kenaz 0.3.2 → 0.3.5

package/core/.formation/tools/worktree-leak.test.js

@@ -1,162 +0,0 @@
-/**
- * TASK_1530 — regression test for the Bifrost/Codex worktree WRITE-LEAK class.
- *
- * Confirmed external-user incident: a Codex (OpenAI adapter) dispatch wrote a
- * modified tracked file to the MAIN repo instead of the isolated worktree, then:
- *   1. spawnBifrost never set CLAUDE_PROJECT_DIR / never captured
- *      originalWorktreeHead → detectWorktreeLeak NEVER fired for Bifrost.
- *   2. detectWorktreeLeak was WARN-ONLY: it wrote a marker + console.error but
- *      did NOT salvage the leaked content nor block the dispatch, so the file
- *      sat in the main working tree where a later force-checkout could discard
- *      it, and the dispatch was still reported as a clean success.
- *
- * This test drives the REAL detectWorktreeLeak (exported from runner.js) against
- * a synthetic temp repo + worktree (no external repo needed), simulating a
- * dispatch that produced ZERO task-branch commits while the main repo has an
- * uncommitted TRACKED modification. It asserts:
- *   • the leak is SALVAGED — recoverable on a branch, and
- *   • the dispatch is BLOCKED — child._formationExitCode flips to 1.
- *
- * Revert-proof: if the runner.js change (salvage + block) is reverted, this test
- * fails (no salvage branch / exit code stays 0).
- *
- * Scaffolding mirrors ./post-complete/worktree-rescue.test.js (TASK_1529).
- */
-
-const test = require('node:test');
-const assert = require('node:assert/strict');
-const fs = require('fs');
-const os = require('os');
-const path = require('path');
-const { execSync } = require('child_process');
-
-const { detectWorktreeLeak } = require('./runner');
-
-function git(cmd, cwd) {
-  return execSync(cmd, { cwd, encoding: 'utf8', timeout: 20000, stdio: ['pipe', 'pipe', 'pipe'] }).trim();
-}
-function gitSafe(cmd, cwd) {
-  try { return git(cmd, cwd); } catch { return null; }
-}
-
-function makeGitRepo(dir) {
-  git('git init', dir);
-  git('git config user.email "test@test.com"', dir);
-  git('git config user.name "Test"', dir);
-  git('git config commit.gpgsign false', dir);
-  gitSafe('git checkout -b userwork', dir);
-  fs.writeFileSync(path.join(dir, 'README.md'), 'init\n');
-  git('git add README.md', dir);
-  git('git commit --no-verify -m "init"', dir);
-  const cur = git('git rev-parse --abbrev-ref HEAD', dir);
-  if (cur !== 'userwork') git(`git branch -m ${cur} userwork`, dir);
-}
-
-function setup() {
-  const tmpRoot = fs.realpathSync(fs.mkdtempSync(path.join(os.tmpdir(), 'tn1530-')));
-  const repo = path.join(tmpRoot, 'repo');
-  fs.mkdirSync(repo, { recursive: true });
-  makeGitRepo(repo);
-  return { tmpRoot, repo };
-}
-
-function teardown(tmpRoot, repo) {
-  try {
-    const wt = git('git worktree list --porcelain', repo);
-    for (const line of wt.split('\n')) {
-      if (line.startsWith('worktree ')) {
-        const p = line.slice('worktree '.length).trim();
-        if (p && path.resolve(p) !== path.resolve(repo)) {
-          gitSafe(`git worktree remove --force "${p}"`, repo);
-        }
-      }
-    }
-  } catch { /* best effort */ }
-  fs.rmSync(tmpRoot, { recursive: true, force: true });
-}
-
-// ─────────────────────────────────────────────────────────────────────────────
-// LEAK: zero task-branch commits + main repo has an uncommitted tracked change.
-// detectWorktreeLeak must SALVAGE the leaked change AND BLOCK the dispatch.
