@opengsd/gsd-pi 1.1.1-dev.2034b16 → 1.1.1-dev.595401e

package/src/resources/extensions/gsd/rule-types.ts

@@ -32,6 +32,8 @@ export interface RuleLifecycle {
   retry_on?: string;
   /** Max times this hook can fire for the same trigger unit. */
   max_cycles?: number;
+  /** Whether this hook is advisory or blocking. */
+  criticality?: PostUnitHookConfig["criticality"];
   /** Idempotency key pattern for this hook. */
   idempotency_key?: string;
 }

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

@@ -0,0 +1,142 @@
+// Project/App: gsd-pi
+// File Purpose: Unit tests for hasBrowserRequiredText heading-depth section guard.
+
+import { describe, test } from 'node:test';
+import assert from 'node:assert/strict';
+
+import { hasBrowserRequiredText } from '../browser-evidence.ts';
+
+describe('hasBrowserRequiredText', () => {
+  test('detects browser requirement in a plain test-cases section', () => {
+    const text = [
+      '## Test Cases',
+      '',
+      '1. Open index.html in a browser and navigate to /dashboard.',
+      '',
+    ].join('\n');
+    assert.ok(hasBrowserRequiredText(text), 'plain browser step should be detected');
+  });
+
+  test('ignores browser mention under a top-level non-requirement heading', () => {
+    const text = [
+      '## Not Proven',
+      '',
+      '- Keyboard usability through a real browser.',
+      '- Browser console cleanliness.',
+      '',
+    ].join('\n');
+    assert.ok(!hasBrowserRequiredText(text), 'browser mention under "Not Proven" should be ignored');
+  });
+
+  test('sub-heading inside a non-requirement section does not re-enable detection', () => {
+    // BUG (pre-fix): ### sub-heading under ## Not Proven resets inNonRequirementSection
+    // to false, causing subsequent lines to be detected as browser requirements.
+    const text = [
+      '## Not Proven By This UAT',
+      '',
+      '- No live browser session was used.',
+      '',
+      '### Visual Checks',
+      '',
+      '- Browser visual polish deferred to next slice.',
+      '- Keyboard interaction in a real browser is not proven here.',
+      '',
+    ].join('\n');
+    assert.ok(
+      !hasBrowserRequiredText(text),
+      'sub-heading under a non-requirement section must not re-enable browser detection',
+    );
+  });
+
+  test('requirement-level heading after non-requirement section re-enables detection', () => {
+    const text = [
+      '## Not Proven',
+      '',
+      '- Browser polish deferred.',
+      '',
+      '## Test Cases',
+      '',
+      '1. Launch browser and open localhost.',
+      '',
+    ].join('\n');
+    assert.ok(
+      hasBrowserRequiredText(text),
+      'browser step under "Test Cases" (same depth as "Not Proven") must still be detected',
+    );
+  });
+
+  test('deferred sub-heading inside a requirement section scopes exclusion to its own block', () => {
+    const text = [
+      '## Test Cases',
+      '',
+      '1. Open browser at localhost.',
+      '',
+      '### Deferred: keyboard check',
+      '',
+      '- Keyboard UAT deferred to next slice.',
+      '',
+      '### Step 2: Verify DOM',
+      '',
+      '1. Navigate to /dashboard in the browser.',
+      '',
+    ].join('\n');
+    assert.ok(
+      hasBrowserRequiredText(text),
+      'browser step under "Step 2" sub-heading must be detected after a sibling "Deferred" sub-heading',
+    );
+  });
+
+  test('deferred sub-heading at same depth as test cases does not escape to parent', () => {
+    const text = [
+      '## Test Cases',
+      '',
+      '### Deferred: responsive layout',
+      '',
+      '- Responsive layout check is deferred to S02.',
+      '',
+    ].join('\n');
+    assert.ok(
+      !hasBrowserRequiredText(text),
+      'content under a "Deferred" sub-heading should be excluded from detection',
+    );
+  });
+
+  test('detects browser requirement written only in a heading', () => {
+    // Regression: the line-by-line scan previously skip-continued past headings,
+    // missing browser obligations expressed only in heading text.
+    const text = '## Open browser session at localhost\n';
+    assert.ok(hasBrowserRequiredText(text), 'browser requirement in heading text must be detected');
+  });
+
+  test('heading that opens a non-requirement section is not itself detected as a requirement', () => {
+    const text = '## Not Proven\n\n- Some note.\n';
+    assert.ok(
+      !hasBrowserRequiredText(text),
+      'a non-requirement section heading should not trigger browser detection',
+    );
+  });
+
+  test('returns false for empty text', () => {
+    assert.ok(!hasBrowserRequiredText(''), 'empty string returns false');
+  });
+
+  test('notes-for-tester heading with sub-headings stays non-requirement', () => {
+    const text = [
+      '## Notes for Tester',
+      '',
+      '### Browser Setup',
+      '',
+      '- Run this spec without a browser; a DOM harness is sufficient.',
+      '- Browser-based visual checks are deferred.',
+      '',
+      '### Follow-up Items',
+      '',
+      '- Track browser session evidence in S02.',
+      '',
+    ].join('\n');
+    assert.ok(
+      !hasBrowserRequiredText(text),
+      'sub-headings under "Notes for Tester" should not re-enable browser detection',
+    );
+  });
+});

