@opengsd/gsd-pi 1.1.1-dev.616a1a1 → 1.1.1-dev.9bb7453

package/src/resources/extensions/gsd/tests/parallel-skill-prompt-integration.test.ts

@@ -22,23 +22,31 @@ import { tmpdir } from "node:os";
 
 import { loadSkills } from "@gsd/pi-coding-agent";
 import {
-  buildResearchSlicePrompt,
+  buildCompleteSlicePrompt,
   buildParallelResearchSlicesPrompt,
+  buildResearchSlicePrompt,
 } from "../auto-prompts.ts";
 
 const SKILL_NAME = "testskill";
+const COMPLETE_SLICE_SKILL_NAME = "complete-slice-policies";
 const SKILL_ACTIVATION_SUBSTRING = `Call Skill({ skill: '${SKILL_NAME}' })`;
+const COMPLETE_SLICE_SKILL_ACTIVATION_SUBSTRING = `Call Skill({ skill: '${COMPLETE_SLICE_SKILL_NAME}' })`;
 
 const tmpDirs: string[] = [];
 let savedCwd: string | undefined;
 
-function setupProjectWithSkill(): string {
+function setupProjectWithSkill(options: {
+  skillName?: string;
+  preferencesLines?: string[];
+} = {}): string {
+  const skillName = options.skillName ?? SKILL_NAME;
   const base = mkdtempSync(join(tmpdir(), "gsd-worker-skill-int-"));
   tmpDirs.push(base);
 
   // Milestone roadmap — buildResearchSlicePrompt inlines the roadmap excerpt.
   const milestoneDir = join(base, ".gsd", "milestones", "M001");
-  mkdirSync(join(milestoneDir, "slices", "S01"), { recursive: true });
+  const sliceOneDir = join(milestoneDir, "slices", "S01");
+  mkdirSync(join(sliceOneDir, "tasks"), { recursive: true });
   mkdirSync(join(milestoneDir, "slices", "S02"), { recursive: true });
   writeFileSync(
     join(milestoneDir, "M001-ROADMAP.md"),
@@ -55,27 +63,41 @@ function setupProjectWithSkill(): string {
     ].join("\n"),
     "utf-8",
   );
+  writeFileSync(
+    join(sliceOneDir, "S01-PLAN.md"),
+    [
+      "# S01: Alpha",
+      "",
+      "**Goal:** Verify worker x skill prompt plumbing.",
+      "**Demo:** Rendered prompts include the skill activation block.",
+      "",
+      "## Tasks",
+      "- [x] **T01: Task** `est:10m`",
+      "",
+    ].join("\n"),
+    "utf-8",
+  );
 
   // Project preferences — buildSkillActivationBlock picks these up via
   // loadEffectiveGSDPreferences(), which reads from `${cwd}/.gsd/PREFERENCES.md`.
   writeFileSync(
     join(base, ".gsd", "PREFERENCES.md"),
-    ["---", `always_use_skills:`, `  - ${SKILL_NAME}`, "---", ""].join("\n"),
+    ["---", ...(options.preferencesLines ?? [`always_use_skills:`, `  - ${skillName}`]), "---", ""].join("\n"),
     "utf-8",
   );
 
   // Project-scoped skill — resolveSkillReference scans `${cwd}/.agents/skills/`.
-  const skillDir = join(base, ".agents", "skills", SKILL_NAME);
+  const skillDir = join(base, ".agents", "skills", skillName);
   mkdirSync(skillDir, { recursive: true });
   writeFileSync(
     join(skillDir, "SKILL.md"),
     [
       "---",
-      `name: ${SKILL_NAME}`,
+      `name: ${skillName}`,
       `description: Integration-test skill for worker × skill prompt plumbing.`,
       "---",
       "",
-      `# ${SKILL_NAME}`,
+      `# ${skillName}`,
       "",
       "Test skill body.",
     ].join("\n"),
@@ -122,6 +144,31 @@ test("worker prompt (buildResearchSlicePrompt) includes <skill_activation> from
   );
 });
 
