gsd-pi 2.80.0-dev.e146beb20 → 2.80.0-dev.e6c48c3af

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

@@ -85,7 +85,11 @@ function makeMockDeps(
     resolveMilestoneFile: () => null,
     reconcileMergeState: () => "clean",
     preflightCleanRoot: () => ({ stashPushed: false, summary: "" }),
-    postflightPopStash: () => {},
+    postflightPopStash: () => ({
+      restored: true,
+      needsManualRecovery: false,
+      message: "restored",
+    }),
     getLedger: () => ({ units: [] }),
     getProjectTotals: () => ({ cost: 0 }),
     formatCost: (c: number) => `$${c.toFixed(2)}`,
@@ -399,6 +403,240 @@ test("runDispatch pauses when complete-milestone summary exists on disk but the
   assert.equal(stopCalls, 0, "mismatch pause should not hard-stop the loop");
 });
 
+test("runDispatch pauses when execute-task artifacts exist but DB status is still open", async (t) => {
+  const capture = createEventCapture();
+  let pauseCalls = 0;
+  let stopCalls = 0;
+  let invalidateCalls = 0;
+  const base = join(tmpdir(), `gsd-stuck-execute-task-${randomUUID()}`);
+  t.after(() => {
+    closeDatabase();
+    rmSync(base, { recursive: true, force: true });
+  });
+
+  const sliceDir = join(base, ".gsd", "milestones", "M001", "slices", "S01");
+  const tasksDir = join(sliceDir, "tasks");
+  mkdirSync(tasksDir, { recursive: true });
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  insertMilestone({ id: "M001", title: "Test", status: "active" });
+  insertSlice({ id: "S01", milestoneId: "M001", title: "Slice", status: "in_progress" });
+  insertTask({ id: "T01", milestoneId: "M001", sliceId: "S01", title: "First task", status: "pending" });
+  writeFileSync(
+    join(sliceDir, "S01-PLAN.md"),
+    [
+      "# S01",
+      "",
+      "## Tasks",
+      "",
+      "- [x] **T01: First task** `est:1h`",
+      "",
+    ].join("\n"),
+  );
+  writeFileSync(join(tasksDir, "T01-PLAN.md"), "# T01 Plan\n");
+  writeFileSync(join(tasksDir, "T01-SUMMARY.md"), "# T01 Summary\n\nDone on disk.\n");
+
+  const deps = makeMockDeps(capture, {
+    pauseAuto: async () => { pauseCalls++; },
+    stopAuto: async () => { stopCalls++; },
+    invalidateAllCaches: () => { invalidateCalls++; },
+    resolveDispatch: async () => ({
+      action: "dispatch" as const,
+      unitType: "execute-task",
+      unitId: "M001/S01/T01",
+      prompt: "execute the task",
+      matchedRule: "executing → execute-task",
+    }),
+  });
+  const ic = makeIC(deps, {
+    s: {
+      ...makeSession(),
+      basePath: base,
+      originalBasePath: base,
+    } as any,
+  });
+  const preData: PreDispatchData = {
+    state: {
+      phase: "executing",
+      activeMilestone: { id: "M001", title: "Test", status: "active" },
+      activeSlice: { id: "S01", title: "Slice" },
+      activeTask: { id: "T01", title: "First task" },
+      registry: [{ id: "M001", status: "active" }],
+      blockers: [],
+    } as any,
+    mid: "M001",
+    midTitle: "Test Milestone",
+  };
+  const loopState: LoopState = {
+    recentUnits: [
+      { key: "execute-task/M001/S01/T01" },
+      { key: "execute-task/M001/S01/T01" },
+    ],
+    stuckRecoveryAttempts: 0,
+    consecutiveFinalizeTimeouts: 0,
+  };
+
+  const result = await runDispatch(ic, preData, loopState);
+
+  assert.equal(result.action, "break");
+  assert.equal((result as any).reason, "execute-task-artifact-db-mismatch");
