@opengsd/gsd-pi 1.2.0-dev.5457a158 → 1.2.0-dev.822c9439

package/src/resources/extensions/gsd/tests/auto-remote-session-lock-cleanup.test.ts

@@ -1,12 +1,13 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { mkdtempSync, mkdirSync, rmSync } from "node:fs";
+import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import { tmpdir } from "node:os";
 
-import { checkRemoteAutoSession } from "../auto.ts";
+import { checkRemoteAutoSession, forceStopAutoRemote } from "../auto.ts";
 import { openDatabase, closeDatabase, _getAdapter } from "../gsd-db.ts";
-import { registerAutoWorker } from "../db/auto-workers.ts";
+import { getAutoWorker, registerAutoWorker } from "../db/auto-workers.ts";
+import { claimMilestoneLease, getMilestoneLease } from "../db/milestone-leases.ts";
 import { normalizeRealPath } from "../paths.ts";
 import { readCrashLock } from "../crash-recovery.ts";
 
@@ -35,6 +36,29 @@ function setWorkerPid(workerId: string, pid: number): void {
   ).run({ ":pid": pid, ":worker_id": workerId });
 }
 
+function insertMilestone(id: string): void {
+  const db = _getAdapter()!;
+  db.prepare(
+    `INSERT INTO milestones (id, title, status, created_at)
+     VALUES (:id, :title, 'active', :created_at)`,
+  ).run({
+    ":id": id,
+    ":title": id,
+    ":created_at": new Date().toISOString(),
+  });
+}
+
+function writeLegacyLock(base: string, pid: number): void {
+  const now = new Date().toISOString();
+  writeFileSync(join(base, ".gsd", "auto.lock"), JSON.stringify({
+    pid,
+    startedAt: now,
+    unitType: "execute-task",
+    unitId: "M001/S01/T01",
+    unitStartedAt: now,
+  }));
+}
+
 function findDeadPidCandidate(): number {
   const candidates = [99_999, 199_999, 299_999, 399_999];
   for (const pid of candidates) {
@@ -62,3 +86,41 @@ test("checkRemoteAutoSession clears stale lock state when lock PID is dead", (t)
   assert.deepEqual(remote, { running: false });
   assert.equal(readCrashLock(base), null, "stale lock should be cleared by remote session check");
 });
+
+test("forceStopAutoRemote escalates a live remote PID and releases worker state", (t) => {
+  const base = makeBase();
+  t.after(() => cleanup(base));
+
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  const workerId = registerAutoWorker({ projectRootRealpath: normalizeRealPath(base) });
+  const pid = 424_242;
+  setWorkerPid(workerId, pid);
+  insertMilestone("M001");
+  writeLegacyLock(base, pid);
+  const lease = claimMilestoneLease(workerId, "M001");
+  assert.equal(lease.ok, true, "precondition: worker holds a milestone lease");
+
+  const signals: Array<NodeJS.Signals | 0> = [];
+  const originalKill = process.kill;
+  process.kill = ((target: number, signal?: NodeJS.Signals | number) => {
+    assert.equal(target, pid);
+    signals.push((signal ?? 0) as NodeJS.Signals | 0);
+    return true;
+  }) as typeof process.kill;
+  t.after(() => {
+    process.kill = originalKill;
+  });
+
+  const result = forceStopAutoRemote(base);
+
+  assert.deepEqual(result, { found: true, pid });
+  assert.ok(signals.includes("SIGTERM"), "force stop should request graceful termination first");
+  assert.ok(signals.includes("SIGKILL"), "force stop should escalate when the PID is still alive");
+  assert.equal(getAutoWorker(workerId)?.status, "stopping");
+  assert.equal(
+    getMilestoneLease("M001")?.status,
+    "released",
+    "force stop should release held milestone leases",
+  );
+  assert.equal(readCrashLock(base), null, "force stop should remove the visible remote lock");
+});

package/src/resources/extensions/gsd/tests/blocked-models.test.ts

@@ -8,7 +8,10 @@ import { join } from "node:path";
 
 import {
   blockModel,
+  blockModelUntil,
+  clearTemporaryModelBlocksForTest,
   isModelBlocked,
+  isModelTemporarilyUnavailable,
   loadBlockedModels,
 } from "../blocked-models.ts";
 
