@opengsd/gsd-pi 1.2.0-dev.9ad8ae33 → 1.2.0-dev.a6376d75

package/src/resources/extensions/gsd/tests/auto-unit-closeout.test.ts

@@ -1,13 +1,23 @@
 // Project/App: gsd-pi
 // File Purpose: Regression tests for auto-unit closeout activity classification.
 
-import test from "node:test";
+import { mkdtempSync, writeFileSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { execSync } from "node:child_process";
+import { join } from "node:path";
+import test, { type TestContext } from "node:test";
 import assert from "node:assert/strict";
 
 import {
   isSuspiciousGhostCompletion,
   snapshotUnitActivity,
 } from "../auto-unit-closeout.ts";
+import {
+  closeUnit,
+  type UnitCloseoutRequest,
+  type UnitCloseoutDeps,
+} from "../unit-closeout.ts";
+import type { NotifySeverity } from "../notification-store.js";
 
 function makeCtx(entries: unknown[]) {
   return {
@@ -17,6 +27,38 @@ function makeCtx(entries: unknown[]) {
   } as any;
 }
 
+function createTempGitRepo(t: TestContext): string {
+  const dir = mkdtempSync(join(tmpdir(), "closeout-test-"));
+  t.after(() => rmSync(dir, { recursive: true, force: true }));
+  execSync("git init", { cwd: dir });
+  execSync("git config user.email test@test.com", { cwd: dir });
+  execSync("git config user.name Test", { cwd: dir });
+  return dir;
+}
+
+function makeMockDeps(overrides: Partial<UnitCloseoutDeps> = {}): UnitCloseoutDeps & { notifications: Array<{ message: string; severity: NotifySeverity }> } {
+  const notifications: Array<{ message: string; severity: NotifySeverity }> = [];
+  return {
+    isolationMode: () => "none",
+    currentBranch: () => "main",
+    commit: () => null,
+    notify: (message, severity) => notifications.push({ message, severity }),
+    notifications,
+    ...overrides,
+  };
+}
+
+function baseRequest(overrides: Partial<UnitCloseoutRequest> = {}): UnitCloseoutRequest {
+  return {
+    basePath: "/tmp/ignored",
+    unitType: "complete-milestone",
+    unitId: "M001",
+    boundary: "milestone",
+    outcome: "complete",
+    ...overrides,
+  };
+}
+
 test("isSuspiciousGhostCompletion rejects fast completions with no assistant output or tools", () => {
   const startedAt = Date.now();
   const ctx = makeCtx([]);
@@ -66,3 +108,129 @@ test("snapshotUnitActivity counts assistant messages and tool calls", () => {
     assistantMessages: 1,
   });
 });
+
+test("closeUnit: nothing-to-commit / isolation none", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "none",
+    commit: () => null,
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "nothing-to-commit");
+  assert.equal(result.commitMessage, null);
+  assert.equal(result.notice, undefined);
+  assert.equal(deps.notifications.length, 0);
+});
+
+test("closeUnit: committed / isolation none", (t) => {
+  const dir = createTempGitRepo(t);
+  writeFileSync(join(dir, "file.txt"), "hello");
+  execSync("git add file.txt", { cwd: dir });
+
+  const deps = makeMockDeps({
+    isolationMode: () => "none",
+    commit: () => "Add file.txt",
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "committed");
+  assert.equal(result.commitMessage, "Add file.txt");
+  assert.equal(result.notice, undefined);
+  assert.equal(deps.notifications.length, 0);
+});
+
+test("closeUnit: milestone-branch / worktree isolation", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "worktree",
+    currentBranch: () => "milestone/M001",
+    commit: () => "Complete M001",
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "milestone-branch");
+  assert.equal(result.commitMessage, "Complete M001");
+  assert.ok(result.notice);
+  assert.equal(deps.notifications.length, 1);
+  assert.equal(deps.notifications[0].severity, "info");
+  assert.match(deps.notifications[0].message, /Merge it to the integration branch/);
+});
+
+test("closeUnit: isolation-bypassed / worktree isolation", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "worktree",
+    currentBranch: () => "main",
+    commit: () => "Complete M001",
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "isolation-bypassed");
+  assert.equal(result.commitMessage, "Complete M001");
+  assert.ok(result.notice);
+  assert.equal(deps.notifications.length, 1);
+  assert.equal(deps.notifications[0].severity, "warning");
+  assert.match(deps.notifications[0].message, /completed outside a milestone worktree\/branch/);
+});
+
+test("closeUnit: isolation-bypassed / branch isolation", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "branch",
+    currentBranch: () => "main",
+    commit: () => "Complete M001",
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "isolation-bypassed");
+  assert.equal(result.commitMessage, "Complete M001");
+  assert.ok(result.notice);
+  assert.equal(deps.notifications.length, 1);
+  assert.equal(deps.notifications[0].severity, "warning");
+  assert.match(deps.notifications[0].message, /completed outside a milestone worktree\/branch/);
+});
+
+test("closeUnit: commit-failed", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "none",
+    commit: () => {
+      throw new Error("simulated commit failure");
+    },
+  });
+
+  const result = closeUnit(baseRequest({ basePath: dir, boundary: "milestone" }), deps);
+
+  assert.equal(result.gitVerdict, "commit-failed");
+  assert.equal(result.commitMessage, null);
+  assert.ok(result.notice);
+  assert.match(result.notice!, /simulated commit failure/);
+  assert.equal(deps.notifications.length, 1);
+  assert.equal(deps.notifications[0].severity, "error");
+});
+
+test("closeUnit: task boundary ignores isolation verdict", (t) => {
+  const dir = createTempGitRepo(t);
+  const deps = makeMockDeps({
+    isolationMode: () => "worktree",
+    currentBranch: () => "main",
+    commit: () => "Complete task",
+  });
+
+  const result = closeUnit(
+    baseRequest({ basePath: dir, boundary: "task", unitType: "execute-task", unitId: "M001/S01/T01" }),
+    deps,
+  );
+
+  assert.equal(result.gitVerdict, "committed");
+  assert.equal(result.commitMessage, "Complete task");
+  assert.equal(result.notice, undefined);
+  assert.equal(deps.notifications.length, 0);
+});
+

