test: review-army + adversarial test hardening

- Tripwire scans execFileSync too (ceiling 8: two more grep-needle string
  exemptions); merge-introduced timeout-less spawnSync in
  question-preference-hook fixed — the tripwire caught a site that landed
  on main AFTER the sweep, on its first day.
- gstack-detach gains TWO watchdog kill regression tests: TERM-immune
  grandchild (the killpg-after-grace escalation) and the leader-dies
  variant (the pgid-at-spawn fix — the case the first test cannot see).
- eval-flake-rank gets its unit suite (final-attempt accounting, artifact
  exclusion, shard recursion, recency bound).
- Groupkill/startup-grace shim markers are per-run unique (pid-suffixed
  sleep durations): sibling Conductor worktrees run free suites with no
  machine lock, and fixed markers let one run pgrep/pkill the other's
  shims — a cross-run flake inside the anti-flake tests.
- flake-ledger test pins the project-scoped local default; stale empty
  section headers in touchfiles-data deleted (they invited entries under
  deliberately retired categories).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

test/eval-flake-rank.test.ts

@@ -0,0 +1,70 @@
+/**
+ * Unit tests for the flake-rank aggregator (WS1's dial). The CLI ranks tests
+ * by retried passes (the flake signature) across finalized eval-store runs —
+ * these pin the accounting: N attempt records = 1 run of that test, the
+ * FINAL attempt decides pass/fail, retried passes count separately, partials
+ * and runner artifacts are excluded, shard dirs recurse, and the recency
+ * bound drops stale files.
+ */
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+import { aggregate, collectEvalFiles } from '../scripts/eval-flake-rank';
+
+const entry = (name: string, passed: boolean, attempt: number) => ({
+  name, suite: 's', tier: 'e2e', passed, attempt, duration_ms: 1000, cost_usd: 0.1,
+});
+
+const run = (tests: object[], extra: object = {}) => JSON.stringify({
+  schema_version: 2, version: '1.0.0', branch: 'b', git_sha: 'x', hostname: 'h',
+  timestamp: '2026-08-31T00:00:00Z', tier: 'e2e',
+  total_tests: tests.length, passed: 0, failed: 0, total_cost_usd: 0, total_duration_ms: 0,
+  tests, ...extra,
+});
+
+describe('eval-flake-rank aggregate', () => {
+  test('final attempt decides; retried pass counts as retriedPass, not a fail', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'flakerank-'));
+    fs.writeFileSync(path.join(dir, 'run1.json'), run([
+      entry('flaky', false, 1), entry('flaky', true, 2),   // pass on retry
+      entry('steady', true, 1),
+      entry('broken', false, 1), entry('broken', false, 2), // fails even retried
+    ]));
+    fs.writeFileSync(path.join(dir, 'run2.json'), run([
+      entry('flaky', true, 1), entry('steady', true, 1),
+    ]));
+    const series = aggregate(collectEvalFiles(dir));
+    expect(series.get('flaky')).toMatchObject({ runs: 2, passes: 2, fails: 0, retriedPasses: 1, totalAttempts: 3 });
+    expect(series.get('steady')).toMatchObject({ runs: 2, passes: 2, fails: 0, retriedPasses: 0 });
+    expect(series.get('broken')).toMatchObject({ runs: 1, passes: 0, fails: 1, retriedPasses: 0, totalAttempts: 2 });
+    fs.rmSync(dir, { recursive: true, force: true });
+  });
+
+  test('partials and runner artifacts are excluded; shard dirs recurse', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'flakerank-'));
+    fs.mkdirSync(path.join(dir, 'shards', 'slug-a'), { recursive: true });
+    fs.writeFileSync(path.join(dir, 'shards', 'slug-a', 'run.json'), run([entry('sharded', true, 1)]));
+    fs.writeFileSync(path.join(dir, '_partial-e2e.json'), run([entry('inflight', false, 1)], { _partial: true }));
+    fs.writeFileSync(path.join(dir, 'manifest.json'), '{"version":1}');
+    fs.writeFileSync(path.join(dir, 'slice-3.json'), '{"version":1}');
+    const files = collectEvalFiles(dir);
+    expect(files).toHaveLength(1);
+    const series = aggregate(files);
+    expect(series.has('sharded')).toBe(true);
+    expect(series.has('inflight')).toBe(false);
+    fs.rmSync(dir, { recursive: true, force: true });
+  });
+
+  test('recency bound drops files older than sinceDays', () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'flakerank-'));
+    const stale = path.join(dir, 'old.json');
+    fs.writeFileSync(stale, run([entry('ancient', true, 1)]));
+    const old = new Date(Date.now() - 90 * 86_400_000);
+    fs.utimesSync(stale, old, old);
+    fs.writeFileSync(path.join(dir, 'new.json'), run([entry('recent', true, 1)]));
+    const files = collectEvalFiles(dir, 60);
+    expect(files.map((f) => path.basename(f))).toEqual(['new.json']);
+    fs.rmSync(dir, { recursive: true, force: true });
+  });
+});