@@ -96,3 +99,19 @@ test("blocked-models: file created under .gsd/runtime/", () => {
     rmSync(base, { recursive: true, force: true });
   }
 });
+
+test("blocked-models: temporary rate-limit blocks expire without persisting", () => {
+  const base = mkBase();
+  try {
+    clearTemporaryModelBlocksForTest();
+    blockModelUntil(base, "openai-codex", "gpt-5.5", Date.now() + 60_000, "session limit");
+    assert.equal(isModelTemporarilyUnavailable(base, "openai-codex", "gpt-5.5"), true);
+    assert.equal(loadBlockedModels(base).length, 0, "rate-limit windows must not persist as account blocks");
+
+    blockModelUntil(base, "openai-codex", "gpt-5.5", Date.now() - 1, "expired");
+    assert.equal(isModelTemporarilyUnavailable(base, "openai-codex", "gpt-5.5"), false);
+  } finally {
+    clearTemporaryModelBlocksForTest();
+    rmSync(base, { recursive: true, force: true });
+  }
+});

package/src/resources/extensions/gsd/tests/browser-automation-contract-fixture.ts

@@ -0,0 +1,39 @@
+import assert from "node:assert/strict";
+
+import {
+	getUatBrowserToolSupportError,
+	hasUatBrowserToolSurface,
+	type UatType,
+} from "../uat-policy.ts";
+
+export const BROWSER_AUTOMATION_CONTRACT_TOOLS = {
+	piProvider: ["read", "browser_navigate"],
+	externalMcpClient: ["read", "mcp__gsd-browser__browser_navigate"],
+	externalMcpWildcard: ["read", "mcp__gsd-browser__*"],
+	otherBrowserMcp: ["read", "mcp__browser-uat__*"],
+	workflowOnly: ["read", "mcp__gsd-workflow__*"],
+	withoutBrowser: ["read", "gsd_uat_exec"],
+} as const;
+
+export function assertBrowserAutomationContractAvailable(tools: readonly string[]): void {
+	assert.equal(hasUatBrowserToolSurface(tools), true, `${tools.join(", ")} should satisfy the Browser Automation Contract`);
+}
+
+export function assertBrowserAutomationContractMissing(tools: readonly string[] | undefined): void {
+	assert.equal(hasUatBrowserToolSurface(tools), false, `${tools?.join(", ") ?? "undefined"} should not satisfy the Browser Automation Contract`);
+}
+
+export function assertBrowserBackedUatCanDispatch(options: {
+	uatType: UatType;
+	activeTools: readonly string[] | undefined;
+	registeredTools?: readonly string[];
+}): void {
+	assert.equal(
+		getUatBrowserToolSupportError({
+			...options,
+			milestoneId: "M001",
+			sliceId: "S01",
+		}),
+		null,
+	);
+}