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

@@ -80,6 +80,20 @@ function createTempProject(): { basePath: string; planPath: string } {
   return { basePath, planPath };
 }
 
+function writeProjectPreferences(basePath: string, yaml: string): void {
+  fs.writeFileSync(path.join(basePath, '.gsd', 'PREFERENCES.md'), `---\n${yaml}---\n`);
+}
+
+async function withWorkingDirectory<T>(cwd: string, action: () => Promise<T>): Promise<T> {
+  const previousCwd = process.cwd();
+  process.chdir(cwd);
+  try {
+    return await action();
+  } finally {
+    process.chdir(previousCwd);
+  }
+}
+
 function makeValidParams() {
   return {
     taskId: 'T01',
@@ -104,6 +118,13 @@ function makeValidParams() {
   };
 }
 
+function makeEscalationOptions() {
+  return [
+    { id: 'continue', label: 'Continue', tradeoffs: 'Keeps execution moving with the default path.' },
+    { id: 'pause', label: 'Pause', tradeoffs: 'Stops execution until the blocker is reviewed.' },
+  ];
+}
+
 // ═══════════════════════════════════════════════════════════════════════════
 // complete-task: Fresh DB is migrated to the current schema version
 // ═══════════════════════════════════════════════════════════════════════════
@@ -359,6 +380,105 @@ console.log('\n=== complete-task: handler happy path ===');
   cleanup(dbPath);
 }
 