+test("complete-slice prompt includes <skill_activation> from unit-specific skill_rules", async () => {
+  const base = setupProjectWithSkill({
+    skillName: COMPLETE_SLICE_SKILL_NAME,
+    preferencesLines: [
+      "skill_rules:",
+      "  - when: complete-slice",
+      "    use:",
+      `      - ${COMPLETE_SLICE_SKILL_NAME}`,
+    ],
+  });
+  savedCwd = process.cwd();
+  process.chdir(base);
+
+  const prompt = await buildCompleteSlicePrompt("M001", "Test Milestone", "S01", "Alpha", base, "minimal");
+
+  assert.ok(
+    prompt.includes("<skill_activation>"),
+    "complete-slice prompt should contain a <skill_activation> block",
+  );
+  assert.ok(
+    prompt.includes(COMPLETE_SLICE_SKILL_ACTIVATION_SUBSTRING),
+    `complete-slice prompt should reference the skill-rule skill '${COMPLETE_SLICE_SKILL_NAME}'`,
+  );
+});
+
 test("subagent dispatch prompt (buildParallelResearchSlicesPrompt) carries <skill_activation> into each embedded per-slice section", async () => {
   const base = setupProjectWithSkill();
   savedCwd = process.cwd();

package/src/resources/extensions/gsd/tests/plan-slice-prompt.test.ts

@@ -83,6 +83,15 @@ test("plan-slice prompt: compact planning gates survive template substitution",
   assert.ok(!result.includes("{{"));
 });
 
+test("plan-slice prompt: absence checks use negated quiet searches", () => {
+  const result = loadPrompt("plan-slice", { ...BASE_VARS, commitInstruction: "Do not commit." });
+  assert.ok(result.includes("For absence checks"));
+  assert.ok(result.includes("`! grep -q 'pattern' file`"));
+  assert.ok(result.includes("`! rg -q 'pattern' file`"));
+  assert.ok(result.includes("do not use `grep -c` or `rg -c`"));
+  assert.ok(result.includes("count commands exit 1 when they find zero matches"));
+});
+
 test("plan-slice prompt: footer references gsd_plan_slice tool, not direct write", () => {
   const result = loadPrompt("plan-slice", { ...BASE_VARS, commitInstruction: "Do not commit." });
   assert.ok(

package/src/resources/extensions/gsd/tests/plan-slice.test.ts

@@ -6,7 +6,7 @@ import { mkdtempSync, mkdirSync, rmSync, readFileSync, existsSync, writeFileSync
 import { join } from 'node:path';
 import { tmpdir } from 'node:os';
 
-import { openDatabase, closeDatabase, insertMilestone, insertSlice, getSlice, getSliceTasks, getTask, getGateResults, updateTaskStatus } from '../gsd-db.ts';
+import { openDatabase, closeDatabase, insertMilestone, insertSlice, insertTask, getSlice, getSliceTasks, getTask, getGateResults, updateTaskStatus } from '../gsd-db.ts';
 import { handlePlanSlice } from '../tools/plan-slice.ts';
 import { parsePlan } from '../parsers-legacy.ts';
 import { parseTaskPlanFile } from '../files.ts';
@@ -265,6 +265,44 @@ test('handlePlanSlice renders plan artifacts under worktree-local .gsd while usi
   }
 });
 
+test('handlePlanSlice preserves completed task closeout state when replanning the same task', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    insertTask({
+      id: 'T01',
+      sliceId: 'S02',
+      milestoneId: 'M001',
+      title: 'Completed implementation',
+      status: 'complete',
+      oneLiner: 'Completed implementation',
+      narrative: 'Already finished.',
+      verificationResult: 'passed',
+      fullSummaryMd: '# T01 Summary\n\nAlready finished.\n',
+    });
+
+    const before = getTask('M001', 'S02', 'T01');
+    assert.equal(before?.status, 'complete');
+    assert.ok(before?.completed_at, 'completed task should have a completion timestamp');
+    assert.equal(before?.full_summary_md, '# T01 Summary\n\nAlready finished.\n');
+
+    const result = await handlePlanSlice(validParams(), base);
+    assert.ok(!('error' in result), `unexpected error: ${'error' in result ? result.error : ''}`);
+
+    const after = getTask('M001', 'S02', 'T01');
+    assert.equal(after?.status, 'complete', 'replanning must not reset completed task status to pending');
+    assert.equal(after?.completed_at, before?.completed_at, 'replanning must not clear completed_at');
+    assert.equal(after?.full_summary_md, before?.full_summary_md, 'replanning must not clear task summary content');
+    assert.equal(after?.verification_result, 'passed', 'replanning must not clear closeout verification');
+    assert.equal(after?.description, 'Implement the slice planning handler.', 'planning fields should still refresh');
+    assert.equal(getTask('M001', 'S02', 'T02')?.status, 'pending', 'new tasks are still inserted as pending');
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanSlice advances DB-derived state out of planning immediately', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));

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";

package/src/resources/extensions/gsd/tests/post-unit-retry-on-orchestrator-bridge.test.ts