+  assert.equal(pauseCalls, 1, "execute-task disk/db mismatch should pause auto-mode");
+  assert.equal(stopCalls, 0, "execute-task disk/db mismatch should not hard-stop the loop");
+  assert.equal(invalidateCalls, 0, "mismatch should not clear caches and continue toward redispatch");
+  assert.equal(loopState.recentUnits.length, 3, "mismatch should keep the stuck window intact");
+  assert.equal(loopState.stuckRecoveryAttempts, 1, "mismatch should not reset the recovery counter");
+});
+
+test("runDispatch pauses at Level 2 when execute-task artifacts exist but DB status is still open", async (t) => {
+  const capture = createEventCapture();
+  let pauseCalls = 0;
+  let stopCalls = 0;
+  let invalidateCalls = 0;
+  const base = join(tmpdir(), `gsd-stuck-execute-task-l2-${randomUUID()}`);
+  t.after(() => {
+    closeDatabase();
+    rmSync(base, { recursive: true, force: true });
+  });
+
+  const sliceDir = join(base, ".gsd", "milestones", "M001", "slices", "S01");
+  const tasksDir = join(sliceDir, "tasks");
+  mkdirSync(tasksDir, { recursive: true });
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  insertMilestone({ id: "M001", title: "Test", status: "active" });
+  insertSlice({ id: "S01", milestoneId: "M001", title: "Slice", status: "in_progress" });
+  insertTask({ id: "T01", milestoneId: "M001", sliceId: "S01", title: "First task", status: "pending" });
+  writeFileSync(
+    join(sliceDir, "S01-PLAN.md"),
+    "# S01\n\n## Tasks\n\n- [x] **T01: First task** `est:1h`\n",
+  );
+  writeFileSync(join(tasksDir, "T01-PLAN.md"), "# T01 Plan\n");
+  writeFileSync(join(tasksDir, "T01-SUMMARY.md"), "# T01 Summary\n\nDone on disk.\n");
+
+  const deps = makeMockDeps(capture, {
+    pauseAuto: async () => { pauseCalls++; },
+    stopAuto: async () => { stopCalls++; },
+    invalidateAllCaches: () => { invalidateCalls++; },
+    resolveDispatch: async () => ({
+      action: "dispatch" as const,
+      unitType: "execute-task",
+      unitId: "M001/S01/T01",
+      prompt: "execute the task",
+      matchedRule: "executing execute-task",
+    }),
+  });
+  const ic = makeIC(deps, {
+    s: {
+      ...makeSession(),
+      basePath: base,
+      originalBasePath: base,
+    } as any,
+  });
+  const preData: PreDispatchData = {
+    state: {
+      phase: "executing",
+      activeMilestone: { id: "M001", title: "Test", status: "active" },
+      activeSlice: { id: "S01", title: "Slice" },
+      activeTask: { id: "T01", title: "First task" },
+      registry: [{ id: "M001", status: "active" }],
+      blockers: [],
+    } as any,
+    mid: "M001",
+    midTitle: "Test Milestone",
+  };
+  const loopState: LoopState = {
+    recentUnits: [
+      { key: "execute-task/M001/S01/T01" },
+      { key: "execute-task/M001/S01/T01" },
+    ],
+    stuckRecoveryAttempts: 1,
+    consecutiveFinalizeTimeouts: 0,
+  };
+
+  const result = await runDispatch(ic, preData, loopState);
+
+  assert.equal(result.action, "break");
+  assert.equal((result as any).reason, "execute-task-artifact-db-mismatch");
+  assert.equal(pauseCalls, 1, "Level 2 execute-task disk/db mismatch should pause auto-mode");
+  assert.equal(stopCalls, 0, "Level 2 execute-task disk/db mismatch should not hard-stop the loop");
+  assert.equal(invalidateCalls, 1, "Level 2 should invalidate caches before the final artifact recheck");