package/src/resources/extensions/gsd/tests/complete-milestone-excerpt.test.ts

@@ -12,6 +12,7 @@ import { tmpdir } from "node:os";
 
 import { buildSliceSummaryExcerpt, buildCompleteMilestonePrompt, buildValidateMilestonePrompt } from "../auto-prompts.ts";
 import { invalidateAllCaches } from "../cache.ts";
+import { closeDatabase, insertMilestone, openDatabase } from "../gsd-db.ts";
 
 // ─── Fixture helpers ──────────────────────────────────────────────────────
 
@@ -364,3 +365,32 @@ test("validate-milestone prompt uses slice excerpts and on-demand paths instead
     "validate prompt must not inline full assessment traces",
   );
 });
+
+test("validate-milestone prompt inlines planned verification classes as canonical rows", async (t) => {
+  const base = createBase();
+  t.after(() => {
+    try { closeDatabase(); } catch { /* ignore */ }
+    cleanup(base);
+  });
+  invalidateAllCaches();
+
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  insertMilestone({
+    id: "M001",
+    planning: {
+      verificationContract: "Local command exits 0.",
+      verificationOperational: "No long-running child process remains.",
+    },
+  });
+  writeRoadmap(base, makeRoadmap());
+  writeSummary(base, "S01", makeFatSummary("S01"));
+  writeSummary(base, "S02", makeFatSummary("S02"));
+
+  const prompt = await buildValidateMilestonePrompt("M001", "Test Milestone", base);
+
+  assert.match(prompt, /### Verification Classes \(from planning\)/);
+  assert.match(prompt, /Every row in this table must appear in `verificationClasses`/);
+  assert.match(prompt, /\| Class \| Planned Check \|/);
+  assert.match(prompt, /\| Contract \| Local command exits 0\. \|/);
+  assert.match(prompt, /\| Operational \| No long-running child process remains\. \|/);
+});

package/src/resources/extensions/gsd/tests/doctor-runtime-checks.test.ts

@@ -45,3 +45,30 @@ test("doctor fix respects git.manage_gitignore false (#4161)", async (t) => {
   assert.equal(readFileSync(join(dir, ".gitignore"), "utf-8"), "node_modules/\n");
   assert.equal(existsSync(join(dir, ".gsd", "PREFERENCES.md")), true);
 });
+
+test("doctor fix resets run-uat counters at the dispatch cap", async (t) => {
+  const dir = createGitProject();
+  t.after(() => rmSync(dir, { recursive: true, force: true }));
+
+  const runtimeDir = join(dir, ".gsd", "runtime");
+  mkdirSync(runtimeDir, { recursive: true });
+  const counterPath = join(runtimeDir, "uat-count-M002-S01.json");
+  writeFileSync(
+    counterPath,
+    JSON.stringify({ count: 3, updatedAt: "2026-06-02T19:40:23.289Z" }) + "\n",
+    "utf-8",
+  );
+
+  const detect = await runGSDDoctor(dir);
+  const issue = detect.issues.find((candidate) => candidate.code === "uat_retry_exhausted");
+  assert.ok(issue, "doctor reports the exhausted UAT retry counter at the dispatch cap");
+  assert.equal(issue.unitId, "M002/S01");
+  assert.match(issue.message, /3 attempt\(s\)/);
+
+  const fixed = await runGSDDoctor(dir, { fix: true, scope: "M002/S02" });
+  assert.ok(
+    fixed.fixesApplied.some((fix) => fix.includes("reset exhausted run-uat retry counter for M002/S01")),
+    "doctor --fix resets the blocked counter even when the current displayed scope has advanced",
+  );
+  assert.equal(existsSync(counterPath), false);
+});

package/src/resources/extensions/gsd/tests/integration/auto-recovery.test.ts

@@ -119,8 +119,8 @@ test("resolveExpectedArtifactPath returns correct path for all slice-level types
 // ─── run-uat artifact path contract (#2873) ──────────────────────────────
 
