: "blob:url-a1" }),
+    );
+    emit(esInstances[0], {
+      type: "upsert",
+      session: { ...OBS_IDLE, thumbnailAttachmentId: "a2" },
+    });
+    expect(store.getSnapshot().thumbnails["a1"]).toBeUndefined();
+    expect(revoke).toHaveBeenCalledWith("blob:url-a1");
+    store.stop();
+  });
+
+  it("remove and reset revoke the session's tracked frame", async () => {
+    const revoke = withRevokeSpy();
+    const { store, esInstances } = harness([{ ...OBS_IDLE, thumbnailAttachmentId: "a1" }]);
+    store.start();
+    await vi.waitFor(() => expect(store.getSnapshot().sessions).toHaveLength(1));
+    store.ensureThumbnail("a1");
+    await vi.waitFor(() => expect(store.getSnapshot().thumbnails["a1"].status).toBe("ready"));
+    emit(esInstances[0], {
+      type: "remove",
+      session: { sessionId: "s1", action: "idle", since: 0 },
+    });
+    expect(revoke).toHaveBeenCalledWith("blob:url-a1");
+    expect(store.getSnapshot().thumbnails["a1"]).toBeUndefined();
+    store.stop();
+  });
+
+  it("a load that finishes after replacement never resurrects the old frame", async () => {
+    const revoke = withRevokeSpy();
+    const { store, esInstances, loadImage } = harness([
+      { ...OBS_IDLE, thumbnailAttachmentId: "a1" },
+    ]);
+    let resolveA1!: (url: string) => void;
+    loadImage.mockImplementation((id: string) =>
+      id === "a1"
+        ? new Promise<string>((resolve) => {
+            resolveA1 = resolve;
+          })
+        : Promise.resolve(`blob:url-${id}`),
+    );
+    store.start();
+    await vi.waitFor(() => expect(store.getSnapshot().sessions).toHaveLength(1));
+    store.ensureThumbnail("a1");
+    // the frame is replaced while a1 is still loading
+    emit(esInstances[0], {
+      type: "upsert",
+      session: { ...OBS_IDLE, thumbnailAttachmentId: "a2" },
+    });
+    resolveA1("blob:url-a1-late");
+    await new Promise((resolve) => setImmediate(resolve));
+    expect(store.getSnapshot().thumbnails["a1"]).toBeUndefined();
+    expect(revoke).toHaveBeenCalledWith("blob:url-a1-late");
+    store.stop();
+  });
+});
+
+describe("SSE (re)open resync", () => {
+  it("refetches the full state when the stream (re)opens", async () => {
+    const { store, fetches, esInstances } = harness([OBS_IDLE]);
+    store.start();
+    await vi.waitFor(() => expect(store.getSnapshot().sessions).toHaveLength(1));
+    const before = fetches.filter((f) => f.url === "/bsk-observation/state").length;
+    esInstances[0].onopen?.();
+    await vi.waitFor(() =>
+      expect(fetches.filter((f) => f.url === "/bsk-observation/state").length).toBe(before + 1),
+    );
+    store.stop();
+  });
+});
@@ -2,7 +2,7 @@
 // traffic isolation, interrupt routing, the owned boundary, and the HTTP/SSE
 // interface. All bsk runs are faked; the scheduler is a manual timer queue.

