@opengsd/gsd-pi 1.2.0-dev.b1abb545 → 1.2.0-dev.d6c5343c

package/src/resources/extensions/gsd/tests/auto-post-unit-evidence-crossref-4909.test.ts

@@ -88,3 +88,49 @@ test("detects session execution tools supported by the evidence collector", () =
 
   assert.equal(_hasExecutionToolCallsInSessionForTest(entries), true);
 });
+
+test("detects execution tool calls in bare agent-end messages (no session-entry wrapper)", () => {
+  // The auto loop passes opts.agentEndMessages as bare {role, content}
+  // messages — not {type: "message", message} session-manager entries.
+  const entries = [
+    {
+      role: "assistant",
+      content: [
+        {
+          type: "toolCall",
+          name: "Bash",
+          arguments: { command: "test -s index.html && grep -q localStorage index.html" },
+        },
+      ],
+    },
+  ];
+
+  assert.equal(_hasExecutionToolCallsInSessionForTest(entries), true);
+});
+
+test("does not suppress for bare agent-end messages without execution tools", () => {
+  const entries = [
+    {
+      role: "assistant",
+      content: [
+        { type: "text", text: "Task complete." },
+        { type: "toolCall", name: "Write", arguments: { file_path: "index.html" } },
+      ],
+    },
+  ];
+
+  assert.equal(_hasExecutionToolCallsInSessionForTest(entries), false);
+});
+
+test("ignores bare user messages with toolCall-shaped content", () => {
+  const entries = [
+    {
+      role: "user",
+      content: [
+        { type: "toolCall", name: "bash", arguments: { command: "echo hi" } },
+      ],
+    },
+  ];
+
+  assert.equal(_hasExecutionToolCallsInSessionForTest(entries), false);
+});

package/src/resources/extensions/gsd/tests/auto-worktree-registry.test.ts

@@ -184,7 +184,7 @@ describe("auto-worktree workspace registry", () => {
     git(["commit", "-m", "add milestone"], tempDir);
 
     createAutoWorktree(tempDir, "M003");
-    const wtDir = join(tempDir, ".gsd", "worktrees", "M003");
+    const wtDir = join(tempDir, ".gsd-worktrees", "M003");
     writeFileSync(join(wtDir, "feature.txt"), "implemented\n");
     git(["add", "feature.txt"], wtDir);
     git(["commit", "-m", "feat: implement M003"], wtDir);
@@ -216,7 +216,7 @@ describe("auto-worktree workspace registry", () => {
     git(["commit", "-m", "add milestone"], tempDir);
 
     createAutoWorktree(tempDir, "M004");
-    const wtDir = join(tempDir, ".gsd", "worktrees", "M004");
+    const wtDir = join(tempDir, ".gsd-worktrees", "M004");
     writeFileSync(join(wtDir, "feature.txt"), "implemented\n");
     git(["add", "feature.txt"], wtDir);
     git(["commit", "-m", "feat: implement M004"], wtDir);

package/src/resources/extensions/gsd/tests/auto-worktree-repair.test.ts

@@ -35,7 +35,8 @@ test("repair target accepts a missing expected milestone worktree", () => {
 
     assert.equal(result.ok, true);
     if (result.ok) {
-      assert.equal(result.expectedPath, join(base, ".gsd", "worktrees", "M001"));
+      // No worktree exists yet, so the expected path is the canonical container.
+      assert.equal(result.expectedPath, join(base, ".gsd-worktrees", "M001"));
     }
   } finally {
     cleanup(base);
@@ -192,7 +193,8 @@ test("paused metadata path resolves to the expected worktree while paused at pro
       baseIsAutoWorktree: false,
     });
 
-    assert.equal(result, join(base, ".gsd", "worktrees", "M001"));
+    // No worktree exists yet, so resolution lands at the canonical container.
+    assert.equal(result, join(base, ".gsd-worktrees", "M001"));
   } finally {
     cleanup(base);
   }

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/checkout-branch-stash-guard.test.ts

@@ -5,7 +5,7 @@ import { mkdirSync, mkdtempSync, readFileSync, realpathSync, rmSync, writeFileSy
 import { join } from "node:path";
 import { tmpdir } from "node:os";
 