+  assert.equal(loopState.recentUnits.length, 3, "Level 2 mismatch should keep the stuck window intact");
+  assert.equal(loopState.stuckRecoveryAttempts, 1, "Level 2 mismatch should not reset the recovery counter");
+});
+
+test("runDispatch clears execute-task stuck state when artifacts and DB status are complete", async (t) => {
+  const capture = createEventCapture();
+  let pauseCalls = 0;
+  let stopCalls = 0;
+  let invalidateCalls = 0;
+  const base = join(tmpdir(), `gsd-stuck-execute-task-complete-${randomUUID()}`);
+  t.after(() => {
+    closeDatabase();
+    rmSync(base, { recursive: true, force: true });
+  });
+
+  const sliceDir = join(base, ".gsd", "milestones", "M001", "slices", "S01");
+  const tasksDir = join(sliceDir, "tasks");
+  mkdirSync(tasksDir, { recursive: true });
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  insertMilestone({ id: "M001", title: "Test", status: "active" });
+  insertSlice({ id: "S01", milestoneId: "M001", title: "Slice", status: "in_progress" });
+  insertTask({ id: "T01", milestoneId: "M001", sliceId: "S01", title: "First task", status: "complete" });
+  writeFileSync(
+    join(sliceDir, "S01-PLAN.md"),
+    "# S01\n\n## Tasks\n\n- [x] **T01: First task** `est:1h`\n",
+  );
+  writeFileSync(join(tasksDir, "T01-PLAN.md"), "# T01 Plan\n");
+  writeFileSync(join(tasksDir, "T01-SUMMARY.md"), "# T01 Summary\n\nDone on disk.\n");
+
+  const deps = makeMockDeps(capture, {
+    pauseAuto: async () => { pauseCalls++; },
+    stopAuto: async () => { stopCalls++; },
+    invalidateAllCaches: () => { invalidateCalls++; },
+    resolveDispatch: async () => ({
+      action: "dispatch" as const,
+      unitType: "execute-task",
+      unitId: "M001/S01/T01",
+      prompt: "execute the task",
+      matchedRule: "executing execute-task",
+    }),
+  });
+  const ic = makeIC(deps, {
+    s: {
+      ...makeSession(),
+      basePath: base,
+      originalBasePath: base,
+    } as any,
+  });
+  const preData: PreDispatchData = {
+    state: {
+      phase: "executing",
+      activeMilestone: { id: "M001", title: "Test", status: "active" },
+      activeSlice: { id: "S01", title: "Slice" },
+      activeTask: { id: "T01", title: "First task" },
+      registry: [{ id: "M001", status: "active" }],
+      blockers: [],
+    } as any,
+    mid: "M001",
+    midTitle: "Test Milestone",
+  };
+  const loopState: LoopState = {
+    recentUnits: [
+      { key: "execute-task/M001/S01/T01" },
+      { key: "execute-task/M001/S01/T01" },
+    ],
+    stuckRecoveryAttempts: 0,
+    consecutiveFinalizeTimeouts: 0,
+  };
+
+  const result = await runDispatch(ic, preData, loopState);
+
+  assert.equal(result.action, "continue");
+  assert.equal(pauseCalls, 0, "closed DB task should not pause auto-mode");
+  assert.equal(stopCalls, 0, "closed DB task should not hard-stop the loop");
+  assert.equal(invalidateCalls, 1, "closed DB task recovery should invalidate caches once");
+  assert.deepEqual(loopState.recentUnits, [], "closed DB task recovery should clear the stuck window");
+  assert.equal(loopState.stuckRecoveryAttempts, 0, "closed DB task recovery should reset the recovery counter");
+});
+
 test("runDispatch clears stuck state after Level 1 artifact recovery", async (t) => {
   const capture = createEventCapture();
   let invalidateCalls = 0;