@@ -8,6 +8,7 @@ import { mock } from "node:test";
 import { postUnitPostVerification, type PostUnitContext } from "../auto-post-unit.ts";
 import { AutoSession } from "../auto/session.ts";
 import { checkPostUnitHooks, resetHookState, resolveHookArtifactPath } from "../post-unit-hooks.ts";
+import { emitJournalEvent } from "../journal.ts";
 import { _clearGsdRootCache } from "../paths.ts";
 import { invalidateAllCaches } from "../cache.ts";
 
@@ -28,6 +29,43 @@ post_unit_hooks:
   writeFileSync(join(basePath, ".gsd", "PREFERENCES.md"), content, "utf-8");
 }
 
+function writeFailingHookPreferences(basePath: string): void {
+  const content = `---
+post_unit_hooks:
+  - name: review-arbiter
+    after:
+      - execute-task
+    prompt: Review {taskId}
+    artifact: REVIEW-DEBATE.md
+    max_cycles: 1
+    enabled: true
+  - name: follow-up-review
+    after:
+      - execute-task
+    prompt: Follow-up review {taskId}
+    enabled: true
+---
+`;
+  writeFileSync(join(basePath, ".gsd", "PREFERENCES.md"), content, "utf-8");
+}
+
+function writeBlockingPreferences(basePath: string): void {
+  const content = `---
+post_unit_hooks:
+  - name: review-arbiter
+    after:
+      - execute-task
+    prompt: Review {taskId}
+    agent: arbiter
+    artifact: REVIEW-DEBATE.md
+    criticality: blocking
+    max_cycles: 2
+    enabled: true
+---
+`;
+  writeFileSync(join(basePath, ".gsd", "PREFERENCES.md"), content, "utf-8");
+}
+
 test("post-unit retry_on marks trigger unit as retry in orchestrator before redispatch", async () => {
   const originalCwd = process.cwd();
   const base = mkdtempSync(join(tmpdir(), "gsd-post-unit-retry-"));
@@ -91,3 +129,144 @@ test("post-unit retry_on marks trigger unit as retry in orchestrator before redi
     rmSync(base, { recursive: true, force: true });
   }
 });
+
+test("failed post-unit hook pauses auto-mode even when its artifact exists", async () => {
+  const originalCwd = process.cwd();
+  const base = mkdtempSync(join(tmpdir(), "gsd-post-unit-hook-failed-"));
+  const taskDir = join(base, ".gsd", "milestones", "M001", "slices", "S01", "tasks");
+  mkdirSync(taskDir, { recursive: true });
+
+  try {
+    process.chdir(base);
+    _clearGsdRootCache();
+    invalidateAllCaches();
+    resetHookState();
+    writeFailingHookPreferences(base);
+
+    const hookDispatch = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.equal(hookDispatch?.hookName, "review-arbiter");
+
+    const artifactPath = resolveHookArtifactPath(base, "M001/S01/T01", "REVIEW-DEBATE.md");
+    writeFileSync(artifactPath, "partial review", "utf-8");
+    emitJournalEvent(base, {
+      ts: "2026-06-03T12:00:00.000Z",
+      flowId: "flow-hook-failed",
+      seq: 3,
+      eventType: "unit-end",
+      data: {
+        unitType: "hook/review-arbiter",
+        unitId: "M001/S01/T01",
+        status: "cancelled",
+        artifactVerified: false,
+      },
+    });
+
+    const pauseAuto = mock.fn(async () => {});
+    const notifications: string[] = [];
+    const s = new AutoSession();
+    s.basePath = base;
+    s.active = true;
+    s.currentUnit = { type: "hook/review-arbiter", id: "M001/S01/T01", startedAt: Date.now() };
+
+    const pctx: PostUnitContext = {
+      s,
+      ctx: {
+        ui: {
+          notify: (message: string) => { notifications.push(message); },
+          setStatus: () => {},
+          setWidget: () => {},
+          setFooter: () => {},
+        },
+        model: { id: "test-model" },
+      } as any,
+      pi: { sendMessage: async () => {}, setModel: async () => true } as any,
+      buildSnapshotOpts: () => ({}),
+      lockBase: () => base,
+      stopAuto: async () => {},
+      pauseAuto,
+      updateProgressWidget: () => {},
+    };
+
+    const result = await postUnitPostVerification(pctx);
+    assert.equal(result, "stopped");
+    assert.equal(pauseAuto.mock.callCount(), 1);
+    assert.ok(
+      notifications.some(message => message.includes("Post-unit hook review-arbiter failed")),
+      "pause notification should explain the failed hook",
+    );
+  } finally {
+    process.chdir(originalCwd);
+    resetHookState();
+    invalidateAllCaches();
+    _clearGsdRootCache();
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("post-unit blocking gate pauses auto-mode on needs-attention verdict", async () => {
+  const originalCwd = process.cwd();
+  const base = mkdtempSync(join(tmpdir(), "gsd-post-unit-gate-"));
+  const taskDir = join(base, ".gsd", "milestones", "M001", "slices", "S01", "tasks");
+  mkdirSync(taskDir, { recursive: true });
+
+  try {
+    process.chdir(base);
+    _clearGsdRootCache();
+    invalidateAllCaches();
+    resetHookState();
+    writeBlockingPreferences(base);
+
+    const hookDispatch = checkPostUnitHooks("execute-task", "M001/S01/T01", base);
+    assert.ok(hookDispatch, "hook should dispatch for execute-task");
+
+    const artifactPath = resolveHookArtifactPath(base, "M001/S01/T01", "REVIEW-DEBATE.md");
+    writeFileSync(artifactPath, "---\nverdict: needs-attention\n---\n\nManual review required.\n", "utf-8");
+
+    const pauseAuto = mock.fn(async () => {});
+    const s = new AutoSession();
+    s.basePath = base;
+    s.active = true;
+    s.currentUnit = { type: "hook/review-arbiter", id: "M001/S01/T01", startedAt: Date.now() };
+    s.orchestration = {
+      start: async () => ({ kind: "started" }),
+      advance: async () => ({ kind: "stopped", reason: "unused" }),
+      completeActiveUnit: async () => {},
+      retryActiveUnit: async () => {},
+      resume: async () => ({ kind: "resumed" }),
+      stop: async (reason: string) => ({ kind: "stopped", reason }),
+      getStatus: () => ({ phase: "running", transitionCount: 0 }),
+    };
+
+    const notifications: string[] = [];
+    const pctx: PostUnitContext = {
+      s,
+      ctx: {
+        ui: {
+          notify: (message: string) => { notifications.push(message); },
+          setStatus: () => {},
+          setWidget: () => {},
+          setFooter: () => {},
+        },
+        model: { id: "test-model" },
+      } as any,
+      pi: { sendMessage: async () => {}, setModel: async () => true } as any,
+      buildSnapshotOpts: () => ({}),
+      lockBase: () => base,
+      stopAuto: async () => {},
+      pauseAuto,
+      updateProgressWidget: () => {},
+    };
+
+    const result = await postUnitPostVerification(pctx);
+    assert.equal(result, "stopped");
+    assert.equal(pauseAuto.mock.callCount(), 1);
+    assert.match(notifications.join("\n"), /Post-unit gate "review-arbiter" blocked execute-task M001\/S01\/T01/);
+    assert.match(notifications.join("\n"), /\/gsd status/);
+  } finally {
+    process.chdir(originalCwd);
+    resetHookState();
+    invalidateAllCaches();
+    _clearGsdRootCache();
+    rmSync(base, { recursive: true, force: true });
+  }
+});