-import { checkoutBranchWithStashGuard } from "../auto-worktree.ts";
+import { checkoutBranchWithStashGuard } from "../worktree-git-recovery.ts";
 
 function git(args: string[], cwd: string): string {
   return execFileSync("git", args, {
@@ -159,4 +159,69 @@ describe("checkoutBranchWithStashGuard", () => {
     const stashList = git(["stash", "list"], repo).trim();
     assert.equal(stashList, "");
   });
+
+  test("auto-resolves combined non-.gsd untracked collision and .gsd index conflict", (t) => {
+    // Regression for: gate checked nonGsdUnmerged.length === 0 but ignored
+    // .gsd/ index conflicts, so a failed pop with both an untracked non-.gsd/
+    // collision AND a .gsd/ tracked conflict would drop the stash while leaving
+    // .gsd/ index unmerged entries unresolved.
+    const repo = createRepo(t);
+    // Add .gsd/DECISIONS.md to base so both branches diverge from it
+    writeFileSync(join(repo, ".gsd", "DECISIONS.md"), "base\n");
+    git(["add", ".gsd/DECISIONS.md"], repo);
+    git(["commit", "-m", "add decisions"], repo);
+
+    git(["checkout", "-b", "milestone/M003"], repo);
+    // Branch has its own version of DECISIONS.md (non-JSONL .gsd/ file)
+    writeFileSync(join(repo, ".gsd", "DECISIONS.md"), "target-version\n");
+    mkdirSync(join(repo, ".harness"), { recursive: true });
+    writeFileSync(join(repo, ".harness", "settings.json"), "{\"theme\":\"dark\"}\n");
+    git(["add", ".gsd/DECISIONS.md", ".harness/settings.json"], repo);
+    git(["commit", "-m", "branch state"], repo);
+    git(["checkout", "main"], repo);
+
+    // On main: tracked .gsd/ change (creates unmerged index entry on pop) +
+    // untracked harness file (triggers "already exists, no checkout").
+    writeFileSync(join(repo, ".gsd", "DECISIONS.md"), "local-version\n");
+    mkdirSync(join(repo, ".harness"), { recursive: true });
+    writeFileSync(join(repo, ".harness", "settings.json"), "{\"theme\":\"light\"}\n");
+
+    checkoutBranchWithStashGuard(repo, "milestone/M003", "test-combined-gsd-conflict");
+
+    const branch = git(["branch", "--show-current"], repo).trim();
+    assert.equal(branch, "milestone/M003");
+    const decisions = readFileSync(join(repo, ".gsd", "DECISIONS.md"), "utf8");
+    assert.equal(decisions, "target-version\n");
+    assert.doesNotMatch(decisions, /<<<<<<<|=======|>>>>>>>/);
+    const status = git(["status", "--porcelain"], repo).trim();
+    assert.equal(status, "");
+    const stashList = git(["stash", "list"], repo).trim();
+    assert.equal(stashList, "");
+  });
+
+  test("auto-resolves non-.gsd untracked restore collisions (e.g. harness config outside .gsd/)", (t) => {
+    const repo = createRepo(t);
+    // Target branch has a harness config file committed; source (main) has it untracked.
+    // Use a path not affected by global gitignore rules (unlike .claude/settings.local.json).
+    git(["checkout", "-b", "milestone/M001"], repo);
+    mkdirSync(join(repo, ".harness"), { recursive: true });
+    writeFileSync(join(repo, ".harness", "settings.json"), "{\"theme\":\"dark\"}\n");
+    git(["add", ".harness/settings.json"], repo);
+    git(["commit", "-m", "add harness settings"], repo);
+    git(["checkout", "main"], repo);
+
+    mkdirSync(join(repo, ".harness"), { recursive: true });
+    writeFileSync(join(repo, ".harness", "settings.json"), "{\"theme\":\"light\"}\n");
+
+    checkoutBranchWithStashGuard(repo, "milestone/M001", "test-non-gsd-untracked-collision");
+
+    const branch = git(["branch", "--show-current"], repo).trim();
+    assert.equal(branch, "milestone/M001");
+    const wtContent = readFileSync(join(repo, ".harness", "settings.json"), "utf8");
+    assert.equal(wtContent, "{\"theme\":\"dark\"}\n");
+    const status = git(["status", "--porcelain"], repo).trim();
+    assert.equal(status, "");
+    const stashList = git(["stash", "list"], repo).trim();
+    assert.equal(stashList, "");
+  });
 });