package/src/resources/extensions/gsd/tests/memory-decay-factor.test.ts

@@ -0,0 +1,90 @@
+// gsd-2 / memoryDecayFactor unit tests
+//
+// Pure-function boundary tests for the V28 time-decay scoring helper.
+// The function maps last_hit_at → multiplier in [0.7, 1.0] used by
+// queryMemoriesRanked to down-weight stale memories without fully suppressing
+// them. These tests pin the contract:
+//
+//   - null / invalid / future timestamps → 1.0 (no decay penalty)
+//   - 0 days ago → 1.0
+//   - linear decay between 0 and 90 days
+//   - 90+ days ago → 0.7 floor
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { memoryDecayFactor } from "../memory-store.ts";
+
+const DAY_MS = 86_400_000;
+
+function isoDaysAgo(days: number): string {
+  return new Date(Date.now() - days * DAY_MS).toISOString();
+}
+
+test("memoryDecayFactor: null lastHitAt returns 1.0 (never-hit = no decay)", () => {
+  assert.equal(memoryDecayFactor(null), 1.0);
+});
+
+test("memoryDecayFactor: invalid timestamp string returns 1.0 (defensive)", () => {
+  assert.equal(memoryDecayFactor("not-a-date"), 1.0);
+  assert.equal(memoryDecayFactor(""), 1.0);
+});
+
+test("memoryDecayFactor: future timestamp clamps to daysAgo=0 → 1.0", () => {
+  // Clock skew or manual DB edits can yield future last_hit_at values.
+  // The factor must stay within [0.7, 1.0] regardless.
+  const future = new Date(Date.now() + 30 * DAY_MS).toISOString();
+  const factor = memoryDecayFactor(future);
+  assert.ok(factor <= 1.0, `factor must not exceed 1.0, got ${factor}`);
+  assert.ok(factor >= 0.7, `factor must not fall below 0.7, got ${factor}`);
+  assert.equal(factor, 1.0);
+});
+
+test("memoryDecayFactor: 0 days ago returns 1.0", () => {
+  const factor = memoryDecayFactor(new Date().toISOString());
+  // Tiny clock drift between now-string and Date.now() inside the function;
+  // assert it's effectively 1.0 within float tolerance.
+  assert.ok(Math.abs(factor - 1.0) < 1e-6, `expected ≈1.0, got ${factor}`);
+});
+
+test("memoryDecayFactor: 30 days ago returns ~0.90 (linear midpoint)", () => {
+  // Formula: 1.0 - 0.3 * (30/90) = 1.0 - 0.1 = 0.90
+  const factor = memoryDecayFactor(isoDaysAgo(30));
+  assert.ok(Math.abs(factor - 0.90) < 1e-3, `expected ≈0.90, got ${factor}`);
+});
+
+test("memoryDecayFactor: 60 days ago returns ~0.80", () => {
+  // Formula: 1.0 - 0.3 * (60/90) = 1.0 - 0.2 = 0.80
+  const factor = memoryDecayFactor(isoDaysAgo(60));
+  assert.ok(Math.abs(factor - 0.80) < 1e-3, `expected ≈0.80, got ${factor}`);
+});
+
+test("memoryDecayFactor: 90 days ago returns 0.70 (floor)", () => {
+  const factor = memoryDecayFactor(isoDaysAgo(90));
+  assert.ok(Math.abs(factor - 0.70) < 1e-3, `expected ≈0.70, got ${factor}`);
+});
+
+test("memoryDecayFactor: 180 days ago stays at 0.70 floor", () => {
+  const factor = memoryDecayFactor(isoDaysAgo(180));
+  assert.equal(factor, 0.70);
+});
+
+test("memoryDecayFactor: result always in [0.7, 1.0] for any input", () => {
+  const samples: (string | null)[] = [
+    null,
+    "",
+    "garbage",
+    new Date(0).toISOString(),
+    isoDaysAgo(0),
+    isoDaysAgo(15),
+    isoDaysAgo(45),
+    isoDaysAgo(89),
+    isoDaysAgo(91),
+    isoDaysAgo(365),
+    new Date(Date.now() + 365 * DAY_MS).toISOString(),
+  ];
+  for (const s of samples) {
+    const f = memoryDecayFactor(s);
+    assert.ok(f >= 0.7 && f <= 1.0, `factor out of [0.7, 1.0] for ${s}: ${f}`);
+  }
+});

package/src/resources/extensions/gsd/tests/migrate-writer-integration.test.ts

@@ -13,6 +13,9 @@ import { parseRoadmap, parsePlan } from '../parsers-legacy.ts';
 import { parseSummary } from '../files.ts';
 import { deriveState } from '../state.ts';
 import { invalidateAllCaches } from '../cache.ts';
+import { ensureDbOpen } from '../bootstrap/dynamic-tools.ts';
+import { closeDatabase, getAllMilestones } from '../gsd-db.ts';
+import { importWrittenMigrationToDb } from '../migrate/command.ts';
 import type {
   GSDProject,
   GSDMilestone,
@@ -250,6 +253,51 @@ test('Scenario 1: Incomplete project — write, parse, deriveState', async () =>
     }
 });
 