+// ═══════════════════════════════════════════════════════════════════════════
+// complete-task: hard-blocker escalation with mid-execution escalation disabled
+// ═══════════════════════════════════════════════════════════════════════════
+
+console.log('\n=== complete-task: disabled hard-blocker escalation rolls back completion ===');
+{
+  const dbPath = tempDbPath();
+  openDatabase(dbPath);
+
+  const { basePath } = createTempProject();
+  writeProjectPreferences(basePath, 'phases:\n  mid_execution_escalation: false\n');
+
+  insertMilestone({ id: 'M001', title: 'Test Milestone' });
+  insertSlice({ id: 'S01', milestoneId: 'M001', title: 'Test Slice' });
+
+  const params = {
+    ...makeValidParams(),
+    blockerDiscovered: true,
+    escalation: {
+      question: 'Should execution pause for the hard blocker?',
+      options: makeEscalationOptions(),
+      recommendation: 'pause',
+      recommendationRationale: 'The blocker should not be silently advanced.',
+      continueWithDefault: false,
+    },
+  };
+
+  const result = await withWorkingDirectory(basePath, () => handleCompleteTask(params, basePath));
+
+  assertTrue('error' in result, 'hard-blocker escalation should fail when escalation handling is disabled');
+  if ('error' in result) {
+    assertMatch(result.error, /hard-blocker escalation/, 'error should mention hard-blocker escalation');
+    assertMatch(result.error, /mid_execution_escalation is disabled/, 'error should mention disabled preference');
+  }
+
+  const task = getTask('M001', 'S01', 'T01');
+  assertTrue(task !== null, 'task row should remain after rollback');
+  assertEq(task!.status, 'pending', 'task status should be rolled back to pending');
+  assertEq(task!.blocker_discovered, true, 'blocker flag should remain recorded for visibility');
+  assertEq(task!.escalation_pending, 0, 'disabled preference should not create a pending escalation flag');
+  assertEq(task!.escalation_awaiting_review, 0, 'disabled preference should not create an awaiting-review flag');
+
+  const adapter = _getAdapter()!;
+  const evRows = adapter.prepare(
+    "SELECT * FROM verification_evidence WHERE task_id = 'T01' AND slice_id = 'S01' AND milestone_id = 'M001'"
+  ).all();
+  assertEq(evRows.length, 0, 'verification evidence should be deleted when completion rolls back');
+
+  cleanupDir(basePath);
+  cleanup(dbPath);
+}
+
+// ═══════════════════════════════════════════════════════════════════════════
+// complete-task: soft escalation with mid-execution escalation disabled
+// ═══════════════════════════════════════════════════════════════════════════
+
+console.log('\n=== complete-task: disabled soft escalation still completes ===');
+{
+  const dbPath = tempDbPath();
+  openDatabase(dbPath);
+
+  const { basePath } = createTempProject();
+  writeProjectPreferences(basePath, 'phases:\n  mid_execution_escalation: false\n');
+
+  insertMilestone({ id: 'M001', title: 'Test Milestone' });
+  insertSlice({ id: 'S01', milestoneId: 'M001', title: 'Test Slice' });
+
+  const params = {
+    ...makeValidParams(),
+    escalation: {
+      question: 'Should execution continue with the default?',
+      options: makeEscalationOptions(),
+      recommendation: 'continue',
+      recommendationRationale: 'The default path is safe enough to continue.',
+      continueWithDefault: true,
+    },
+  };
+
+  const result = await withWorkingDirectory(basePath, () => handleCompleteTask(params, basePath));
+
+  assertTrue(!('error' in result), 'soft escalation should still complete when escalation handling is disabled');
+  if (!('error' in result)) {
+    assertTrue(!result.escalation, 'disabled preference should not return escalation metadata');
+  }
+
+  const task = getTask('M001', 'S01', 'T01');
+  assertTrue(task !== null, 'task row should exist');
+  assertEq(task!.status, 'complete', 'soft escalation should leave task complete');
+
+  const adapter = _getAdapter()!;
+  const evRows = adapter.prepare(
+    "SELECT * FROM verification_evidence WHERE task_id = 'T01' AND slice_id = 'S01' AND milestone_id = 'M001'"
+  ).all();
+  assertEq(evRows.length, 1, 'verification evidence should remain for soft escalation completion');
+
+  cleanupDir(basePath);
+  cleanup(dbPath);
+}
+
 // ═══════════════════════════════════════════════════════════════════════════
 // complete-task: Handler validation errors
 // ═══════════════════════════════════════════════════════════════════════════
@@ -414,6 +534,9 @@ console.log('\n=== complete-task: handler idempotency ===');
   // First call should succeed
   const r1 = await handleCompleteTask(params, basePath);
   assertTrue(!('error' in r1), 'first call should succeed');
+  if ('error' in r1) {
+    throw new Error(r1.error);
+  }
 