 test("resolveExpectedArtifactPath for run-uat returns ASSESSMENT path, not UAT (#2873)", (t) => {
-  // The run-uat prompt instructs the agent to call gsd_summary_save with
-  // artifact_type: "ASSESSMENT", which writes S##-ASSESSMENT.md. The artifact
+  // The run-uat prompt instructs the agent to call gsd_uat_result_save, which
+  // writes S##-ASSESSMENT.md through the workflow persistence path. The artifact
   // verification path must match — otherwise verification fails and auto-mode
   // retries the unit in an infinite loop.
   const base = makeTmpBase();
@@ -147,12 +147,12 @@ test("diagnoseExpectedArtifact for run-uat references ASSESSMENT (#2873)", (t) =
 });
 
 test("verifyExpectedArtifact passes for run-uat when ASSESSMENT file exists (#2873)", (t) => {
-  // Regression test: run-uat writes S##-ASSESSMENT.md via gsd_summary_save,
+  // Regression test: run-uat writes S##-ASSESSMENT.md via gsd_uat_result_save,
   // but verification looked for S##-UAT.md, causing false stuck retries.
   const base = makeTmpBase();
   t.after(() => cleanup(base));
 
-  // Write the ASSESSMENT file (what gsd_summary_save actually produces)
+  // Write the ASSESSMENT file (what gsd_uat_result_save actually produces)
   const assessPath = join(base, ".gsd", "milestones", "M001", "slices", "S01", "S01-ASSESSMENT.md");
   writeFileSync(assessPath, "---\nverdict: PASS\n---\n# UAT Assessment\n");
 

package/src/resources/extensions/gsd/tests/integration/run-uat.test.ts

@@ -72,6 +72,38 @@ function makeBrowserObservableUatContent(mode = 'artifact-driven'): string {
   ].join('\n');
 }
 
+function makeDeferredBrowserUatContent(): string {
+  return [
+    '# UAT File',
+    '',
+    '## UAT Type',
+    '',
+    '- UAT mode: artifact-driven',
+    '- Why this mode is sufficient: Node interaction tests exercise the real app.js render/event/localStorage loop through a DOM harness. Live browser, keyboard, responsive, and visual-polish UAT remain intentionally deferred to S02.',
+    '',
+    '## Smoke Test',
+    '',
+    'Run `node --test tests/s01-static-interactions.test.js` and confirm all tests pass.',
+    '',
+    '## Test Cases',
+    '',
+    '1. Click the todo row edit control in the DOM harness.',
+    '2. Save changed text and reload/recreate the app from persisted localStorage.',
+    '3. Expected: the stored record shape remains unchanged.',
+    '',
+    '## Not Proven By This UAT',
+    '',
+    '- Final visual polish of edit controls.',
+    '- Keyboard usability through a real browser.',
+    '- Browser console and local network cleanliness.',
+    '',
+    '## Notes for Tester',
+    '',
+    'S02 should capture browser evidence for the full loop rather than changing this persisted model.',
+    '',
+  ].join('\n');
+}
+
 describe('run-uat', () => {
 test('(a) artifact-driven', () => {
   assert.deepStrictEqual(
@@ -232,8 +264,8 @@ test('(k) run-uat prompt template', () => {
     `prompt contains detected dynamic uatType value "${uatType}" after substitution`,
   );
   assert.ok(
-    promptResult?.includes(`uatType: ${uatType}`) ?? false,
-    `prompt contains dynamic uatType frontmatter value "${uatType}" after substitution`,
+    promptResult?.includes(`uatType: "${uatType}"`) ?? false,
+    `prompt contains dynamic uatType field "${uatType}" after substitution`,
   );
   assert.ok(
     !/\{\{[^}]+\}\}/.test(promptResult ?? ''),
@@ -249,7 +281,7 @@ test('(k) run-uat prompt template', () => {
   );
 });
 
-test('(k2) run-uat prompt references gsd_summary_save, not direct write', () => {
+test('(k2) run-uat prompt references gsd_uat_result_save, not direct write', () => {
   const promptResult = loadPromptFromWorktree('run-uat', {
     workingDirectory: '/tmp/test-project',
     milestoneId: 'M001',
@@ -261,17 +293,25 @@ test('(k2) run-uat prompt references gsd_summary_save, not direct write', () =>
   });
 