+test('Scenario 1b: written migration imports into authoritative DB state', async () => {
+    const base = mkdtempSync(join(tmpdir(), 'gsd-writer-db-import-'));
+    try {
+      const project = buildIncompleteProject();
+      await writeGSDDirectory(project, base);
+
+      assert.equal(await ensureDbOpen(base), true, 'db import: ensureDbOpen creates authoritative DB');
+
+      invalidateAllCaches();
+      const before = await deriveState(base);
+      assert.equal(before.activeMilestone, null, 'db import: markdown-only migration is invisible before DB import');
+
+      const imported = await importWrittenMigrationToDb(base);
+      assert.deepStrictEqual(imported.hierarchy, { milestones: 1, slices: 2, tasks: 3 }, 'db import: hierarchy counts');
+
+      invalidateAllCaches();
+      const after = await deriveState(base);
+      assert.deepStrictEqual(after.phase, 'executing', 'db import: deriveState sees imported DB hierarchy');
+      assert.deepStrictEqual(after.activeMilestone?.id, 'M001', 'db import: active milestone');
+      assert.deepStrictEqual(after.activeSlice?.id, 'S02', 'db import: active slice');
+      assert.deepStrictEqual(after.activeTask?.id, 'T03', 'db import: active task');
+    } finally {
+      closeDatabase();
+      rmSync(base, { recursive: true, force: true });
+    }
+});
+
+test('Scenario 1c: DB import verification fails when preview counts do not match', async () => {
+    const base = mkdtempSync(join(tmpdir(), 'gsd-writer-db-check-'));
+    try {
+      const project = buildIncompleteProject();
+      await writeGSDDirectory(project, base);
+
+      const preview = generatePreview(project);
+      await assert.rejects(
+        () => importWrittenMigrationToDb(base, { ...preview, totalTasks: preview.totalTasks + 1 }),
+        /migration DB import verification failed: tasks 3\/4/,
+      );
+      assert.deepStrictEqual(getAllMilestones(), [], 'db import: failed verification rolls back hierarchy rewrite');
+    } finally {
+      closeDatabase();
+      rmSync(base, { recursive: true, force: true });
+    }
+});
+
   // ─── Scenario 2: Fully complete project ────────────────────────────────
 