-// ─────────────────────────────────────────────────────────────────────────────
-
-test('TASK_1530: Bifrost write-leak (zero task commits + dirty main) is salvaged AND blocks dispatch', () => {
-  const { tmpRoot, repo } = setup();
-  try {
-    // Baseline tracked file the user cares about.
-    const lostContent = 'module.exports = { version: 2, critical: true };\n';
-    fs.writeFileSync(path.join(repo, 'kb.js'), 'module.exports = { version: 1 };\n');
-    git('git add kb.js', repo);
-    git('git commit --no-verify -m "seed kb.js baseline"', repo);
-
-    // Agent worktree on a task branch — snapshot HEAD BEFORE any agent work
-    // (exactly what spawnBifrost now captures as originalWorktreeHead).
-    const worktree = path.join(tmpRoot, 'wt');
-    git(`git worktree add -b task/task_repro1530 "${worktree}" HEAD`, repo);
-    const originalWorktreeHead = git('git rev-parse HEAD', worktree);
-
-    // THE BUG: the Codex agent produced ZERO commits on the task branch, and its
-    // modified tracked file landed in the MAIN repo (CLAUDE_PROJECT_DIR misroute).
-    fs.writeFileSync(path.join(repo, 'kb.js'), lostContent);
-
-    // Simulate runner.js's child object so the function can flip the exit code.
-    const child = { _formation: { sessionId: 'sess1530', skill: 'feature_agent' } };
-
-    // Drive the REAL detector against the synthetic repo (mainRoot = temp repo).
-    const res = detectWorktreeLeak(worktree, originalWorktreeHead, child, repo);
-
-    // 1) A leak must be reported.
-    assert.ok(res && res.leaked === true, `expected leak to be detected, got ${JSON.stringify(res)}`);
-
-    // 2) The dispatch must be BLOCKED: exit code flipped to 1.
-    assert.equal(child._formationExitCode, 1,
-      `expected child._formationExitCode=1 (blocked), got ${child._formationExitCode}`);
-    assert.ok(child._formationLeak, 'expected child._formationLeak to be set');
-
-    // 3) The leaked modification must be SALVAGED onto a recoverable branch.
-    assert.ok(res.salvageBranch, `expected a salvage branch, got ${res.salvageBranch}`);
-    const salvaged = gitSafe(`git show ${res.salvageBranch}:kb.js`, repo);
-    assert.equal(salvaged, lostContent.trim(),
-      `salvage branch must contain the leaked kb.js content. Got: ${JSON.stringify(salvaged)}`);
-
-    // 4) A LEAK_DETECTED marker is persisted in the worktree.
-    assert.ok(fs.existsSync(path.join(worktree, 'LEAK_DETECTED')), 'expected LEAK_DETECTED marker');
-  } finally {
-    teardown(tmpRoot, repo);
-  }
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// REGRESSION GUARD: a healthy dispatch (>=1 task-branch commit) must NOT be
-// flagged as a leak even if the main repo happens to be dirty.
-// ─────────────────────────────────────────────────────────────────────────────
-
-test('TASK_1530 regression: dispatch WITH task-branch commits is NOT flagged as leak', () => {
-  const { tmpRoot, repo } = setup();
-  try {
-    fs.writeFileSync(path.join(repo, 'kb.js'), 'v1\n');
-    git('git add kb.js', repo);
-    git('git commit --no-verify -m "seed"', repo);
-
-    const worktree = path.join(tmpRoot, 'wt');
-    git(`git worktree add -b task/task_ok1530 "${worktree}" HEAD`, repo);
-    const originalWorktreeHead = git('git rev-parse HEAD', worktree);
-
-    // Agent committed real work on the task branch (the healthy path).
-    fs.writeFileSync(path.join(worktree, 'feature.js'), 'export const x = 1;\n');
-    git('git add feature.js', worktree);
-    git('git commit --no-verify -m "[TASK_OK1530] add feature"', worktree);
-
-    // Main repo is incidentally dirty (e.g. unrelated state files).