   assert.ok(
-    promptResult.includes('gsd_summary_save'),
-    'run-uat prompt should reference gsd_summary_save tool',
+    promptResult.includes('gsd_uat_result_save'),
+    'run-uat prompt should reference gsd_uat_result_save tool',
+  );
+  assert.ok(
+    promptResult.includes('presentedTools') && promptResult.includes('blockedTools'),
+    'run-uat prompt should specify the tool presentation contract',
   );
   assert.ok(
-    promptResult.includes('artifact_type: "ASSESSMENT"'),
-    'run-uat prompt should specify ASSESSMENT artifact type',
+    !promptResult.includes('Call `gsd_summary_save`'),
+    'run-uat prompt should not instruct direct summary-save UAT persistence',
   );
   assert.ok(
     !promptResult.includes('MUST write'),
     'run-uat prompt should not instruct direct file write in footer',
   );
+  assert.ok(
+    !promptResult.includes('Call `gsd_summary_save` with `artifact_type: "ASSESSMENT"`'),
+    'run-uat prompt should not instruct the legacy summary-save UAT path',
+  );
 });
 
 test('(l) dispatch preconditions via resolveSliceFile', () => {
@@ -482,8 +522,8 @@ test('(n) stale replay guard', async () => {
 });
 
 test('(q) verdict in ASSESSMENT file skips UAT dispatch (file-based path)', async () => {
-    // Regression test for #2644: run-uat prompt writes the verdict to
-    // S{sid}-ASSESSMENT.md (via gsd_summary_save artifact_type:"ASSESSMENT"),
+    // Regression test for #2644: run-uat writes the verdict to
+    // S{sid}-ASSESSMENT.md through the structured UAT save path,
     // but checkNeedsRunUat only checked S{sid}-UAT.md — causing a stuck loop.
     const base = createFixtureBase();
     try {
@@ -679,11 +719,27 @@ test('(u) run-uat prompt promotes artifact-driven browser specs to browser-execu
       const prompt = await buildRunUatPrompt('M001', 'S01', uatRel, uatContent, base);
 
       assert.match(prompt, /\*\*Detected UAT mode:\*\*\s*`browser-executable`/);
-      assert.match(prompt, /uatType: browser-executable/);
+      assert.match(prompt, /uatType: "browser-executable"/);
       assert.match(prompt, /use gsd-browser tools/i);
     } finally {
       cleanup(base);
     }
 });
 
+test('(v) run-uat prompt keeps deferred browser work artifact-driven', async () => {
+    const base = createFixtureBase();
+    try {
+      const uatRel = '.gsd/milestones/M001/slices/S01/S01-UAT.md';
+      const uatContent = makeDeferredBrowserUatContent();
+      writeSliceFile(base, 'M001', 'S01', 'UAT', uatContent);
+
+      const prompt = await buildRunUatPrompt('M001', 'S01', uatRel, uatContent, base);
+
+      assert.match(prompt, /\*\*Detected UAT mode:\*\*\s*`artifact-driven`/);
+      assert.match(prompt, /uatType: "artifact-driven"/);
+      assert.doesNotMatch(prompt, /uatType: "browser-executable"/);
+    } finally {
+      cleanup(base);
+    }
+});
 });

package/src/resources/extensions/gsd/tests/post-unit-hooks.test.ts

@@ -11,6 +11,7 @@ import {
   resetHookState,
   isRetryPending,
   consumeRetryTrigger,
+  consumeGateBlock,
   resolveHookArtifactPath,
   runPreDispatchHooks,
   persistHookState,
@@ -20,6 +21,7 @@ import {
   formatHookStatus,
   triggerHookManually,
 } from "../post-unit-hooks.ts";
+import { invalidateAllCaches } from "../cache.ts";
 
 // ─── Fixture Helpers ───────────────────────────────────────────────────────
 
@@ -29,6 +31,11 @@ function createFixtureBase(): string {
   return base;
 }
 
+function writeHookPreferences(base: string, hookYaml: string): void {
+  writeFileSync(join(base, ".gsd", "PREFERENCES.md"), `---\npost_unit_hooks:\n${hookYaml}\n---\n`, "utf-8");
+  invalidateAllCaches();
+}
+
 // ═══════════════════════════════════════════════════════════════════════════
 // Phase 1: Post-Unit Hook Tests
 // ═══════════════════════════════════════════════════════════════════════════
@@ -104,6 +111,156 @@ test('consumeRetryTrigger clears state', () => {
   assert.ok(!isRetryPending(), "no retry initially");
 });
 