 test('Scenario 2: Fully complete project — deriveState phase', async () => {

package/src/resources/extensions/gsd/tests/migration-auto-check.test.ts

@@ -0,0 +1,127 @@
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import test from "node:test";
+
+import { ensureDbOpen } from "../bootstrap/dynamic-tools.ts";
+import {
+  _getAdapter,
+  closeDatabase,
+  getAllMilestones,
+  getSliceTasks,
+} from "../gsd-db.ts";
+import {
+  autoImportMarkdownHierarchyIfDbMismatch,
+  countMarkdownHierarchy,
+} from "../migration-auto-check.ts";
+import { writeGSDDirectory } from "../migrate/writer.ts";
+import type { GSDProject } from "../migrate/types.ts";
+
+function makeBase(): string {
+  return mkdtempSync(join(tmpdir(), "gsd-migration-auto-check-"));
+}
+
+function cleanup(base: string): void {
+  closeDatabase();
+  rmSync(base, { recursive: true, force: true });
+}
+
+function projectFixture(): GSDProject {
+  return {
+    projectContent: "# Legacy Project\n",
+    decisionsContent: "",
+    requirements: [],
+    milestones: [
+      {
+        id: "M001",
+        title: "Legacy Milestone",
+        vision: "Carry forward previous work",
+        successCriteria: ["Existing task is visible"],
+        research: null,
+        boundaryMap: [],
+        slices: [
+          {
+            id: "S01",
+            title: "Legacy Slice",
+            risk: "medium",
+            depends: [],
+            done: false,
+            demo: "Legacy slice demo",
+            goal: "Legacy slice demo",
+            research: null,
+            summary: null,
+            tasks: [
+              {
+                id: "T01",
+                title: "Legacy Task",
+                description: "Task carried from markdown",
+                done: false,
+                estimate: "",
+                files: ["src/index.ts"],
+                mustHaves: [],
+                summary: null,
+              },
+            ],
+          },
+        ],
+      },
+    ],
+  };
+}
+
+test("migration auto-check imports markdown hierarchy when DB is empty", async () => {
+  const base = makeBase();
+  try {
+    await writeGSDDirectory(projectFixture(), base);
+    assert.deepEqual(countMarkdownHierarchy(base), { milestones: 1, slices: 1, tasks: 1 });
+
+    assert.equal(await ensureDbOpen(base), true);
+    assert.equal(getAllMilestones().length, 0, "fresh authoritative DB starts empty");
+
+    const result = await autoImportMarkdownHierarchyIfDbMismatch(base);
+    assert.equal(result.action, "imported");
+    assert.equal(result.reason, "db-empty");
+    assert.deepEqual(result.afterDb, { milestones: 1, slices: 1, tasks: 1 });
+    assert.equal(getAllMilestones().length, 1);
+    assert.equal(getSliceTasks("M001", "S01").length, 1);
+  } finally {
+    cleanup(base);
+  }
+});
+
+test("migration auto-check repairs DB hierarchy count mismatch", async () => {
+  const base = makeBase();
+  try {
+    await writeGSDDirectory(projectFixture(), base);
+    await autoImportMarkdownHierarchyIfDbMismatch(base);
+
+    _getAdapter()!.prepare("DELETE FROM tasks WHERE milestone_id = ? AND slice_id = ? AND id = ?").run("M001", "S01", "T01");
+    assert.equal(getSliceTasks("M001", "S01").length, 0, "test fixture simulates stale DB task count");
+
+    const result = await autoImportMarkdownHierarchyIfDbMismatch(base);
+    assert.equal(result.action, "imported");
+    assert.equal(result.reason, "count-mismatch");
+    assert.deepEqual(result.beforeDb, { milestones: 1, slices: 1, tasks: 0 });
+    assert.deepEqual(result.afterDb, { milestones: 1, slices: 1, tasks: 1 });
+    assert.equal(getSliceTasks("M001", "S01").length, 1);
+  } finally {
+    cleanup(base);
+  }
+});
+
+test("migration auto-check leaves matching DB hierarchy alone", async () => {
+  const base = makeBase();
+  try {
+    await writeGSDDirectory(projectFixture(), base);
+    await autoImportMarkdownHierarchyIfDbMismatch(base);
+
+    const result = await autoImportMarkdownHierarchyIfDbMismatch(base);
+    assert.equal(result.action, "none");
+    assert.equal(result.reason, "in-sync");
+    assert.deepEqual(result.markdown, { milestones: 1, slices: 1, tasks: 1 });
+    assert.deepEqual(result.afterDb, { milestones: 1, slices: 1, tasks: 1 });
+  } finally {
+    cleanup(base);
+  }
+});

package/src/resources/extensions/gsd/tests/pre-execution-checks.test.ts

@@ -1,3 +1,6 @@
+// Project/App: GSD-2
+// File Purpose: Unit tests for pre-execution validation checks.
+
 /**
  * pre-execution-checks.test.ts — Unit tests for pre-execution validation checks.
  *
@@ -1542,6 +1545,27 @@ describe("checkFilePathConsistency directory inputs (#4446)", () => {
     assert.equal(results[0].blocking, true);
   });
 