test/flake-ledger.test.ts

@@ -52,8 +52,13 @@ describe('flake ledger', () => {
     fs.rmSync(dir, { recursive: true, force: true });
   });
 
-  test('env override wins over the tmpdir default', () => {
+  test('env override wins; local default is project-scoped, never machine-global', () => {
     expect(flakeLedgerPath({ GSTACK_FLAKE_LEDGER: '/x/y.jsonl' } as NodeJS.ProcessEnv)).toBe('/x/y.jsonl');
-    expect(flakeLedgerPath({} as NodeJS.ProcessEnv)).toContain('gstack-flake-ledger.jsonl');
+    // Without the env override, the default lives under the PROJECT dir
+    // (sibling worktrees of different repos must not interleave one series);
+    // tmpdir is only the last-resort fallback when slug detection fails.
+    const local = flakeLedgerPath({} as NodeJS.ProcessEnv);
+    expect(local).toMatch(/flake-ledger\.jsonl$/);
+    expect(local.includes(path.join('.gstack', 'projects')) || local.startsWith(os.tmpdir())).toBe(true);
   });
 });

test/gstack-detach.test.ts

@@ -69,6 +69,59 @@ describe('gstack-detach', () => {
     } finally { fs.rmSync(dir, { recursive: true, force: true }); }
   }, 16000);
 