package/src/resources/extensions/gsd/tests/preferences.test.ts

@@ -561,6 +561,35 @@ test("post-unit hook max_cycles clamping via validatePreferences", () => {
   assert.equal(p4.post_unit_hooks![0].max_cycles, 3, "valid value passes through");
 });
 
+test("post-unit hook criticality and on_block validation", () => {
+  const base = { name: "h", after: ["execute-task"], prompt: "do something", artifact: "REVIEW.md" };
+
+  const { preferences, errors } = validatePreferences({
+    post_unit_hooks: [{
+      ...base,
+      criticality: "blocking",
+      on_block: { action: "retry-unit", artifact: "NEEDS-REWORK.md" },
+    }],
+  } as any);
+  assert.equal(errors.length, 0);
+  assert.equal(preferences.post_unit_hooks![0].criticality, "blocking");
+  assert.deepEqual(preferences.post_unit_hooks![0].on_block, {
+    action: "retry-unit",
+    artifact: "NEEDS-REWORK.md",
+  });
+
+  const missingArtifact = validatePreferences({
+    post_unit_hooks: [{ name: "blocking", after: ["execute-task"], prompt: "do something", criticality: "blocking" }],
+  } as any);
+  assert.match(missingArtifact.errors.join("\n"), /criticality blocking requires artifact/);
+  assert.equal(missingArtifact.preferences.post_unit_hooks, undefined);
+
+  const invalidAction = validatePreferences({
+    post_unit_hooks: [{ ...base, on_block: { action: "teleport" } }],
+  } as any);
+  assert.match(invalidAction.errors.join("\n"), /invalid on_block action/);
+});
+
 test("pre-dispatch hook action validation via validatePreferences", () => {
   const base = { name: "h", before: ["execute-task"] };
 