+  test("runtime directory annotation is skipped as a pre-execution file dependency", (t) => {
+    const tempDir = join(tmpdir(), `pre-exec-dir-runtime-${Date.now()}`);
+    mkdirSync(tempDir, { recursive: true });
+    t.after(() => rmSync(tempDir, { recursive: true, force: true }));
+
+    const tasks = [
+      createTask({
+        id: "T02",
+        inputs: ["entries/ directory (runtime)"],
+        expected_output: ["src/commands/delete.ts", "src/index.ts"],
+      }),
+    ];
+
+    const results = checkFilePathConsistency(tasks, tempDir);
+    assert.deepEqual(
+      results,
+      [],
+      "Runtime-only directory inputs are created during command execution, not required before the task starts",
+    );
+  });
+
   test("tilde-prefixed input is matched against $HOME, not the project basePath", (t) => {
     const fakeHome = join(tmpdir(), `pre-exec-tilde-home-${Date.now()}`);
     const projectDir = join(tmpdir(), `pre-exec-tilde-proj-${Date.now()}`);
@@ -1597,6 +1621,20 @@ describe("checkTaskOrdering directory inputs (#4446)", () => {
       "Directory reference should not be treated as reading a file created later",
     );
   });
+
+  test("runtime directory annotation does not produce an ordering violation", () => {
+    const tasks = [
+      createTask({
+        id: "T02",
+        sequence: 0,
+        inputs: ["entries/ directory (runtime)"],
+        expected_output: [],
+      }),
+    ];
+
+    const results = checkTaskOrdering(tasks, "/tmp");
+    assert.deepEqual(results, []);
+  });
 });
 
 // ─── Regression Tests: checkTaskOrdering false positive for pre-execution refs (#4071) ──

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

@@ -0,0 +1,40 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, readdirSync } from "node:fs";
+import { join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { dirname } from "node:path";
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+const promptsDir = join(__dirname, "..", "prompts");
+
+test("prompt templates do not reference legacy milestone-root .gsd paths", () => {
+  const offenders: string[] = [];
+  for (const file of readdirSync(promptsDir)) {
+    if (!file.endsWith(".md")) continue;
+    const content = readFileSync(join(promptsDir, file), "utf-8");
+    const legacyPatterns = [
+      /\.gsd\/\{\{(?:milestoneId|mid)\}\}\//g,
+      /\.gsd\/<milestone-id>\//g,
+      /\.gsd\/<ID>\//g,
+    ];
+    for (const pattern of legacyPatterns) {
+      if (pattern.test(content)) {
+        offenders.push(`${file}: ${pattern.source}`);
+      }
+    }
+  }
+
+  assert.deepEqual(
+    offenders,
+    [],
+    "Milestone artifacts must use .gsd/milestones/<MID>/..., not legacy .gsd/<MID>/...",
+  );
+});
+
+test("quick task prompt delegates commit policy to quick.ts", () => {
+  const content = readFileSync(join(promptsDir, "quick-task.md"), "utf-8");
+  assert.match(content, /\{\{commitInstruction\}\}/);
+  assert.doesNotMatch(content, /Stage only relevant files/);
+});

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

@@ -65,6 +65,25 @@ describe('prompt step ordering (#3696)', () => {
     assert.ok(learningsIdx < completeMilestoneIdx, 'learnings extraction must happen before gsd_complete_milestone');
   });
 
+  test('complete-milestone duplicate guard checks milestone status before durable writes', () => {
+    const guardMatch = completeMilestoneMd.match(/^\d+\.\s.*gsd_milestone_status/m);
+    const reqUpdateMatch = completeMilestoneMd.match(/^\d+\.\s.*gsd_requirement_update/m);
+    assert.ok(guardMatch, 'complete-milestone must start with a gsd_milestone_status duplicate guard');
+    assert.ok(reqUpdateMatch, 'gsd_requirement_update should appear in a numbered step');
+
+    const guardIdx = completeMilestoneMd.indexOf(guardMatch![0]);
+    const reqUpdateIdx = completeMilestoneMd.indexOf(reqUpdateMatch![0]);
+    assert.ok(
+      guardIdx < reqUpdateIdx,
+      'duplicate guard must run before requirement/project/learnings writes',
+    );
+    assert.match(
+      completeMilestoneMd,
+      /status(?:`|\*\*)?\s+(?:is\s+)?(?:`complete`|"complete")/i,
+      'duplicate guard must tell the agent to stop when status is complete',
+    );
+  });
+
   test('complete-slice.md uses gsd_requirement_update', () => {
     assert.match(completeSliceMd, /gsd_requirement_update/,
       'complete-slice.md should reference gsd_requirement_update');