-import { mkdtempSync, writeFileSync } from "node:fs";
+import { existsSync, mkdtempSync, writeFileSync } from "node:fs";
 import type { IncomingMessage, ServerResponse } from "node:http";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
@@ -344,6 +344,7 @@ describe("HTTP/SSE interface", () => {

  function fakeRes() {
    const chunks: string[] = [];
+    const closeHandlers: (() => void)[] = [];
    const res = {
      status: 0,
      writeHead(status: number) {
@@ -356,14 +357,35 @@ describe("HTTP/SSE interface", () => {
      end(chunk?: string) {
        if (chunk !== undefined) chunks.push(chunk);
      },
+      on(event: string, fn: () => void) {
+        if (event === "close") closeHandlers.push(fn);
+      },
+      close: () => closeHandlers.forEach((fn) => fn()),
      body: () => chunks.join(""),
    };
    return {
-      res: res as unknown as ServerResponse & { status: number; body: () => string },
+      res: res as unknown as ServerResponse & {
+        status: number;
+        body: () => string;
+        close: () => void;
+      },
      chunks,
    };
  }

+  /** Loopback GET/POST request stubs the fence accepts. */
+  function fakeReq(overrides: Record<string, unknown> = {}): IncomingMessage {
+    return {
+      method: "GET",
+      headers: { host: "127.0.0.1:3999" },
+      on: (event: string, fn: (chunk?: string) => void) => {
+        if (event === "data") fn(JSON.stringify({ sessionId: "s1" }));
+        if (event === "end") fn();
+      },
+      ...overrides,
+    } as unknown as IncomingMessage;
+  }
+
  it("serves state, streams events, and routes interrupt", async () => {
    const registry = new SessionRegistry(5);
    own(registry, "s1");
@@ -379,35 +401,25 @@ describe("HTTP/SSE interface", () => {
    const stateRoute = routes.get("/bsk-observation/state");
    expect(stateRoute).toBeDefined();
    const stateRes = fakeRes();
-    await stateRoute?.handler({ method: "GET" } as IncomingMessage, stateRes.res as never);
+    await stateRoute?.handler(fakeReq(), stateRes.res as never);
    expect(stateRes.res.status).toBe(200);
    expect(JSON.parse(stateRes.res.body()).sessions[0].sessionId).toBe("s1");

    // events (SSE): one write per service event
    const eventsRoute = routes.get("/bsk-observation/events");
    const eventsRes = fakeRes();
-    const closeHandlers: (() => void)[] = [];
-    const req = {
-      method: "GET",
-      on: (event: string, fn: () => void) => {
-        if (event === "close") closeHandlers.push(fn);
-      },
-    } as unknown as IncomingMessage;
-    await eventsRoute?.handler(req, eventsRes.res as never);
+    await eventsRoute?.handler(fakeReq(), eventsRes.res as never);
    service.beginAction("s1", "clicking");
    expect(eventsRes.res.body()).toContain('"action":"clicking"');
-    closeHandlers.forEach((fn) => fn());
+    eventsRes.res.close();

    // interrupt (POST {sessionId})
    const interruptRoute = routes.get("/bsk-observation/interrupt");
    const interruptRes = fakeRes();
-    const postReq = {
+    const postReq = fakeReq({
      method: "POST",
-      on: (event: string, fn: (chunk?: string) => void) => {
-        if (event === "data") fn(JSON.stringify({ sessionId: "s1" }));
-        if (event === "end") fn();
-      },
-    } as unknown as IncomingMessage;
+      headers: { host: "127.0.0.1:3999", "content-type": "application/json" },
+    });
    await interruptRoute?.handler(postReq, interruptRes.res as never);
    expect(JSON.parse(interruptRes.res.body())).toEqual({ interrupted: true });

@@ -416,6 +428,57 @@ describe("HTTP/SSE interface", () => {
    service.dispose();
  });

+  it("fences off non-loopback, cross-site, and simple-request traffic", async () => {
+    const registry = new SessionRegistry(5);
+    own(registry, "s1");
+    const { service } = setup({ registry });
+    service.addSession("s1");
+    const { routes, webServer } = routeHarness();
+    const ctx = { get: (key: string) => (key === "webServer" ? webServer : undefined) } as never;
+    const dispose = registerObservationRoutes(ctx, service);
+    const stateRoute = routes.get("/bsk-observation/state");
+    const interruptRoute = routes.get("/bsk-observation/interrupt");
+
+    const getState = async (headers: Record<string, string>) => {
+      const res = fakeRes();
+      await stateRoute?.handler(fakeReq({ headers }), res.res as never);
+      return res.res.status;
+    };
+
+    // Host must be a loopback authority.
+    expect(await getState({ host: "192.168.1.20:3999" })).toBe(403);
+    expect(await getState({ host: "attacker.example" })).toBe(403);
+    expect(await getState({ host: "localhost:3999" })).toBe(200);
+    expect(await getState({ host: "[::1]:3999" })).toBe(200);
+    // A present Origin must match the Host (DNS-rebinding reads die here).
+    expect(await getState({ host: "127.0.0.1:3999", origin: "http://evil.example" })).toBe(403);
+    expect(await getState({ host: "127.0.0.1:3999", origin: "http://127.0.0.1:3999" })).toBe(200);
+    // Explicitly cross-site fetches are refused even without an Origin.
+    expect(await getState({ host: "127.0.0.1:3999", "sec-fetch-site": "cross-site" })).toBe(403);
+    expect(await getState({ host: "127.0.0.1:3999", "sec-fetch-site": "same-origin" })).toBe(200);
+
+    // POST must be application/json — a cross-site simple request can never comply.
