diff --git a/browse/src/terminal-agent.ts b/browse/src/terminal-agent.ts
index c5417bd38..88e89a719 100644
--- a/browse/src/terminal-agent.ts
+++ b/browse/src/terminal-agent.ts
@@ -301,11 +301,12 @@ function buildTabAwarenessHint(stateDir: string): string {
     'You are running inside the gstack browser sidebar with live access to the user\'s browser tabs.',
     '',
     'Tab state files (kept fresh automatically by the extension):',
-    `  ${tabsFile}        — all open tabs (id, url, title, active, pinned)`,
-    `  ${activeFile}    — the currently active tab`,
-    'Read these any time the user asks about "tabs", "the current page", or anything multi-tab. Do NOT shell out to $B tabs just to learn what\'s open — read the file.',
+    `  ${tabsFile}        — all open tabs: tabs[] of {tabId, url, title, active, pinned, windowId}`,
+    `  ${activeFile}    — the currently active tab: {tabId, url, title}`,
+    'Read these any time the user asks about "tabs", "the current page", or anything multi-tab. They\'re kept current, so read them instead of running $B tabs when you only need to know what\'s open.',
     '',
     'Tab manipulation commands (via $B):',
+    '  ($B is the gstack browse CLI. This shell does not set it, so resolve it the way the /browse skill does: the project\'s .claude/skills/gstack/browse/dist/browse if present, else ~/.claude/skills/gstack/browse/dist/browse.)',
     '  $B tab <id>                 — switch to a tab',
     '  $B newtab [url]             — open a new tab',
     '  $B closetab [id]            — close a tab (current if no id)',
diff --git a/scripts/typecheck-test-baseline.json b/scripts/typecheck-test-baseline.json
index a1477b4ef..ee19e3b9f 100644
--- a/scripts/typecheck-test-baseline.json
+++ b/scripts/typecheck-test-baseline.json
@@ -485,7 +485,6 @@
     "test/skill-e2e-shared-libs-paths.test.ts\tTS2769\tNo overload matches this call. The last overload gave the following error. Argument of type 'unknown' is not assignable to parameter of type 'string'.": 1,
     "test/skill-e2e-ship-docsync.test.ts\tTS2345\tArgument of type 'EvalCollector | null' is not assignable to parameter of type 'EvalCollector'. Type 'null' is not assignable to type 'EvalCollector'.": 9,
     "test/skill-e2e-third-party-actions.test.ts\tTS2345\tArgument of type '{ maxTurns: 8; allowedTools: readonly ['Read', 'Bash']; timeout: 240000; runId: string; env: { PATH: string; }; prompt: string; workingDirectory: string; testName: string; }' is not assignable to parameter of type '{ prompt: string; workingDirectory: string; maxTurns?: number | undefined; appendSystemPrompt?: string | undefined; completionReserveMs?: number | undefined; allowedTools?: string[] | undefined; ... 9 more ...; nativeLifecycle?: { ...; } | undefined; }'. Types of property 'allowedTools' are incompatible. The type 'readonly [\"Read\", \"Bash\"]' is 'readonly' and cannot be assigned to the mutable type 'string[]'.": 5,
-    "test/skill-e2e-triage.test.ts\tTS2353\tObject literal may only specify known properties, and 'has_in_branch_classification' does not exist in type 'Partial<EvalTestEntry>'.": 1,
     "test/skill-routing-e2e.test.ts\tTS2345\tArgument of type '\"e2e-routing\"' is not assignable to parameter of type '\"e2e\" | \"llm-judge\"'.": 1,
     "test/strict-output.test.ts\tTS2739\tType '{ failedTests: number; unhandledBetweenTests: number; terminalFileCounts: number[]; }' is missing the following properties from type 'BunTestOutputSummary': terminalTestCounts, skippedTests, passedTests": 5,
     "test/test-free-shards.test.ts\tTS2345\tArgument of type 'number | ReadableStream<Uint8Array<ArrayBuffer>> | undefined' is not assignable to parameter of type 'BodyInit | null | undefined'. Type 'number' is not assignable to type 'BodyInit | null | undefined'.": 2,
diff --git a/test/auq-matrix-format-gate.test.ts b/test/auq-matrix-format-gate.test.ts
new file mode 100644
index 000000000..48b9f95e3
--- /dev/null
+++ b/test/auq-matrix-format-gate.test.ts
@@ -0,0 +1,19 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { auqMachineFormatProblems, scoreAuqFormat } from './helpers/auq-sdk-capture';
+
+const briefs = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/auq-matrix/briefs.json'), 'utf8')) as
+  { known_good: Record<string, string>; known_bad: Record<string, string> };
+
+describe('AUQ matrix format gate', () => {
+  test.each(Object.entries(briefs.known_good))('passes %s', (_name, text) => {