+  test('watchdog group-SIGKILLs TERM-immune grandchildren (no orphan survives)', () => {
+    // Regression pin for the 2026-08 escalation change: the watchdog used to
+    // follow its killpg(SIGTERM) + 5s grace with a DIRECT proc.kill() — a
+    // grandchild that ignores TERM survived and burned cores/API for hours
+    // (the observed 15-hour-orphan class). Now the grace escalates to
+    // killpg(SIGKILL). The child here traps TERM and spawns a TERM-immune
+    // grandchild; only a GROUP SIGKILL clears both. Markers are per-run
+    // unique (pid) so concurrent worktree suites can't cross-kill.
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'gd-'));
+    const log = path.join(dir, 'run.log');
+    const g1 = `6091.${process.pid}`;
+    const g2 = `6092.${process.pid}`;
+    const alive = (m: string) => spawnSync('pgrep', ['-f', `sleep ${m.replace('.', '\\.')}`], { stdio: 'pipe', timeout: 5_000 }).status === 0;
+    try {
+      spawnSync(DETACH, ['--log', log, '--timeout', '1', '--', 'bash', '-c',
+        `trap '' TERM; (trap '' TERM; sleep ${g1}) & exec sleep ${g2}`],
+        { encoding: 'utf-8', timeout: 10000 });
+      expect(waitFor(() => logHas(log, '### gstack-detach EXIT=timeout ###'), 15000)).toBe(true);
+      // Grace is 5s after the TERM that both processes ignore — the SIGKILL
+      // escalation must clear the whole group shortly after the sentinel.
+      expect(waitFor(() => !alive(g1) && !alive(g2), 10000),
+        'TERM-immune child/grandchild survived the watchdog — killpg(SIGKILL) escalation regressed').toBe(true);
+    } finally {
+      spawnSync('pkill', ['-9', '-f', `sleep 609[12]\\.${process.pid}`], { stdio: 'ignore', timeout: 5_000 });
+      fs.rmSync(dir, { recursive: true, force: true });
+    }
+  }, 30000);
+
+  test('watchdog kills the grandchild even when the LEADER dies on the SIGTERM', () => {
+    // The pgid-after-grace bug: killpg(getpgid(proc.pid), SIGKILL) raised
+    // ESRCH once the leader had honored the TERM, and the except fell back
+    // to proc.kill() on a corpse — the TERM-immune grandchild lived forever.
+    // The fix captures the pgid AT SPAWN. This variant is the one the
+    // TERM-immune-leader test above cannot see.
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'gd-'));
+    const log = path.join(dir, 'run.log');
+    const g = `6093.${process.pid}`;
+    const alive = () => spawnSync('pgrep', ['-f', `sleep ${g.replace('.', '\\.')}`], { stdio: 'pipe', timeout: 5_000 }).status === 0;
+    try {
+      // Leader: no trap — dies on the watchdog's SIGTERM. Grandchild:
+      // TERM-immune, same group — only a saved-pgid SIGKILL reaches it.
+      spawnSync(DETACH, ['--log', log, '--timeout', '1', '--', 'bash', '-c',
+        `(trap '' TERM; sleep ${g}) & sleep 60`],
+        { encoding: 'utf-8', timeout: 10000 });
+      expect(waitFor(() => logHas(log, '### gstack-detach EXIT=timeout ###'), 15000)).toBe(true);
+      expect(waitFor(() => !alive(), 10000),
+        'grandchild survived a dead leader — the pgid must be captured at spawn, not resolved after the grace').toBe(true);
+    } finally {
+      spawnSync('pkill', ['-9', '-f', `sleep 6093\\.${process.pid}`], { stdio: 'ignore', timeout: 5_000 });
+      fs.rmSync(dir, { recursive: true, force: true });
+    }
+  }, 30000);
+
   test('machine --lock serializes concurrent runs (second WAITS for the first)', () => {
     const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'gd-'));
     const lock = `gstack-detach-test-${process.pid}`;

test/helpers/touchfiles-data.ts

@@ -229,8 +229,6 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   'retro':             ['bin/gstack-retro-metrics', 'retro/**', 'test/skill-e2e-retro.test.ts'],
   'retro-base-branch': ['bin/gstack-retro-metrics', 'retro/**', 'test/skill-e2e-retro.test.ts'],
 
-  // Global discover
-
   // CSO
   'cso-full-audit':   ['cso/**', 'test/skill-e2e-cso.test.ts'],
   'cso-diff-mode':    ['cso/**', 'test/skill-e2e-cso.test.ts'],
@@ -297,8 +295,6 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     'test/skill-e2e-docsync-spawned.test.ts',
   ],
 
-  // Plan completion audit + verification
-
   // Design
   'design-consultation-core':       ['design-consultation/**', 'scripts/gen-skill-docs.ts', 'test/helpers/llm-judge.ts', 'test/skill-e2e-design.test.ts'],
   'design-consultation-existing':   ['design-consultation/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-design.test.ts'],
@@ -307,8 +303,6 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   'plan-design-review-no-ui-scope': ['plan-design-review/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-design.test.ts'],
   'design-review-fix':              ['design-review/**', 'browse/src/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-design.test.ts'],
 