package/src/resources/extensions/gsd/tests/browser-contract.test.ts

@@ -0,0 +1,44 @@
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  BROWSER_CONTRACT_TOOL_NAMES,
+  BROWSER_EVIDENCE_SIGNAL_TOOL_NAMES,
+  hasBrowserContractPrefix,
+  isBrowserContractToolName,
+} from "../../shared/browser-contract.ts";
+import { isUatBrowserToolName } from "../uat-policy.ts";
+import { BROWSER_REQUIREMENT_RE, BROWSER_RUNTIME_RE } from "../browser-evidence.ts";
+
+// Note: RUN_UAT_BROWSER_TOOL_NAMES and MANAGED_GSD_BROWSER_TOOL_NAMES are
+// reference-equal aliases of BROWSER_CONTRACT_TOOL_NAMES, and the managed
+// adapter's spec table is Record-keyed by BrowserContractToolName — both
+// derivations are pinned by the type system, not by runtime assertions here.
+describe("Browser Automation Contract parity", () => {
+  it("every contract name satisfies the UAT browser-tool predicate, bare and MCP-prefixed", () => {
+    for (const name of BROWSER_CONTRACT_TOOL_NAMES) {
+      assert.equal(isUatBrowserToolName(name), true, name);
+      assert.equal(isUatBrowserToolName(`mcp__gsd-browser__${name}`), true, `mcp__gsd-browser__${name}`);
+    }
+  });
+
+  it("contract names are canonical browser_* names with no duplicates", () => {
+    assert.equal(new Set(BROWSER_CONTRACT_TOOL_NAMES).size, BROWSER_CONTRACT_TOOL_NAMES.length);
+    for (const name of BROWSER_CONTRACT_TOOL_NAMES) {
+      assert.equal(hasBrowserContractPrefix(name), true, name);
+      assert.equal(isBrowserContractToolName(name), true, name);
+    }
+    assert.equal(isBrowserContractToolName("browser_not_a_real_tool"), false);
+    assert.equal(hasBrowserContractPrefix("gsd_uat_exec"), false);
+  });
+
+  it("evidence-signal names stay a subset of the contract and drive the detection regexes", () => {
+    for (const name of BROWSER_EVIDENCE_SIGNAL_TOOL_NAMES) {
+      assert.equal(isBrowserContractToolName(name), true, name);
+      // Identifier-shaped names keep the regex splice in browser-evidence.ts escape-free.
+      assert.match(name, /^browser_[a-z_]+$/);
+      assert.match(`Verified via ${name} call`, BROWSER_REQUIREMENT_RE);
+      assert.match(`Verified via ${name} call`, BROWSER_RUNTIME_RE);
+    }
+  });
+});

package/src/resources/extensions/gsd/tests/browser-daemon-auto-prep.test.ts

@@ -0,0 +1,144 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  prepareBrowserDaemonForUat,
+  shouldWarmBrowserDaemonForUat,
+} from "../browser-daemon-auto-prep.ts";
+import { commitBrowserEngineResolution } from "../../browser-tools/engine/selection.ts";
+import { resolveGsdBrowserCliAvailability } from "../../shared/gsd-browser-cli.ts";
+
+const GSD_BROWSER_ENGINE = { GSD_BROWSER_ENGINE: "gsd-browser" } as const;
+
+test("shouldWarmBrowserDaemonForUat skips artifact-driven UAT", () => {
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "artifact-driven",
+      sessionProvider: "claude-code",
+      projectRoot: "/tmp/project",
+    }),
+    false,
+  );
+});
+
+test("shouldWarmBrowserDaemonForUat enables Claude Code browser UAT when gsd-browser is available", (t) => {
+  const availability = resolveGsdBrowserCliAvailability();
+  if (!availability.available) {
+    t.skip("bundled gsd-browser CLI unavailable");
+  }
+
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "browser-executable",
+      sessionProvider: "claude-code",
+      sessionAuthMode: "externalCli",
+      projectRoot: "/tmp/project",
+      env: GSD_BROWSER_ENGINE,
+    }),
+    true,
+  );
+});
+
+test("shouldWarmBrowserDaemonForUat enables warm-up for Claude Code oauth/apiKey when engine is gsd-browser", (t) => {
+  const availability = resolveGsdBrowserCliAvailability();
+  if (!availability.available) {
+    t.skip("bundled gsd-browser CLI unavailable");
+  }
+
+  for (const sessionAuthMode of ["oauth", "apiKey"] as const) {
+    assert.equal(
+      shouldWarmBrowserDaemonForUat({
+        uatType: "browser-executable",
+        sessionProvider: "claude-code",
+        sessionAuthMode,
+        sessionBaseUrl: "https://api.anthropic.com",
+        projectRoot: "/tmp/project",
+        env: GSD_BROWSER_ENGINE,
+      }),
+      true,
+      `expected warm-up for sessionAuthMode=${sessionAuthMode}`,
+    );
+  }
+});
+
+test("shouldWarmBrowserDaemonForUat skips legacy Playwright engine for Claude Code", () => {
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "browser-executable",
+      sessionProvider: "claude-code",
+      sessionAuthMode: "oauth",
+      projectRoot: "/tmp/project",
+      env: { GSD_BROWSER_ENGINE: "legacy" },
+    }),
+    false,
+  );
+});
+
+test("shouldWarmBrowserDaemonForUat uses session-committed ambient engine for non-Claude providers", () => {
+  const projectRoot = "/tmp/ambient-engine-project";
+  commitBrowserEngineResolution(projectRoot, {
+    engine: "legacy",
+    source: "probe",
+    reason: "gsd-browser daemon connect failed (test); using legacy Playwright",
+  });
+
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "browser-executable",
+      sessionProvider: "openai",
+      projectRoot,
+    }),
+    false,
+  );
+});
+
+test("shouldWarmBrowserDaemonForUat skips when browser MCP is disabled", () => {
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "browser-executable",
+      sessionProvider: "claude-code",
+      projectRoot: "/tmp/project",
+      env: { GSD_BROWSER_MCP_ENABLED: "0" },
+    }),
+    false,
+  );
+});
+
+test("shouldWarmBrowserDaemonForUat skips when warm-up is disabled", () => {
+  assert.equal(
+    shouldWarmBrowserDaemonForUat({
+      uatType: "browser-executable",
+      sessionProvider: "claude-code",
+      projectRoot: "/tmp/project",
+      env: { GSD_BROWSER_WARMUP: "0" },
+    }),
+    false,
+  );
+});
+
+test("prepareBrowserDaemonForUat returns null when warm-up is not required", () => {
+  assert.equal(
+    prepareBrowserDaemonForUat({
+      uatType: "artifact-driven",
+      sessionProvider: "claude-code",
+      sessionAuthMode: "externalCli",
+      projectRoot: "/tmp/example-project",
+    }),
+    null,
+  );
+});
+
+test("prepareBrowserDaemonForUat returns actionable error when daemon start fails", () => {
+  const error = prepareBrowserDaemonForUat({
+    uatType: "browser-executable",
+    sessionProvider: "claude-code",
+    sessionAuthMode: "externalCli",
+    projectRoot: "/tmp/example-project",
+    env: {
+      ...GSD_BROWSER_ENGINE,
+      GSD_BROWSER_MCP_COMMAND: "/definitely/missing/gsd-browser",
+    },
+  });
+
+  assert.match(error ?? "", /gsd-browser daemon failed to start/i);
+});