   // Verify complete-task did not duplicate T01. S01-PLAN.md is a projection,
   // so the remaining plan task is not imported implicitly.
@@ -421,16 +544,29 @@ console.log('\n=== complete-task: handler idempotency ===');
   assertEq(tasks.length, 1, 'should only have the completed DB task after first call');
   assertEq(tasks.filter(t => t.id === 'T01').length, 1, 'should have exactly one T01 row after first call');
 
-  // Second call with same params — state machine guard rejects (task is already complete)
+  // If the DB row is complete but the projection was lost, the duplicate call
+  // should repair the missing summary from full_summary_md instead of forcing a
+  // reopen/re-complete loop.
+  fs.unlinkSync(r1.summaryPath);
+  assertTrue(!fs.existsSync(r1.summaryPath), 'fixture should remove the task summary before repair');
   const r2 = await handleCompleteTask(params, basePath);
-  assertTrue('error' in r2, 'second call should return error (task already complete)');
+  assertTrue(!('error' in r2), 'second call should repair missing summary for DB-complete task');
   if ('error' in r2) {
-    assertMatch(r2.error, /already complete/, 'error should mention already complete');
+    throw new Error(r2.error);
+  }
+  assertTrue(fs.existsSync(r2.summaryPath), 'missing summary should be restored on disk');
+  assertEq(r2.duplicate, true, 'repair should be reported as a duplicate/no-op state mutation');
+
+  // Third call with the summary present — state machine guard rejects (task is already complete)
+  const r3 = await handleCompleteTask(params, basePath);
+  assertTrue('error' in r3, 'third call should return error (task already complete)');
+  if ('error' in r3) {
+    assertMatch(r3.error, /already complete/, 'error should mention already complete');
   }
 
-  // Still no duplicate rows from the rejected second call.
+  // Still no duplicate rows from the repair or rejected third call.
   const tasksAfter = getSliceTasks('M001', 'S01');
-  assertEq(tasksAfter.length, 1, 'should still only have T01 after rejected second call');
+  assertEq(tasksAfter.length, 1, 'should still only have T01 after duplicate repair and rejected third call');
   assertEq(tasksAfter.filter(t => t.id === 'T01').length, 1, 'should still have exactly one T01 row');
 
   cleanupDir(basePath);

package/src/resources/extensions/gsd/tests/deep-project-auto-loop.test.ts

@@ -6,7 +6,8 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { randomUUID } from "node:crypto";
 
-import { runDispatch, runPreDispatch } from "../auto/phases.ts";
+import { runDispatch } from "../auto/dispatch.ts";
+import { runPreDispatch } from "../auto/pre-dispatch.ts";
 import { AutoSession } from "../auto/session.ts";
 import { resolveUnitSupervisionTimeouts } from "../auto-timers.ts";
 import { bootstrapAutoSession } from "../auto-start.ts";

package/src/resources/extensions/gsd/tests/derive-state-helpers.test.ts

@@ -569,6 +569,42 @@ describe('derive-state-helpers', () => {
     }
   });
 