-    fs.writeFileSync(path.join(repo, 'kb.js'), 'v1\nincidental\n');
-
-    const child = { _formation: { sessionId: 'sessok', skill: 'feature_agent' } };
-    const res = detectWorktreeLeak(worktree, originalWorktreeHead, child, repo);
-
-    assert.equal(res, null, `healthy dispatch must not be flagged as leak, got ${JSON.stringify(res)}`);
-    assert.equal(child._formationExitCode, undefined, 'exit code must remain unset for healthy dispatch');
-    assert.ok(!child._formationLeak, 'no leak record for healthy dispatch');
-  } finally {
-    teardown(tmpRoot, repo);
-  }
-});

package/core/.specialist/tools/__tests__/task-1863-boot-lease-reuse.test.js

@@ -1,99 +0,0 @@
-// TASK_1863 — boot ghost-lease fix: re-running ansuz-boot-fast.js within the SAME
-// conversation must reuse the existing stream lease (via stream-lease register's
-// findLeaseByConversationId reuse path), not mint a new boot-<ts>-<pid> ghost.
-// Isolated sandbox via CLAUDE_PROJECT_DIR (stream-lease resolveBase honors it);
-// never touches the real repo's leases.
-//
-// Arms:
-//   1  (AC#2): 3 consecutive boots, same conversation id -> readAllLeases must hold
-//      exactly ONE lease for that conversation; boots 2/3 must not print a
-//      [Multi-Ansuz] banner (no self-ghost).
-//   2  (control): a boot under a DIFFERENT conversation id still creates its own
-//      new lease (reuse is conversation-scoped, not a global singleton).
-//
-// Usage: node .specialist/tools/__tests__/task-1863-boot-lease-reuse.test.js
-// Exit 0=all pass, non-zero=at least one failure.
-
-const { spawnSync } = require('child_process');
-const fs = require('fs');
-const path = require('path');
-const os = require('os');
-
-function resolveRoot() {
-  var d = path.resolve(__dirname);
-  for (var i = 0; i < 10; i++) {
-    if (fs.existsSync(path.join(d, '.kenaz')) && fs.existsSync(path.join(d, '.specialist'))) return d;
-    var parent = path.dirname(d); if (parent === d) break; d = parent;
-  }
-  if (process.env.CLAUDE_PROJECT_DIR) return process.env.CLAUDE_PROJECT_DIR;
-  throw new Error('Cannot resolve repo root from ' + __dirname);
-}
-const WT = resolveRoot();
-const SB = path.join(os.tmpdir(), 't1863_boot_' + process.pid);
-const BOOT = path.join(WT, '.specialist/tools/ansuz-boot-fast.js');
-const STREAM_LEASE = path.join(WT, '.specialist/tools/stream-lease.js');
-const CONV1 = 'conv-t1863-boot';
-const CONV2 = 'conv-t1863-other';
-
-for (const d of [
-  '.kenaz/tasks/backlog', '.kenaz/tasks/active', '.kenaz/tasks/done',
-  '.kenaz/tasks/blocked', '.kenaz/tasks/cancel',
-  '.kenaz/specialist/sessions/leases',
-]) fs.mkdirSync(path.join(SB, d), { recursive: true });
-
-const env = { ...process.env, CLAUDE_PROJECT_DIR: SB };
-
-const boot = (convId) => spawnSync(process.execPath, [
-  BOOT,
-  '--project-path=' + SB,
-  '--conversation-id=' + convId,
-  '--mcp-base-url=http://127.0.0.1:1', // dead port: preamble arm fails instantly, offline-safe
-], { encoding: 'utf8', env, windowsHide: true, cwd: SB, timeout: 30000 });
-
-// Count via the exported readAllLeases (card: do NOT re-implement lease scanning).