+    const post = async (contentType: string | undefined) => {
+      const res = fakeRes();
+      const req = fakeReq({
+        method: "POST",
+        headers: {
+          host: "127.0.0.1:3999",
+          ...(contentType !== undefined ? { "content-type": contentType } : {}),
+        },
+      });
+      await interruptRoute?.handler(req, res.res as never);
+      return res.res.status;
+    };
+    expect(await post("text/plain")).toBe(403);
+    expect(await post("application/x-www-form-urlencoded")).toBe(403);
+    expect(await post(undefined)).toBe(403);
+    expect(await post("application/json; charset=utf-8")).toBe(200);
+
+    dispose();
+    service.dispose();
+  });
+
  it("registers nothing when no webServer is mounted", () => {
    const { service } = setup({});
    const ctx = { get: () => undefined } as never;
@@ -434,7 +497,12 @@ describe("availability and dead sessions", () => {
    service.addSession("s1");
    for (let i = 1; i <= 3; i++) {
      scheduler.runNext();
-      await waitFor(() => runner.calls.filter((c) => c.args[0] === "screenshot").length === i);
+      // the frame's scratch-file unlink lands before the next timer is queued
+      await waitFor(
+        () =>
+          runner.calls.filter((c) => c.args[0] === "screenshot").length === i &&
+          scheduler.pending().length === 1,
+      );
    }
    expect(service.isAvailable()).toBe(false);
    expect(events.some((e) => e.type === "availability" && e.available === false)).toBe(true);
@@ -498,3 +566,79 @@ describe("actionForLabel", () => {
    expect(actionForLabel("session start")).toBe("starting");
  });
 });
+
+describe("scratch files and thumbnail references", () => {
+  it("reuses one scratch path per session and deletes it after reading", async () => {
+    const scheduler = fakeScheduler();
+    const runner = fakeRunner();
+    const { service } = setup({ runner, scheduler });
+    service.addSession("s1");
+    // first frame (scheduled with delay 0)
+    scheduler.runNext();
+    await waitFor(
+      () =>
+        service.getState()[0]?.thumbnailAttachmentId !== undefined &&
+        scheduler.pending().length === 1,
+    );
+    const firstOut = runner.calls.filter((c) => c.args[0] === "screenshot").at(-1)?.args;
+    const pathOf = (args: string[]) => args[args.indexOf("--out") + 1];
+    expect(pathOf(firstOut ?? [])).toBe(join(tmpdir(), "bsk-obs-s1.png"));
+    expect(existsSync(pathOf(firstOut ?? []))).toBe(false);
+    // second frame: same path, deleted again
+    service.endAction("s1");
+    scheduler.runNext();
+    await waitFor(
+      () =>
+        runner.calls.filter((c) => c.args[0] === "screenshot").length >= 2 &&
+        scheduler.pending().length === 1,
+    );
+    const secondOut = runner.calls.filter((c) => c.args[0] === "screenshot").at(-1)?.args;
+    expect(pathOf(secondOut ?? [])).toBe(join(tmpdir(), "bsk-obs-s1.png"));
+    await waitFor(() => !existsSync(pathOf(secondOut ?? [])));
+    service.dispose();
+  });
+
+  it("drops the replaced frame reference and clears it on remove", async () => {
+    let counter = 0;
+    const attachments = {
+      saveImage: async (input: { data: Uint8Array; name?: string }) => {
+        counter += 1;
+        return {
+          attachmentId: `att-${counter}`,
+          mediaType: "image/png",
+          bytes: input.data.byteLength,
+          width: 4,
+          height: 4,
+          name: input.name,
+        };
+      },
+      readImage: async (ref: { attachmentId: unknown }) => ({
+        data: new Uint8Array([1, 2, 3]),
+        attachment: ref,
+      }),
+    };
+    const scheduler = fakeScheduler();
+    const { service } = setup({ scheduler, attachments });
+    service.addSession("s1");
+    scheduler.runNext();
+    await waitFor(
+      () =>
+        service.getState()[0]?.thumbnailAttachmentId === "att-1" &&
+        scheduler.pending().length === 1,
+    );
+    service.endAction("s1");
+    scheduler.runNext();
+    await waitFor(
+      () =>
+        service.getState()[0]?.thumbnailAttachmentId === "att-2" &&
+        scheduler.pending().length === 1,
+    );
+    // the replaced frame's read path is gone
+    expect(await service.readThumbnail("att-1")).toBeUndefined();
+    expect((await service.readThumbnail("att-2"))?.mediaType).toBe("image/png");
+    // removing the session clears its live reference too
+    service.removeSession("s1");
+    expect(await service.readThumbnail("att-2")).toBeUndefined();
+    service.dispose();
+  });
+});
@@ -0,0 +1,85 @@
+// KeyedExecutor: per-key FIFO guarantees, including the abort-while-queued
+// race that used to release the tail early (letting a later task overlap the
+// one still running on the same session).