+    expect(auqMachineFormatProblems(text)).toEqual([]);
+  });
+  test.each(Object.entries(briefs.known_bad))('fails %s', (_name, text) => {
+    expect(auqMachineFormatProblems(text).length).toBeGreaterThan(0);
+  });
+  test('a brief missing only presentation elements no longer fails, but they are still reported', () => {
+    expect(scoreAuqFormat(briefs.known_good['no-emoji-or-net']!).missing).toEqual(['Pros / cons:', '✅', '❌', 'Net:']);
+  });
+});
diff --git a/test/browse-judge-floors.test.ts b/test/browse-judge-floors.test.ts
new file mode 100644
index 000000000..53a1a160f
--- /dev/null
+++ b/test/browse-judge-floors.test.ts
@@ -0,0 +1,13 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { browseJudgeFloorsMet } from './helpers/workflow-judge-cache';
+
+type Scores = { clarity: number; completeness: number; actionability: number };
+const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/browse-judge/panel-means.json'), 'utf8')) as
+  { known_good: Record<string, Scores>; known_bad: Record<string, Scores> };
+
+describe('browse judge floors', () => {
+  test.each(Object.entries(stored.known_good))('passes %s', (_name, scores) => expect(browseJudgeFloorsMet(scores)).toBe(true));
+  test.each(Object.entries(stored.known_bad))('fails %s', (_name, scores) => expect(browseJudgeFloorsMet(scores)).toBe(false));
+});
diff --git a/test/ceo-mode-option.test.ts b/test/ceo-mode-option.test.ts
index 0fe3becfa..85b0b3f0f 100644
--- a/test/ceo-mode-option.test.ts
+++ b/test/ceo-mode-option.test.ts
@@ -827,6 +827,22 @@ const mutations:Record<string,(x:any)=>void>={
  'wrong actual choice':x=>{const c=decision(x);c.answers[c.questions[0].question]=c.questions[0].options[1].label},
 };
 for(const [name,mutate] of Object.entries(mutations))test(name,()=>{const x=clone();mutate(x);expect(check(x)).toBe(false)});
+// Wording is free: reworded labels, a different question title, no "PLAN.md lists" phrase and
+// rationale outside the old keyword list still prove HOLD when the outcome is the same.
+const relabel=(x:any,keep:string,defer:string)=>{const c=decision(x);const q=c.questions[0];q.options[0].label=keep;q.options[1].label=defer;c.answers[q.question]=keep;x.tools.find((t:any)=>t.kind==='use'&&t.toolUseId===c.toolUseId).input.questions=structuredClone(c.questions)};
+const rewordings:Record<string,(x:any)=>void>={
+ 'reworded option labels':x=>relabel(x,'A) Retain the update endpoint (recommended)','B) Move update to a later TODO'),
+ 'reworded question title':x=>editQuestion(x,q=>q.question=q.question.replace(/^D\d+ — [^\n]*/,'D4: Should the update endpoint stay in this review?')),
+ 'no plan-lists phrase':x=>editQuestion(x,q=>q.question=q.question.replace('PLAN.md lists','The plan has').replace('the plan already states','the plan names')),
+ 'rationale outside the old keyword list':x=>editQuestion(x,q=>q.question=q.question.replace(/ELI10:[\s\S]*?Stakes if/,'ELI10: Renaming a saved view in place is part of the written approach, and dropping it would make every correction a two-step chore that members notice right away.\nStakes if').replace(/Stakes if[\s\S]*?Recommendation:/,'Stakes if we pick wrong: deferring leaves members unable to correct a saved view; keeping it costs one more endpoint in this review.\nRecommendation:')),
+};
+for(const [name,change] of Object.entries(rewordings))test(`still HOLD with ${name}`,()=>{const x=clone();change(x);expect(check(x)).toBe(true)});
+const newBad:Record<string,(x:any)=>void>={
+ 'both options keep':x=>relabel(x,'Keep update in scope (recommended)','Keep update and add sharing'),
+ 'kept item outside the plan':x=>relabel(x,'Keep sharing links (recommended)','Defer sharing links to TODOS.md'),
+ 'answer selects defer after relabel':x=>{relabel(x,'Keep update in scope (recommended)','Defer update to TODOS.md');const c=decision(x);c.answers[c.questions[0].question]=c.questions[0].options[1].label},
+};
+for(const [name,mutate] of Object.entries(newBad))test(`not HOLD with ${name}`,()=>{const x=clone();mutate(x);expect(check(x)).toBe(false)});
 test('later quoted withdrawal is not current withdrawal',()=>{const x=clone();x.transcript.assistantMessages.push({sessionId:decision(x).sessionId,timestamp:new Date().toISOString(),text:'Example: "I withdraw this decision."'});expect(check(x)).toBe(true)});
 {
   const briefs = require('./fixtures/ceo-hold-note-briefs-36597762183.json');
diff --git a/test/ceo-split-collection.test.ts b/test/ceo-split-collection.test.ts
index 2dd9a39c6..3bd93d3ec 100644