-// Child process so resolveBase picks up CLAUDE_PROJECT_DIR=SB at module load.
-const readLeases = () => {
-  const r = spawnSync(process.execPath, ['-e',
-    'const sl=require(' + JSON.stringify(STREAM_LEASE) + ');process.stdout.write(JSON.stringify(sl.readAllLeases()));',
-  ], { encoding: 'utf8', env, windowsHide: true });
-  return JSON.parse(r.stdout || '[]');
-};
-
-let pass = 0, fail = 0;
-const check = (name, cond, detail) => {
-  console.log((cond ? 'PASS' : 'FAIL') + ' | ' + name + (cond ? '' : ' | ' + detail));
-  cond ? pass++ : fail++;
-};
-
-// ── Arm 1: 3 boots, same conversation ──
-const outs = [];
-for (let i = 0; i < 3; i++) {
-  const r = boot(CONV1);
-  outs.push((r.stdout || '') + (r.stderr || ''));
-  check('boot#' + (i + 1) + ' runs (banner printed)', (r.stdout || '').includes('Ansuz'),
-    'status=' + r.status + ' | ' + ((r.stdout || '') + (r.stderr || '')).slice(0, 300));
-}
-const all = readLeases();
-const mine = all.filter((l) => l.conversation_id === CONV1);
-check('AC#2: exactly ONE lease for the conversation after 3 boots', mine.length === 1,
-  'count=' + mine.length + ' sids=' + all.map((l) => l.session_id).join(','));
-check('boots 2+3 show no [Multi-Ansuz] self-ghost',
-  !outs[1].includes('[Multi-Ansuz]') && !outs[2].includes('[Multi-Ansuz]'),
-  'boot2/3 banner: ' + (outs[1] + outs[2]).split(String.fromCharCode(10)).filter((l) => l.includes('Multi-Ansuz') || l.includes('[live]') || l.includes('[stale]')).join(' / '));
-
-// ── Arm 2 (control): different conversation still gets its OWN new lease ──
-const r2 = boot(CONV2);
-const all2 = readLeases();
-const mine2 = all2.filter((l) => l.conversation_id === CONV2);
-const mine1 = all2.filter((l) => l.conversation_id === CONV1);
-check('control: new conversation creates a NEW lease (reuse is conversation-scoped)',
-  mine2.length === 1 && mine1.length === 1 && all2.length === 2,
-  'conv2=' + mine2.length + ' conv1=' + mine1.length + ' total=' + all2.length);
-check('control: second conversation SEES the first as a live peer (banner intact)',
-  ((r2.stdout || '')).includes('[Multi-Ansuz]'),
-  'no peer banner: ' + (r2.stdout || '').slice(0, 400));
-
-fs.rmSync(SB, { recursive: true, force: true });
-console.log('---');
-console.log('RESULT: ' + pass + ' pass, ' + fail + ' fail');
-process.exit(fail ? 1 : 0);

package/core/.specialist/tools/__tests__/task-1977-banner-cap.test.js

@@ -1,129 +0,0 @@
-// TASK_1977 R3/R4(d) — ansuz-boot-fast.js handoff banner fan-out cap.
-//
-// MEASURED 2026-09-20: a run of 19 accumulated handoff slots printed 33.8 KB
-// into the fast-boot banner, most repeating the same pending_decisions text.
-// Fix: sort newest-first, print at most 3 full slots (deduped by topic
-// prefix), fold the rest into one summary line. Archiving must still touch
-// EVERY slot regardless of what gets rendered.
-//
-// Arms:
-//   1. 5 distinct-topic fixture slots -> 3 printed in full + 1 summary line
-//      naming the other 2 topics; all 5 archived.
-//   2. dedup case: several slots share a topic prefix -> duplicates fold
-//      into the summary instead of eating a "printed" slot.
-//
-// Usage: node .specialist/tools/__tests__/task-1977-banner-cap.test.js
-// Exit 0=all pass, non-zero=at least one failure.