+
+import { describe, expect, it } from "vitest";
+import { KeyedExecutor } from "../src/queue";
+
+describe("KeyedExecutor", () => {
+  it("runs same-key tasks in FIFO order", async () => {
+    const queue = new KeyedExecutor();
+    const order: string[] = [];
+    const gate = new Promise<void>((resolve) => setTimeout(resolve, 5));
+    await Promise.all([
+      queue.run("k", async () => {
+        await gate;
+        order.push("a");
+      }),
+      queue.run("k", async () => {
+        order.push("b");
+      }),
+      queue.run("k", async () => {
+        order.push("c");
+      }),
+    ]);
+    expect(order).toEqual(["a", "b", "c"]);
+  });
+
+  it("a task aborted while queued does not let the next one overlap the runner", async () => {
+    const queue = new KeyedExecutor();
+    let releaseA!: () => void;
+    const aRuns = new Promise<void>((resolve) => {
+      releaseA = resolve;
+    });
+    const events: string[] = [];
+
+    // A holds the session until releaseA.
+    const a = queue.run("s1", async () => {
+      events.push("a:start");
+      await aRuns;
+      events.push("a:end");
+    });
+    // B is queued behind A and aborted before A finishes.
+    const abortB = new AbortController();
+    const b = queue.run(
+      "s1",
+      async () => {
+        events.push("b:start");
+      },
+      abortB.signal,
+    );
+    // C is queued behind B.
+    const c = queue.run("s1", async () => {
+      events.push("c:start");
+    });
+
+    abortB.abort();
+    await expect(b).rejects.toMatchObject({ name: "AbortError" });
+    // Give C every chance to (wrongly) start while A is still in flight.
+    await new Promise((resolve) => setTimeout(resolve, 20));
+    expect(events).toEqual(["a:start"]);
+
+    releaseA();
+    await a;
+    await c;
+    expect(events).toEqual(["a:start", "a:end", "c:start"]);
+  });
+
+  it("a failed task never strands the queue", async () => {
+    const queue = new KeyedExecutor();
+    await expect(
+      queue.run("k", async () => {
+        throw new Error("boom");
+      }),
+    ).rejects.toThrow("boom");
+    await expect(queue.run("k", async () => "ok")).resolves.toBe("ok");
+  });
+
+  it("drops the tail entry once the queue drains", async () => {
+    const queue = new KeyedExecutor();
+    const internals = queue as unknown as { tails: Map<string, Promise<void>> };
+    await queue.run("k", async () => {});
+    await new Promise((resolve) => setImmediate(resolve));
+    expect(internals.tails.size).toBe(0);
+  });
+});
@@ -1,8 +1,8 @@
-import { mkdtempSync, writeFileSync } from "node:fs";
+import { existsSync, mkdtempSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import type { ToolDefinition, ToolRunContext } from "@deepseek-ai/dsh-tools";
-import { describe, expect, it } from "vitest";
+import { describe, expect, it, vi } from "vitest";
 import { ObservationService } from "../src/observation";
 import { KeyedExecutor } from "../src/queue";
 import type { BskRunner, BskRunOptions, BskRunResult } from "../src/runner";
@@ -621,3 +621,180 @@ describe("presentation", () => {
    expect(view).toMatchObject({ card: "terminal", output: "[session s1] screenshot of tab 7" });
  });
 });
+
+describe("screenshot scratch file lifecycle", () => {
+  /** Runner that honors --out like the real bsk (writes the PNG there). */
+  function outWritingRunner() {
+    const calls: FakeCall[] = [];
+    return {
+      calls,
+      runner: {
+        async run(args: string[], options: BskRunOptions = {}): Promise<BskRunResult> {
+          calls.push({ args, options });
+          if (args[0] === "session") {
+            return {
+              code: 0,
+              stdout: JSON.stringify({ session_id: "s1", browser_instance_id: "chrome-1" }),
+              stderr: "",
+              timedOut: false,
+              aborted: false,
+            };
+          }
+          if (args[0] === "screenshot") {
+            const out = args[args.indexOf("--out") + 1];
+            writeFileSync(out, Buffer.from([0x89, 0x50, 0x4e, 0x47]));
+            return {
+              code: 0,
+              stdout: JSON.stringify({
+                tab_id: 7,
+                width: 4,
+                height: 4,
+                format: "png",
+                path: out,
+                byte_size: 4,
+              }),
+              stderr: "",
+              timedOut: false,
+              aborted: false,
+            };
+          }
+          return { code: 2, stdout: "{}", stderr: "", timedOut: false, aborted: false };
+        },