+  // ─── Batch slice query: multi-milestone slices loaded in one query ─────
+  test('buildRegistryAndFindActive: batched slice query preserves ordering across milestones', async () => {
+    const base = createFixtureBase();
+    try {
+      // M001 is complete so the loop advances past it.
+      writeFile(base, 'milestones/M001/M001-ROADMAP.md', ROADMAP_CONTENT);
+      writeFile(base, 'milestones/M001/M001-SUMMARY.md', '# M001 Summary\n\nDone.');
+      // M002 is a queued shell milestone with no context and no slices — should be deferred.
+      mkdirSync(join(base, '.gsd', 'milestones', 'M002'), { recursive: true });
+      // M003 is the real active milestone with slices in non-trivial sequence order.
+      writeFile(base, 'milestones/M003/M003-CONTEXT.md', '# M003: Real\n\nReal milestone.');
+
+      openDatabase(':memory:');
+      insertMilestone({ id: 'M001', title: 'Complete', status: 'complete' });
+      insertMilestone({ id: 'M002', title: 'Shell', status: 'queued' });
+      insertMilestone({ id: 'M003', title: 'Real', status: 'active' });
+      // Slices intentionally inserted with out-of-order sequences to prove ordering.
+      insertSlice({ id: 'S01', milestoneId: 'M001', title: 'M1 Second', status: 'complete', risk: 'low', depends: [], sequence: 2 });
+      insertSlice({ id: 'S02', milestoneId: 'M001', title: 'M1 First', status: 'complete', risk: 'low', depends: [], sequence: 1 });
+      insertSlice({ id: 'S03', milestoneId: 'M003', title: 'M3 Second', status: 'active', risk: 'low', depends: [], sequence: 5 });
+      insertSlice({ id: 'S04', milestoneId: 'M003', title: 'M3 First', status: 'active', risk: 'low', depends: [], sequence: 3 });
+
+      invalidateStateCache();
+      const state = await deriveStateFromDb(base);
+
+      assert.equal(state.activeMilestone?.id, 'M003', 'batched: M003 is active after complete M001 and shell M002');
+      assert.equal(state.activeSlice?.id, 'S04', 'batched: first slice by sequence is active');
+      assert.equal(state.registry.find(e => e.id === 'M001')?.status, 'complete', 'batched: M001 complete');
+      assert.equal(state.registry.find(e => e.id === 'M002')?.status, 'pending', 'batched: shell M002 deferred to pending');
+      assert.equal(state.registry.find(e => e.id === 'M003')?.status, 'active', 'batched: M003 active');
+    } finally {
+      closeDatabase();
+      cleanup(base);
+    }
+  });
+
   // ─── Deferred queued shell: shell milestone deferred, real one promoted ──
   test('buildRegistryAndFindActive: queued shell deferred, later real milestone becomes active (#3470)', async () => {
     const base = createFixtureBase();

package/src/resources/extensions/gsd/tests/dispatch-history.test.ts

@@ -23,6 +23,7 @@ import { recordDispatchClaim, markFailed, markCanceled } from "../db/unit-dispat
 import {
   buildDispatchKey,
   createDispatchHistory,
+  lookupLatestLedgerError,
   normalizeDispatchKey,
   parseDispatchKey,
   STUCK_WINDOW_SIZE,
@@ -157,6 +158,31 @@ test("recordDispatch attaches the latest ledger error on repeats so repeat-error
   assert.match(verdict?.reason ?? "", /Same error repeated/);
 });
 