-
-const { spawnSync } = require('child_process');
-const fs = require('fs');
-const os = require('os');
-const path = require('path');
-
-function resolveRoot() {
-  let d = path.resolve(__dirname);
-  for (let i = 0; i < 10; i++) {
-    if (fs.existsSync(path.join(d, '.kenaz')) && fs.existsSync(path.join(d, '.specialist'))) return d;
-    const parent = path.dirname(d);
-    if (parent === d) break;
-    d = parent;
-  }
-  if (process.env.CLAUDE_PROJECT_DIR) return process.env.CLAUDE_PROJECT_DIR;
-  throw new Error('Cannot resolve repo root from ' + __dirname);
-}
-const WT = resolveRoot();
-const BOOT = path.join(WT, '.specialist/tools/ansuz-boot-fast.js');
-
-let pass = 0, fail = 0;
-function check(name, cond, detail) {
-  console.log((cond ? 'PASS' : 'FAIL') + ' | ' + name + (cond ? '' : ' | ' + detail));
-  cond ? pass++ : fail++;
-}
-
-function makeSandbox(prefix) {
-  const sb = path.join(os.tmpdir(), prefix + '_' + process.pid + '_' + Date.now());
-  for (const d of [
-    '.kenaz/tasks/backlog', '.kenaz/tasks/active', '.kenaz/tasks/done',
-    '.kenaz/tasks/blocked', '.kenaz/tasks/cancel',
-    '.kenaz/specialist/sessions/leases',
-    '.kenaz/specialist/handoffs',
-  ]) fs.mkdirSync(path.join(sb, d), { recursive: true });
-  return sb;
-}
-
-function writeSlot(sb, name, topic, generatedAt) {
-  const p = path.join(sb, '.kenaz/specialist/handoffs', name);
-  fs.writeFileSync(p, JSON.stringify({
-    generated_at: generatedAt,
-    isbar: {
-      situation: { current_topic: topic, pending_decisions: 'pending for ' + topic },
-      recommendation: { first_action: 'continue ' + topic },
-      assessment: { unfinished: 'unfinished ' + topic },
-    },
-  }));
-}
-
-function boot(sb) {
-  return spawnSync(process.execPath, [
-    BOOT,
-    '--project-path=' + sb,
-    '--conversation-id=t1977-banner-' + path.basename(sb),
-    '--mcp-base-url=http://127.0.0.1:1', // dead port: preamble arm fails instantly, offline-safe
-  ], { encoding: 'utf8', env: { ...process.env, CLAUDE_PROJECT_DIR: sb }, windowsHide: true, cwd: sb, timeout: 30000 });
-}
-
-// ── Arm 1: 5 distinct-topic slots, oldest to newest ──
-const sb1 = makeSandbox('t1977_banner_distinct');
-writeSlot(sb1, 'slot-1.json', 'Topic ONE',   '2026-09-15T00:00:00.000Z');
-writeSlot(sb1, 'slot-2.json', 'Topic TWO',   '2026-09-16T00:00:00.000Z');
-writeSlot(sb1, 'slot-3.json', 'Topic THREE', '2026-09-17T00:00:00.000Z');
-writeSlot(sb1, 'slot-4.json', 'Topic FOUR',  '2026-09-18T00:00:00.000Z');
-writeSlot(sb1, 'slot-5.json', 'Topic FIVE',  '2026-09-19T00:00:00.000Z'); // newest
-const r1 = boot(sb1);
-const out1 = (r1.stdout || '') + (r1.stderr || '');
-check('arm1: boot runs (banner printed)', out1.includes('Ansuz'), 'status=' + r1.status + ' | ' + out1.slice(0, 300));
-// TASK_2280: the banner prints String(slot_file).slice(0, 8) (TASK_1988 diet),
-// so "slot-5.json" shows as "[slot-5.j]" -- match the 8-char prefix, not the filename.