-  // Design Shotgun
-
   // /diagram (diagram-render bundle consumers). Triplet = deterministic
   // functional (gate); authoring quality = LLM-judged benchmark (periodic).
   'diagram-triplet':            ['diagram/**', 'lib/diagram-render/**', 'browse/src/write-commands.ts', 'browse/src/read-commands.ts', 'test/skill-e2e-diagram.test.ts'],
@@ -696,8 +690,6 @@ export const E2E_TIERS: Record<string, 'gate' | 'periodic'> = {
   'retro': 'periodic',
   'retro-base-branch': 'gate',
 
-  // Global discover
-
   // CSO — gate for security guardrails, periodic for quality
   'cso-full-audit': 'periodic',  // D2a demotion 2026-08: 250s/$0.57 full audit; cso targeted tests stay gate
   'cso-diff-mode': 'gate',
@@ -827,8 +819,6 @@ export const LLM_JUDGE_TOUCHFILES: Record<string, string[]> = {
   'design-review/SKILL.md fix loop':      ['design-review/SKILL.md', 'design-review/SKILL.md.tmpl', 'test/skill-llm-eval.test.ts'],
   'design-consultation/SKILL.md research': ['design-consultation/SKILL.md', 'design-consultation/SKILL.md.tmpl', 'test/skill-llm-eval.test.ts'],
 
-  // Office Hours
-
   // Deploy skills
   'land-and-deploy/SKILL.md workflow':    ['land-and-deploy/SKILL.md', 'land-and-deploy/SKILL.md.tmpl', 'land-and-deploy/sections/**', 'test/skill-llm-eval.test.ts'],
   'canary/SKILL.md monitoring loop':      ['canary/SKILL.md', 'canary/SKILL.md.tmpl', 'test/skill-llm-eval.test.ts'],

test/question-preference-hook.test.ts

@@ -730,6 +730,7 @@ describe('Conductor spawned deny (#2733)', () => {
       const scriptKind = spawnSync(BIN, [], {
         env: { PATH: process.env.PATH ?? '/usr/bin:/bin', ...env },
         encoding: 'utf-8',
+        timeout: 30_000,
       }).stdout.trim();
       expect(
         spawnedByEnv(env),

test/session-runner-groupkill.test.ts

@@ -36,12 +36,16 @@ describe('session-runner timeout kills the whole process group', () => {
     const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'groupkill-'));
     const shimDir = path.join(dir, 'bin');
     fs.mkdirSync(shimDir);
-    // Unique-ish sleep durations double as pgrep markers: they appear only
+    // Markers are unique PER RUN (fractional seconds carry this process's
-    // in the shim's children's argv, never in this test process's cmdline.
+    // pid): sibling Conductor worktrees run free suites concurrently with no
+    // machine lock, and fixed markers let one run pgrep/pkill the OTHER
+    // run's shims (review finding — a cross-run flake inside the anti-flake
+    // tests). GNU sleep accepts decimals, argv stays greppable.
+    const mark = (n: number) => `${n}.${process.pid}`;
     const shim = [
       '#!/bin/bash',
-      'sleep 6041 &',        // the orphan-candidate grandchild
+      `sleep ${mark(6041)} &`,   // the orphan-candidate grandchild
-      'exec sleep 6042',     // the shim itself, wedged forever, no NDJSON
+      `exec sleep ${mark(6042)}`, // the shim itself, wedged forever, no NDJSON
     ].join('\n');
     fs.writeFileSync(path.join(shimDir, 'claude'), `${shim}\n`, { mode: 0o755 });
 