package/src/resources/extensions/gsd/tests/complete-slice-verification-gate.test.ts

@@ -258,6 +258,48 @@ describe('complete-slice verification gate (#3580)', () => {
     }
   });
 
+  test('allows a browser UAT declared as a bare keyword under ## UAT Type (M006/S01 format drift)', async () => {
+    // Regression: the agent wrote `## UAT Type\nbrowser-executable` (no
+    // `- UAT mode:` bullet). The old parser defaulted that to artifact-driven
+    // and the gate rejected the slice in a loop.
+    const body = [
+      '## UAT Type',
+      'browser-executable',
+      '',
+      '## Smoke Test',
+      '1. Open the page in a browser and perform add/edit/complete/delete once.',
+    ].join('\n');
+    const result = await handleCompleteSlice(
+      makeParams({ uatContent: body }),
+      basePath,
+    );
+    if ('error' in result) {
+      assert.doesNotMatch(
+        result.error,
+        /requires browser verification/i,
+        `bare browser-executable declaration must pass the browser gate, got: ${result.error}`,
+      );
+    }
+  });
+
+  test('explains the missing declaration when a browser UAT has no parseable UAT mode', async () => {
+    // When nothing was declared, the error must not claim the agent
+    // "declared artifact-driven" — it must show the expected bullet format.
+    const body = [
+      '## Smoke Test',
+      '1. Open the page in a browser and perform add/edit/complete/delete once.',
+    ].join('\n');
+    const result = await handleCompleteSlice(
+      makeParams({ uatContent: body }),
+      basePath,
+    );
+    assert.ok('error' in result, 'expected handler to reject an undeclared browser UAT');
+    const error = (result as { error: string }).error;
+    assert.match(error, /no parseable UAT mode declaration/i);
+    assert.match(error, /- UAT mode: browser-executable/);
+    assert.doesNotMatch(error, /but declares "UAT mode: artifact-driven"/);
+  });
+
   test('allows a browser UAT when it is declared mixed (mixed receives browser tools)', async () => {
     const body = BROWSER_UAT_BODY.replace('artifact-driven', 'mixed (artifact-driven + browser)');
     const result = await handleCompleteSlice(