-const fullBlocks1 =(out1.match(/📜 Handoff \[slot-\d\.j\]:/g) || []).length;
-check('arm1: exactly 3 full slots printed', fullBlocks1 === 3, 'count=' + fullBlocks1 + '\n' + out1);
-check('arm1: newest 3 topics printed in full (FIVE, FOUR, THREE)',
-  out1.includes('Topic FIVE') && out1.includes('Topic FOUR') && out1.includes('Topic THREE'),
-  out1);
-check('arm1: one summary line for the remaining 2, naming their topics',
-  /2 more handoff slots archived \(topics: Topic TWO; Topic ONE\)/.test(out1),
-  out1);
-const archiveDir1 = path.join(sb1, '.kenaz/specialist/handoff_archive');
-const archived1 = fs.existsSync(archiveDir1) ? fs.readdirSync(archiveDir1) : [];
-check('arm1: ALL 5 slots archived regardless of banner selection', archived1.length === 5, 'archived=' + JSON.stringify(archived1));
-const remaining1 = fs.readdirSync(path.join(sb1, '.kenaz/specialist/handoffs'));
-check('arm1: handoffs dir drained after archiving', remaining1.length === 0, 'remaining=' + JSON.stringify(remaining1));
-
-// ── Arm 2: dedup -- 4 slots share a topic prefix, 1 is distinct ──
-const sb2 = makeSandbox('t1977_banner_dedup');
-const REPEATED_TOPIC = 'Repeated handoff topic used verbatim across several stale Codex sessions that keeps recurring';
-writeSlot(sb2, 'dup-1.json', REPEATED_TOPIC, '2026-09-16T00:00:00.000Z');
-writeSlot(sb2, 'dup-2.json', REPEATED_TOPIC, '2026-09-17T00:00:00.000Z');
-writeSlot(sb2, 'dup-3.json', REPEATED_TOPIC, '2026-09-18T00:00:00.000Z'); // newest of the dupes
-writeSlot(sb2, 'unique.json', 'A distinct one-off topic', '2026-09-15T00:00:00.000Z'); // oldest overall
-const r2 = boot(sb2);
-const out2 = (r2.stdout || '') + (r2.stderr || '');
-check('arm2: boot runs (banner printed)', out2.includes('Ansuz'), 'status=' + r2.status + ' | ' + out2.slice(0, 300));
-const fullBlocks2 = (out2.match(/📜 Handoff \[(dup-\d\.js|unique\.j)\]:/g) || []).length;
-check('arm2: only 2 full slots printed (one per distinct topic prefix, dupes folded)', fullBlocks2 === 2, 'count=' + fullBlocks2 + '\n' + out2);
-check('arm2: newest dupe (dup-3) is the one printed in full, not dup-1/dup-2',
-  out2.includes('📜 Handoff [dup-3.js]:') && !out2.includes('📜 Handoff [dup-1.js]:') && !out2.includes('📜 Handoff [dup-2.js]:'),
-  out2);
-check('arm2: the distinct topic is also printed in full', out2.includes('📜 Handoff [unique.j]:'), out2);
-check('arm2: summary line folds the 2 duplicate slots', /2 more handoff slots archived/.test(out2), out2);
-const archiveDir2 = path.join(sb2, '.kenaz/specialist/handoff_archive');
-const archived2 = fs.existsSync(archiveDir2) ? fs.readdirSync(archiveDir2) : [];
-check('arm2: ALL 4 slots archived regardless of dedup folding', archived2.length === 4, 'archived=' + JSON.stringify(archived2));
-
-// ── cleanup ──
-for (const sb of [sb1, sb2]) {
-  try { fs.rmSync(sb, { recursive: true, force: true }); } catch { /* best-effort */ }
-}
-
-console.log('---');
-console.log('RESULT: ' + pass + ' pass, ' + fail + ' fail');
-process.exit(fail > 0 ? 1 : 0);