@@ -71,13 +75,13 @@ describe('session-runner timeout kills the whole process group', () => {
       // The kill is SIGKILL on the GROUP: give the OS a beat to reap, then
       // require both the wedged shim and its grandchild gone.
       await new Promise((r) => setTimeout(r, 1_000));
-      expect(aliveWithArg('sleep 6042'), 'the fake claude itself survived the timeout kill').toBe(false);
+      expect(aliveWithArg(`sleep ${mark(6042)}`), 'the fake claude itself survived the timeout kill').toBe(false);
-      expect(aliveWithArg('sleep 6041'), 'the grandchild ORPHANED — group kill regressed to a direct-child kill').toBe(false);
+      expect(aliveWithArg(`sleep ${mark(6041)}`), 'the grandchild ORPHANED — group kill regressed to a direct-child kill').toBe(false);
     } finally {
       process.env.PATH = realPath;
       // Belt and braces: never leak the markers into later tests even on
       // assertion failure.
-      spawnSync('pkill', ['-f', 'sleep 604[12]'], { stdio: 'ignore', timeout: 5_000 });
+      spawnSync('pkill', ['-f', `sleep 604[12]\\.${process.pid}`], { stdio: 'ignore', timeout: 5_000 });
       fs.rmSync(dir, { recursive: true, force: true });
     }
   }, 60_000);

test/session-runner-startup-grace.test.ts

@@ -42,7 +42,7 @@ describe('session-runner two-phase timeout', () => {
       '#!/bin/bash',
       'sleep 2',
       'echo \'{"type":"system","subtype":"init"}\'',
-      'exec sleep 6071',
+      `exec sleep 6071.${process.pid}`,
     ].join('\n') + '\n', { mode: 0o755 });
 
     const realPath = process.env.PATH;
@@ -65,7 +65,7 @@ describe('session-runner two-phase timeout', () => {
       expect(wall).toBeLessThan(20_000);
     } finally {
       process.env.PATH = realPath;
-      Bun.spawnSync(['pkill', '-f', 'sleep 6071'], { timeout: 5_000 });
+      Bun.spawnSync(['pkill', '-f', `sleep 6071\\.${process.pid}`], { timeout: 5_000 });
       fs.rmSync(dir, { recursive: true, force: true });
     }
   }, 60_000);
@@ -74,7 +74,7 @@ describe('session-runner two-phase timeout', () => {
     const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'grace-'));
     const shimDir = path.join(dir, 'bin');
     fs.mkdirSync(shimDir);
-    fs.writeFileSync(path.join(shimDir, 'claude'), '#!/bin/bash\nexec sleep 6072\n', { mode: 0o755 });
+    fs.writeFileSync(path.join(shimDir, 'claude'), `#!/bin/bash\nexec sleep 6072.${process.pid}\n`, { mode: 0o755 });
 
     const realPath = process.env.PATH;
     process.env.PATH = `${shimDir}:${realPath}`;
@@ -95,7 +95,7 @@ describe('session-runner two-phase timeout', () => {
       expect(result.costEstimate.turnsUsed).toBe(0);
     } finally {
       process.env.PATH = realPath;
-      Bun.spawnSync(['pkill', '-f', 'sleep 6072'], { timeout: 5_000 });
+      Bun.spawnSync(['pkill', '-f', `sleep 6072\\.${process.pid}`], { timeout: 5_000 });
       fs.rmSync(dir, { recursive: true, force: true });
     }
   }, 60_000);

test/spawnsync-timeout-tripwire.test.ts

@@ -37,7 +37,7 @@ const SCAN_ROOTS = [
   'browser-skills',
 ];
 
-const SYNC_SPAWN = /\b(?:spawnSync|execSync|Bun\.spawnSync)\s*\(/;
+const SYNC_SPAWN = /\b(?:spawnSync|execSync|execFileSync|Bun\.spawnSync)\s*\(/;
 /** Generous on purpose: multi-line arg arrays push the options object far
  *  below the call line (observed: +13 lines in codex-model-probe). A wide
  *  window trades a sliver of false-negative risk for zero rename churn. */
@@ -48,7 +48,7 @@ const EXEMPT_MARKER = /tripwire-exempt:/;
 const COMMENT_LINE = /^\s*(?:\/\/|\*|\/\*)/;
 
 /** Shrink-only: lower it when exemptions burn down; never raise it. */
-const EXEMPT_CEILING = 6;
+const EXEMPT_CEILING = 8;
 
 const SELF = path.join('test', 'spawnsync-timeout-tripwire.test.ts');