package/src/resources/extensions/gsd/tests/clear-stale-autostart.test.ts

@@ -73,4 +73,48 @@ describe("clear stale pending auto-start (#3667)", () => {
       "pending auto-start gate must clear stale map entries for completed discussions",
     );
   });
+
+  test("guided-flow recovers a finished-but-unconsumed discussion instead of dead-ending", () => {
+    // CONTEXT exists + no live turn means the discussion completed but the
+    // agent_end handoff never consumed the entry (e.g. an external-engine
+    // post-hoc gate re-arm wiped the depth verification after the save).
+    // Without recovery, every /gsd prints "Discussion already in progress"
+    // forever: the stale heuristic requires CONTEXT to be absent and
+    // discussPlanComplete requires a ROADMAP that planning never produced.
+    const source = readFileSync(join(__dirname, "..", "guided-flow.ts"), "utf-8");
+    assert.ok(
+      source.includes("milestoneHasContext && !isAgentTurnInFlight(ctx)"),
+      "pending-entry guard must have a recovery branch for CONTEXT-present, no-turn-in-flight entries",
+    );
+    assert.ok(
+      source.includes("extractDepthVerificationMilestoneId(pendingGateId) === entry.milestoneId"),
+      "recovery must only clear a pending gate belonging to the entry's own milestone",
+    );
+    assert.ok(
+      source.includes("if (checkAutoStartAfterDiscuss(basePath)) return;"),
+      "recovery must re-run the discuss→auto handoff after clearing the stale gate",
+    );
+  });
+
+  test("guided-flow does not treat a live discuss turn as a stale pending entry", () => {
+    const source = readFileSync(join(__dirname, "..", "guided-flow.ts"), "utf-8");
+    assert.ok(
+      source.includes("!isAgentTurnInFlight(ctx)"),
+      "stale-entry deletion must be gated on no agent turn being in flight — a dispatched " +
+      "discuss turn can think for over 30s before writing its first artifact, and deleting " +
+      "its entry re-dispatches the workflow (duplicate interview + duplicate completion message)",
+    );
+    assert.ok(
+      source.includes('const milestoneHasDraft = !!resolveMilestoneFile(basePath, entry.milestoneId, "CONTEXT-DRAFT");'),
+      "stale-entry check must treat an existing CONTEXT-DRAFT as proof of an in-progress interview",
+    );
+    assert.ok(
+      source.includes("!milestoneHasDraft"),
+      "stale-entry deletion must require the CONTEXT-DRAFT to be absent",
+    );
+    assert.ok(
+      source.includes("ctx.hasPendingMessages"),
+      "in-flight detection must also cover dispatched-but-not-yet-started queued messages",
+    );
+  });
 });

package/src/resources/extensions/gsd/tests/commands-verdict.test.ts

@@ -472,18 +472,19 @@ test("auto-dispatch needs-attention pause message references /gsd verdict", asyn
   }
 });
 
-test("state.ts needs-remediation blocker messages reference /gsd verdict", async () => {
+test("guidance.ts needs-remediation blocker messages reference /gsd verdict", async () => {
   // We don't need to invoke deriveState — just assert the substring is in the
-  // source. The blocker strings are constructed inline and shipped to the user
-  // verbatim, so a static check is sufficient and avoids fragile DB setup.
-  const stateSource = readFileSync(
-    new URL("../state.ts", import.meta.url).pathname,
+  // source. The blocker strings are constructed in the guidance catalog and
+  // shipped to the user verbatim, so a static check is sufficient and avoids
+  // fragile DB setup.
+  const guidanceSource = readFileSync(
+    new URL("../guidance.ts", import.meta.url).pathname,
     "utf-8",
   );
-  const occurrences = stateSource.match(/`\/gsd verdict /g) ?? [];
+  const occurrences = guidanceSource.match(/`\/gsd verdict /g) ?? [];
   assert.ok(
     occurrences.length >= 2,
-    `expected at least 2 references to /gsd verdict in state.ts blockers, found ${occurrences.length}`,
+    `expected at least 2 references to /gsd verdict in guidance.ts blockers, found ${occurrences.length}`,
   );
 });
 

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(