package/src/resources/extensions/gsd/tests/prompt-contracts.test.ts

@@ -2,6 +2,7 @@ import test from "node:test";
 import assert from "node:assert/strict";
 import { readFileSync } from "node:fs";
 import { join } from "node:path";
+import { RUN_UAT_WORKFLOW_TOOL_NAMES } from "../tool-presentation-plan.ts";
 
 const promptsDir = join(process.cwd(), "src/resources/extensions/gsd/prompts");
 const templatesDir = join(process.cwd(), "src/resources/extensions/gsd/templates");
@@ -25,11 +26,41 @@ test("reactive-execute prompt keeps task summaries with subagents and avoids bat
 test("run-uat prompt branches on dynamic UAT mode and supports runtime evidence", () => {
   const prompt = readPrompt("run-uat");
   assert.match(prompt, /\*\*Detected UAT mode:\*\*\s*`\{\{uatType\}\}`/);
-  assert.match(prompt, /uatType:\s*\{\{uatType\}\}/);
+  assert.match(prompt, /uatType:\s*"\{\{uatType\}\}"/);
+  assert.match(prompt, /gsd_uat_result_save/);
+  assert.match(prompt, /presentedTools/);
+  assert.match(prompt, /blockedTools/);
   assert.match(prompt, /live-runtime/);
   assert.match(prompt, /browser\/runtime\/network/i);
   assert.match(prompt, /NEEDS-HUMAN/);
   assert.doesNotMatch(prompt, /uatType:\s*artifact-driven/);
+  assert.doesNotMatch(prompt, /Call `gsd_summary_save`/);
+});
+
+test("run-uat prompt lists canonical gsd_uat_exec intent values", () => {
+  const prompt = readPrompt("run-uat");
+  assert.match(prompt, /`uat-artifact-check`/);
+  assert.match(prompt, /`uat-runtime-check`/);
+  assert.match(prompt, /`uat-browser-check`/);
+  assert.match(prompt, /`uat-service-start`/);
+  assert.match(prompt, /`uat-log-inspection`/);
+  assert.match(prompt, /do not use `artifact`, `runtime`, or `human-follow-up` as `intent`/i);
+});
+
+test("run-uat prompt gives the complete UAT result-save presentation contract", () => {
+  const prompt = readPrompt("run-uat");
+  assert.match(prompt, /Call `gsd_uat_result_save` once after all checks are complete/);
+  assert.doesNotMatch(prompt, /Call `gsd_summary_save` with `artifact_type: "ASSESSMENT"`/);
+
+  for (const toolName of RUN_UAT_WORKFLOW_TOOL_NAMES) {
+    assert.ok(prompt.includes(`"${toolName}"`), `prompt should include required presented tool ${toolName}`);
+  }
+
+  for (const toolName of ["gsd_exec", "gsd_summary_save", "gsd_save_gate_result"] as const) {
+    assert.ok(prompt.includes(`name: "${toolName}"`), `prompt should include blocked tool ${toolName}`);
+  }
+
+  assert.ok(prompt.includes("forbidden during run-uat"), "prompt should explain blocked run-uat tools");
 });
 
 test("workflow-start prompt defaults to autonomy instead of per-phase confirmation", () => {
@@ -465,6 +496,27 @@ test("reactive-execute prompt references tool calls instead of checkbox updates"
   assert.match(prompt, /completion tool calls/);
 });
 
+test("parallel subagent prompts forbid serialized tasks arrays", () => {
+  const expectations = [
+    { name: "reactive-execute", agent: "worker" },
+    { name: "parallel-research-slices", agent: "scout" },
+    { name: "gate-evaluate", agent: "tester" },
+  ] as const;
+
+  for (const { name, agent } of expectations) {
+    const prompt = readPrompt(name);
+    assert.match(prompt, /tasks:\s*\[\.\.\.\]/, `${name} must show the native array placeholder`);
+    assert.match(prompt, /native JSON array/i, `${name} must explicitly require a native JSON array`);
+    assert.match(prompt, /Do NOT JSON\.stringify/i, `${name} must forbid JSON.stringify on tasks`);
+    assert.match(prompt, /must be array/i, `${name} must mention the subagent validation failure`);
+    assert.match(
+      prompt,
+      new RegExp(`tasks:\\s*\\[\\{\\s*agent:\\s*"${agent}"`, "i"),
+      `${name} must show the concrete ${agent} task call shape`,
+    );
+  }
+});
+
 // ─── Project-shape classifier + 3-or-4-options-with-Other-hatch contract ──
 
 test("guided-discuss-project classifies project shape and persists the verdict to PROJECT.md", () => {