+        killAll() {},
+        killFor: () => 0,
+      } as BskRunner,
+    };
+  }
+
+  function setupWithRunner(runner: BskRunner, services: Record<string, unknown> = {}) {
+    const { ctx, tools } = makeCtx(services);
+    const registry = new SessionRegistry(5);
+    registerTools({
+      ctx: ctx as never,
+      runner,
+      registry,
+      config: CONFIG,
+      observation: disabledObservation({ ctx, runner, registry }),
+      queue: new KeyedExecutor(),
+    });
+    return { tools, registry };
+  }
+
+  const IMAGE_SERVICES = {
+    attachments: {
+      imageLimits: { mediaTypes: ["image/png"], maxImageBytes: 1e7, maxMessageImageBytes: 1e7 },
+      saveImage: async (input: { data: Uint8Array; mediaType: string }) => ({
+        attachmentId: "att-1",
+        mediaType: input.mediaType,
+        bytes: input.data.byteLength,
+        width: 4,
+        height: 4,
+      }),
+    },
+    llm: { resolveModelInfo: async () => ({ inputModalities: ["text", "image"] }) },
+  };
+
+  it("deletes the scratch PNG once the bytes are in the attachment store", async () => {
+    const { runner, calls } = outWritingRunner();
+    const { tools } = setupWithRunner(runner, IMAGE_SERVICES);
+    await startSession(tools);
+    const exec = makeExec({
+      agent: {
+        session: { requestHeader: () => ({ config: { provider: "p", model: "vl" } }) },
+        options: {},
+      },
+    });
+    const screenshot = tools.get("browser_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);
+    expect(out).toBeDefined();
+    expect(existsSync(out ?? "")).toBe(false);
+  });
+
+  it("keeps the PNG only when it is the model-facing artifact (no store)", async () => {
+    const { runner, calls } = outWritingRunner();
+    const { tools } = setupWithRunner(runner);
+    await startSession(tools);
+    const screenshot = tools.get("browser_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);
+    expect(existsSync(out ?? "")).toBe(true);
+  });
+});
+
+describe("observation action instrumentation timing", () => {
+  it("a queued call never overwrites the in-flight action label", async () => {
+    const { ctx, tools } = makeCtx();
+    const registry = new SessionRegistry(5);
+    let releaseNavigate!: () => void;
+    const navigateGate = new Promise<void>((resolve) => {
+      releaseNavigate = resolve;
+    });
+    const runner: BskRunner = {
+      async run(args: string[]): Promise<BskRunResult> {
+        if (args[0] === "session") {
+          return {
+            code: 0,
+            stdout: JSON.stringify({ session_id: "s1", browser_instance_id: "chrome-1" }),
+            stderr: "",
+            timedOut: false,
+            aborted: false,
+          };
+        }
+        if (args[0] === "navigate") {
+          await navigateGate;
+          return {
+            code: 0,
+            stdout: JSON.stringify({ tab_id: 7, url: "http://x" }),
+            stderr: "",
+            timedOut: false,
+            aborted: false,
+          };
+        }
+        return {
+          code: 0,
+          stdout: JSON.stringify({ text: "ok", ref_count: 0, tab_id: 7, truncated: false }),
+          stderr: "",
+          timedOut: false,
+          aborted: false,
+        };
+      },
+      killAll() {},
+      killFor: () => 0,
+    };
+    const queue = new KeyedExecutor();
+    const observation = new ObservationService({
+      ctx: ctx as never,
+      runner,
+      registry,
+      queue,
+      options: { enabled: true, thumbnailIntervalMs: 1500, idleIntervalMs: 8000 },
+    });
+    registerTools({ 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 a = navigate?.execute({ url: "http://x" }, makeExec()) as Promise<unknown>;
+    // A starts immediately (empty queue) and blocks on the gate.
+    await vi.waitFor(() =>
+      expect(observation.getState().find((s) => s.sessionId === "s1")?.action).toBe("navigating"),
+    );
+    // B queues behind A: the label must stay "navigating" until A settles.
+    const b = snapshot?.execute({}, makeExec()) as Promise<unknown>;
+    await new Promise((resolve) => setTimeout(resolve, 30));
+    expect(observation.getState().find((s) => s.sessionId === "s1")?.action).toBe("navigating");
+    releaseNavigate();
+    await a;
+    await vi.waitFor(() =>
+      expect(observation.getState().find((s) => s.sessionId === "s1")?.action).toBe("snapshotting"),
+    );
+    await b;
+    await vi.waitFor(() =>
+      expect(observation.getState().find((s) => s.sessionId === "s1")?.action).toBe("idle"),
+    );
+    observation.dispose();
+  });
+});