test(webui): let the browser acceptance matrix survive a CI run

packages/webui/test/browser/acceptance-record.ts

@@ -18,7 +18,7 @@
 // run, so a re-run cannot leave a case behind from the previous one, and the
 // suites cannot overwrite each other's records.
 
-import { mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
 import { release } from "node:os";
 import { resolve } from "node:path";
 
@@ -79,6 +79,55 @@ const PLAN_DIRECTORIES = {
   "workspace-browser-tab": "../../../../.plan/workspace-browser-tab/complete",
 } as const;
 
+/**
+ * The record files as the checkout had them.
+ *
+ * A record file is **tracked**: the plan's `complete/` directory is where the
+ * accepted evidence lives, and the file in the checkout is the record of the
+ * run that was accepted. A run rewrites it — every record carries the time
+ * and the machine it ran on — so the working tree is dirty once the suite is
+ * over.
+ *
+ * That dirt is not a source change, and the release packaging refuses to run
+ * with a dirty tree ("Packaging requires a clean committed working tree"), so
+ * on CI the run puts the files back when it is over.
+ *
+ * Locally the records stay: they are what the run just produced, and the
+ * published matrix is written from them. A suite that recorded into a file
+ * the checkout does not have leaves no file behind either.
+ */
+const originals = new Map<string, string | null>();
+
+/** Remembers a record file's content the first time the run writes to it. */
+function rememberOriginal(path: string): void {
+  if (originals.has(path)) return;
+  originals.set(path, existsSync(path) ? readFileSync(path, "utf8") : null);
+}
+
+/**
+ * Puts the record files back the way the checkout had them.
+ *
+ * On CI. A record file is tracked, so a run leaves the working tree dirty —
+ * every record carries the time and the machine it ran on — and the release
+ * packaging refuses to run with a dirty tree ("Packaging requires a clean
+ * committed working tree"). That dirt is the run's own timestamps, not a
+ * source change, so the suite puts the files back when it is done.
+ *
+ * Locally the records stay: they are what the run just produced, and the
+ * published matrix is written from them.
+ */
+export function restoreAcceptanceRecords(): void {
+  if (process.env.CI !== "true") return;
+  for (const [path, content] of originals) {
+    if (content === null) {
+      rmSync(path, { force: true });
+      continue;
+    }
+    writeFileSync(path, content, "utf8");
+  }
+  originals.clear();
+}
+
 /** A plan whose acceptance results are recorded. */
 export type WebuiAcceptancePlan = keyof typeof PLAN_DIRECTORIES;
 
@@ -119,7 +168,9 @@ export function acceptanceRecordsFor(plan: WebuiAcceptancePlan): AcceptanceRecor
    */
   function beginSuite(suite: string): void {
     mkdirSync(recordDir, { recursive: true });
-    writeFileSync(resolve(recordDir, `${suite}.acceptance.json`), "[]\n", "utf8");
+    const path = resolve(recordDir, `${suite}.acceptance.json`);
+    rememberOriginal(path);
+    writeFileSync(path, "[]\n", "utf8");
     started.add(suite);
   }
 
@@ -139,6 +190,7 @@ export function acceptanceRecordsFor(plan: WebuiAcceptancePlan): AcceptanceRecor
     };
     mkdirSync(recordDir, { recursive: true });
     const path = resolve(recordDir, `${suite}.acceptance.json`);
+    rememberOriginal(path);
     if (!started.has(suite)) {
       started.add(suite);
       writeFileSync(path, "[]\n", "utf8");

packages/webui/test/browser/browser-bridge-real-browser.spec.ts

@@ -24,7 +24,7 @@ import { rm } from "node:fs/promises";
 
 import { expect, test } from "@playwright/test";
 
-import { beginAcceptanceSuite, observeAcceptance } from "./acceptance-record.js";
+import { beginAcceptanceSuite, observeAcceptance, restoreAcceptanceRecords } from "./acceptance-record.js";
 import {
   browserVersionOf,
   bridgeSettings,
@@ -275,4 +275,10 @@ test.describe("the browser bridge, against a real browser", () => {
       }
     }
   });
+
+  // On CI the record file is tracked and the run's own timestamps are not a
+  // source change, so the file goes back the way the checkout had it.
+  test.afterAll(() => {
+    restoreAcceptanceRecords();
+  });
 });

packages/webui/test/browser/browser-bridge-real-external.spec.ts

@@ -50,7 +50,7 @@ async function extensionLastError(
 }
 
 import type { WebuiBrowserTargetState } from "../../src/shared/contracts/browser.js";
-import { beginAcceptanceSuite, observeAcceptance } from "./acceptance-record.js";
+import { beginAcceptanceSuite, observeAcceptance, restoreAcceptanceRecords } from "./acceptance-record.js";
 import {
   browserVersionOf,
   bridgeSettings,
@@ -327,6 +327,12 @@ test.describe("the external bridge, against a real browser and a real page", ()
       }
     });
   }
+
+  // On CI the record file is tracked and the run's own timestamps are not a
+  // source change, so the file goes back the way the checkout had it.
+  test.afterAll(() => {
+    restoreAcceptanceRecords();
+  });
 });
 