+test("lookupLatestLedgerError matches the bare unit id, not the compound key", (t) => {
+  const f = makeLedgerFixture(t);
+  const dispatchId = f.claim("execute-task", "M001/S01/T01");
+  markFailed(dispatchId, { errorSummary: "boom: deterministic failure" });
+
+  // The ledger keys rows by the bare unit id with the unit type in its own
+  // column. The shared lookup (also used by dispatch.ts's runDispatch path)
+  // must use the bare id; a compound `unitType/unitId` value misses entirely,
+  // which previously silently dropped repeat-error detection on that path.
+  assert.equal(
+    lookupLatestLedgerError("execute-task", "M001/S01/T01"),
+    "boom: deterministic failure",
+  );
+  assert.equal(
+    lookupLatestLedgerError("execute-task", "execute-task/M001/S01/T01"),
+    undefined,
+    "a compound key must not match the bare-id ledger row",
+  );
+  assert.equal(
+    lookupLatestLedgerError("plan-slice", "M001/S01/T01"),
+    undefined,
+    "a different unit type on the same id must not be attached",
+  );
+});
+
 test("recordDispatch never attaches another unit type's ledger error for the same unit id", (t) => {
   const f = makeLedgerFixture(t);
   const dispatchId = f.claim("plan-slice", "M001/S01");
@@ -262,6 +288,35 @@ test("#482 regression: a re-dispatch loop spanning a session restart is detected
   assert.match(verdict?.reason ?? "", /execute-task:M001\/S01\/T01 derived 3 consecutive times/);
 });
 
+test("rehydrate mirrors recordDispatch error attachment so repeat-error detection fires after the next dispatch", (t) => {
+  const f = makeLedgerFixture(t);
+  // Session 1: two failed dispatches for the same unit, same error summary.
+  for (let i = 0; i < 2; i++) {
+    const id = f.claim("execute-task", "M001/S01/T01");
+    markFailed(id, { errorSummary: "boom: deterministic failure" });
+  }
+
+  // Session 2: fresh history rehydrates from the ledger. Just like the live
+  // recordDispatch path, the first occurrence of a unit skips the ledger
+  // lookup; only repeats carry the error. So rehydration alone does not trip
+  // Rule 1 — keeping the post-restart window no more aggressive than the live
+  // one (Rule 1 has no retry-budget suppression).
+  const restarted = historyFor(f.base);
+  const count = restarted.rehydrate();
+  assert.equal(count, 2);
+  const window = restarted.getRecentWindow();
+  assert.equal(window[0].error, undefined);
+  assert.equal(window[1].error, "boom: deterministic failure");
+  assert.equal(restarted.detectStuck(), null);
+
+  // The next dispatch of the same unit attaches the error again, giving two
+  // consecutive matching errors → Rule 1 fires, exactly as in the live path.
+  restarted.recordDispatch("execute-task", "M001/S01/T01");
+  const verdict = restarted.detectStuck();
+  assert.equal(verdict?.stuck, true);
+  assert.match(verdict?.reason ?? "", /Same error repeated/);
+});
+
 test("rehydrate degrades to an empty window without a scope or ledger", () => {
   const noScope = historyFor(null);
   assert.equal(noScope.rehydrate(), 0);

package/src/resources/extensions/gsd/tests/dist-redirect.mjs

@@ -7,6 +7,10 @@ const require = createRequire(import.meta.url);
 const ROOT = new URL("../../../../../", import.meta.url);
 
 export function resolve(specifier, context, nextResolve) {
+  if (specifier.startsWith('node:')) {
+    return { url: specifier, format: 'builtin', shortCircuit: true };
+  }
+
   // 1. Redirect all workspace package bare imports to source.
   //    CI portability runs don't build any packages/ dist artifacts, so every
   //    @gsd/* specifier (including transitive ones pulled in by pi-coding-agent
@@ -100,6 +104,10 @@ export function resolve(specifier, context, nextResolve) {
 }
 
 export function load(url, context, nextLoad) {
+  if (url.startsWith('node:') || context.format === 'builtin') {
+    return { format: 'builtin', source: '', shortCircuit: true };
+  }
+
   // jiti/CJS may still enter through stale packages/*/dist/index.js — redirect to src.
   if (url.includes('/packages/pi-ai/dist/index.js')) {
     url = url.replace('/dist/index.js', '/src/index.ts');