+test('Advisory hook keeps artifact idempotency without verdict frontmatter', () => {
+  resetHookState();
+  const base = createFixtureBase();
+  try {
+    writeHookPreferences(base, `  - name: docs-hint
+    after:
+      - execute-task
+    prompt: Review docs
+    artifact: DOCS-HINT.md
+`);
+    writeFileSync(resolveHookArtifactPath(base, "M001/S01/T01", "DOCS-HINT.md"), "plain advisory note", "utf-8");
+
+    const result = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.deepStrictEqual(result, null, "existing advisory artifact remains idempotent");
+    assert.deepStrictEqual(consumeGateBlock(), null, "advisory hook does not create gate block");
+  } finally {
+    resetHookState();
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test('Blocking hook skips only after passing frontmatter verdict', () => {
+  resetHookState();
+  const base = createFixtureBase();
+  try {
+    writeHookPreferences(base, `  - name: security-review
+    after:
+      - execute-task
+    prompt: Review security
+    artifact: SECURITY-REVIEW.md
+    criticality: blocking
+`);
+    writeFileSync(
+      resolveHookArtifactPath(base, "M001/S01/T01", "SECURITY-REVIEW.md"),
+      "---\nverdict: pass\n---\n\nNo blocking findings.\n",
+      "utf-8",
+    );
+
+    const result = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.deepStrictEqual(result, null, "passing gate artifact is idempotent");
+    assert.deepStrictEqual(consumeGateBlock(), null, "passing gate does not block");
+  } finally {
+    resetHookState();
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test('Blocking hook reruns invalid artifact once then blocks at cycle budget', () => {
+  resetHookState();
+  const base = createFixtureBase();
+  try {
+    writeHookPreferences(base, `  - name: security-review
+    after:
+      - execute-task
+    prompt: Review security
+    artifact: SECURITY-REVIEW.md
+    criticality: blocking
+`);
+    writeFileSync(resolveHookArtifactPath(base, "M001/S01/T01", "SECURITY-REVIEW.md"), "partial output", "utf-8");
+
+    const dispatch = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.ok(dispatch, "invalid gate artifact dispatches the blocking hook");
+    assert.equal(dispatch.unitType, "hook/security-review");
+
+    const afterHook = checkPostUnitHooks("hook/security-review", "M001/S01/T01", base);
+    assert.deepStrictEqual(afterHook, null, "no further hook dispatch after max_cycles=1");
+    const block = consumeGateBlock();
+    assert.ok(block, "gate block is recorded");
+    assert.equal(block.hookName, "security-review");
+    assert.match(block.reason, /missing frontmatter verdict/);
+  } finally {
+    resetHookState();
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test('Blocking hook restored from disk does not trust artifact without clean hook completion', () => {
+  resetHookState();
+  const base = createFixtureBase();
+  try {
+    writeHookPreferences(base, `  - name: security-review
+    after:
+      - execute-task
+    prompt: Review security
+    artifact: SECURITY-REVIEW.md
+    criticality: blocking
+    max_cycles: 2
+`);
+    const firstDispatch = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.ok(firstDispatch, "gate dispatches first cycle");
+    persistHookState(base);
+
+    writeFileSync(
+      resolveHookArtifactPath(base, "M001/S01/T01", "SECURITY-REVIEW.md"),
+      "---\noutcome:\n  verdict: pass\n---\n",
+      "utf-8",
+    );
+
+    resetHookState();
+    restoreHookState(base);
+
+    const resumed = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.ok(resumed, "persisted active gate reruns when clean hook completion was not observed");
+    assert.equal(resumed.unitType, "hook/security-review");
+  } finally {
+    resetHookState();
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test('Blocking hook needs-rework verdict requests trigger unit retry', () => {
+  resetHookState();
+  const base = createFixtureBase();
+  try {
+    writeHookPreferences(base, `  - name: review-arbiter
+    after:
+      - execute-task
+    prompt: Review task
+    artifact: REVIEW-DEBATE.md
+    criticality: blocking
+    max_cycles: 2
+    on_block:
+      action: retry-unit
+`);
+    const dispatch = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.ok(dispatch, "gate dispatches");
+    writeFileSync(
+      resolveHookArtifactPath(base, "M001/S01/T01", "REVIEW-DEBATE.md"),
+      "---\nverdict: needs-rework\n---\n\nRework required.\n",
+      "utf-8",
+    );
+
+    const afterHook = checkPostUnitHooks("hook/review-arbiter", "M001/S01/T01", base);
+    assert.deepStrictEqual(afterHook, null, "needs-rework routes via retry signal");
+    assert.ok(isRetryPending(), "retry is pending");
+    assert.deepStrictEqual(consumeRetryTrigger(), {
+      unitType: "execute-task",
+      unitId: "M001/S01/T01",
+    });
+  } finally {
+    resetHookState();
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
 // ─── Variable substitution in prompts ──────────────────────────────────────
 test('Variable substitution', () => {
   const base = "/project";