 /** Waits for the extension to announce a tab. */

packages/webui/test/browser/browser-bridge-real-packaged.spec.ts

@@ -31,7 +31,7 @@ import { resolve } from "node:path";
 
 import { expect, test } from "@playwright/test";
 
-import { beginAcceptanceSuite, observeAcceptance } from "./acceptance-record.js";
+import { beginAcceptanceSuite, observeAcceptance, restoreAcceptanceRecords } from "./acceptance-record.js";
 import {
   browserVersionOf,
   bridgeSettings,
@@ -262,4 +262,10 @@ test.describe("the packaged production entry, against a real browser", () => {
       }
     });
   }
+
+  // On CI the record file is tracked and the run's own timestamps are not a
+  // source change, so the file goes back the way the checkout had it.
+  test.afterAll(() => {
+    restoreAcceptanceRecords();
+  });
 });

packages/webui/test/integration/workspace-browser-tab-acceptance.test.ts

@@ -41,7 +41,7 @@ import {
   SEND_BROWSER_COMMAND_OPERATION_NAME,
   SEND_BROWSER_INPUT_OPERATION_NAME,
 } from "../../src/shared/operation-names.js";
-import { acceptanceRecordsFor } from "../browser/acceptance-record.js";
+import { acceptanceRecordsFor, restoreAcceptanceRecords } from "../browser/acceptance-record.js";
 import type {
   WebuiBrowserAuthorization,
   WebuiBrowserTargetId,
@@ -54,6 +54,18 @@ const SUITE = "workspace-browser-tab";
 /** How long a page is given to settle after a move. */
 const SETTLE_MS = 750;
 
+/**
+ * How long a case may take.
+ *
+ * A case drives a real browser: the first one launches it, opens a tab and
+ * waits for the strip to gain it. The default budget (5s) is smaller than a
+ * cold browser start on a loaded runner, and a case that runs out of time
+ * leaves the browser half launched — the next case then finds a browser that
+ * exited during startup and fails too. The sibling suite
+ * (`browser-panel-live-navigation.test.ts`) has always given its cases 30s.
+ */
+const TIMEOUT_MS = 30_000;
+
 async function chromeIsAvailable(): Promise<boolean> {
   for (const candidate of ["/usr/bin/google-chrome", "/usr/bin/chromium"]) {
     try {
@@ -123,6 +135,9 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
   afterAll(async () => {
     await service?.close();
     if (dataDir) await rm(dataDir, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
+    // On CI the record file is tracked and the run's own timestamps are not a
+    // source change, so the file goes back the way the checkout had it.
+    restoreAcceptanceRecords();
   });
 
   // -- the paths the cases drive ------------------------------------------
@@ -300,7 +315,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(state.tabs[0]?.content.kind).toBe("web");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("the Agent navigates after opening, and the picture follows", async () => {
     await record(
@@ -340,7 +355,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(frames.length).toBeGreaterThan(0);
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("the Agent acts on a page it did not open, and a stale reference is refused", async () => {
     await record(
@@ -377,7 +392,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         ).rejects.toMatchObject({ code: "browser_target_stale" });
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a person opens a tab while the Agent works, and the two pages differ", async () => {
     await record(
@@ -408,7 +423,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await readTitle(personPage!)).toContain("person");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a person opens a tab while the run has no selection, and the run gets it", async () => {
     await record(
@@ -437,7 +452,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await readTitle(page!)).toContain("acted on");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a person types into their own tab while the Agent works", async () => {
     await record(
@@ -505,7 +520,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await readField(agentPage)).toBe("");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a person types a url in the launcher, and the tab becomes that page", async () => {
     await record(
@@ -537,7 +552,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect((commands[1] as { readonly type: string }).type).toBe("open-web-tab");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a person closes a web tab, and the page is released", async () => {
     await record(
@@ -577,7 +592,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await runSelection("run-6")).toBeUndefined();
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("back and forward move through the page's own history", async () => {
     await record(
@@ -625,7 +640,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await readTitle(back)).toContain("history two");
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("the surface's switch is off, and both paths are refused", async () => {
     await record(
@@ -671,7 +686,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         }
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a run with no selection is refused every action but open_tab", async () => {
     await record(
@@ -695,7 +710,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(opened).toBeDefined();
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a session ends, and its pages are released", async () => {
     await record(
@@ -719,7 +734,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(await runSelection("run-8")).toBeUndefined();
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a second web tab in one session is addressable, and the strip shows both", async () => {
     await record(
@@ -759,7 +774,7 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         expect(webuiWorkspaceTabStripReconcile([], published)).toHaveLength(2);
       },
     );
-  });
+  }, TIMEOUT_MS);
 
   it("a page navigates, and a stale reference is refused on every tab", async () => {
     await record(
@@ -798,5 +813,5 @@ describe.skipIf(!available)("the acceptance of the workspace browser tab model",
         ).resolves.toMatchObject({ accepted: true });
       },
     );
-  });
+  }, TIMEOUT_MS);
 });