--- a/test/ceo-split-collection.test.ts
+++ b/test/ceo-split-collection.test.ts
@@ -181,7 +181,7 @@ mock.module(${JSON.stringify(path.join(ROOT, 'test/helpers/e2e-gate.ts'))}, () =
 mock.module(${JSON.stringify(path.join(ROOT, 'test/helpers/plan-review-decisions.ts'))}, () => ({
   evaluatePlanReviewDecisions: async input => {
     facts.evaluators++; save();
-    expect(input.kind).toBe('scope'); expect(input.floor).toBe(4);
+    expect(input.kind).toBe('scope'); expect(input.floor).toBe(2);
     expect(input.ceiling).toBeUndefined(); expect(input.targets).toEqual(CEO_SCOPE_CANDIDATES);
     expect(input.deadlineAt - Date.now()).toBeGreaterThan(1_490_000);
     expect(input.deadlineAt - Date.now()).toBeLessThanOrEqual(1_500_000);
diff --git a/test/ceo-split-floor.test.ts b/test/ceo-split-floor.test.ts
new file mode 100644
index 000000000..135236572
--- /dev/null
+++ b/test/ceo-split-floor.test.ts
@@ -0,0 +1,49 @@
+import { expect, test } from 'bun:test';
+import type { AskUserQuestionFingerprint } from './helpers/claude-pty-runner';
+import type { NativeQuestion } from './helpers/plan-skill-questions';
+import { CEO_SCOPE_CANDIDATES } from './helpers/plan-review-cases';
+import { CEO_SPLIT_CALL_FLOOR } from './helpers/ceo-split-question-policy';
+import { validatePlanReviewDecisionResponse, type PlanReviewDecision, type PlanReviewDecisionInput } from './helpers/plan-review-decisions';
+
+const PLATFORMS = ['Slack', 'Discord', 'Microsoft Teams', 'Telegram', 'Mattermost'];
+const question = (platform: string, n: number): NativeQuestion => ({
+  header: platform, multiSelect: false,
+  question: `D3.${n} — ${platform} integration\nELI10: Decide this integration independently. Recommendation: Include.`,
+  options: ['Include', 'Defer', 'Cut', 'Hold'].map(label => ({ label, description: `Choose ${label} for ${platform}.` })),
+});
+type NativeCall = AskUserQuestionFingerprint & { toolUseId: string; questions: NativeQuestion[] };
+const call = (id: string, questions: NativeQuestion[]) => ({
+  toolUseId: id, questions, selectedOptions: questions.map(() => 1), signature: id,
+  promptSnippet: 'diagnostic', options: [], observedAtMs: 1, preReview: true,
+}) as unknown as NativeCall;
+const row = (fp: NativeCall, tab: number, target: string): PlanReviewDecision => ({
+  toolUseId: fp.toolUseId, questionIndex: tab, kind: 'scope', targetIds: [target], independentDecisions: 1,
+  evidence: [{ field: 'question', optionIndex: null, quote: fp.questions[tab - 1]!.question.split('\n')[0]! }],
+  reason: 'This acknowledged question presents the independent decision described by this target.',
+  optionActions: (['include', 'defer', 'cut', 'hold'] as const).map((action, i) => ({ optionIndex: i + 1, action })),
+});
+/** Stored shapes: each inner array is one AskUserQuestion call; numbers index PLATFORMS. */
+function run(groups: number[][]) {
+  const fingerprints = groups.map((group, i) => call(`native-${i}`, group.map(p => question(PLATFORMS[p]!, p + 1))));
+  const input: PlanReviewDecisionInput = { plan: 'Review the five independent integrations in this plan.', kind: 'scope',
+    targets: CEO_SCOPE_CANDIDATES, fingerprints, floor: CEO_SPLIT_CALL_FLOOR, deadlineAt: Date.now() + 60_000 };
+  const questions = groups.flatMap((group, i) => group.map((p, tab) => row(fingerprints[i]!, tab + 1, `E${p + 1}`)));
+  return () => validatePlanReviewDecisionResponse(input, { questions });
+}
+
+test('the split floor is the minimum number of calls for five options at four per call', () => {
+  expect(CEO_SPLIT_CALL_FLOOR).toBe(Math.ceil(PLATFORMS.length / 4));
+});
+test.each([
+  ['five sequential calls', [[0], [1], [2], [3], [4]]],
+  ['two batched calls', [[0, 1, 2, 3], [4]]],
+  ['three batched calls', [[0, 1], [2, 3], [4]]],
+])('known-good split passes: %s', (_name, groups) => {
+  expect(run(groups)().coveredTargetIds).toEqual(CEO_SCOPE_CANDIDATES.map(c => c.id));
+});
+test.each([
+  ['a candidate silently dropped', [[0, 1, 2, 3]], 'missing target decisions'],
+  ['one call for a dropped fifth candidate', [[0, 1, 2], [3]], 'missing target decisions'],
+])('known-bad split fails: %s', (_name, groups, message) => {
+  expect(run(groups)).toThrow(message);
+});
diff --git a/test/ceo-split-question-policy.test.ts b/test/ceo-split-question-policy.test.ts
index 184754c79..4a63f19b5 100644
--- a/test/ceo-split-question-policy.test.ts
+++ b/test/ceo-split-question-policy.test.ts
@@ -285,7 +285,7 @@ const save = () => fs.writeFileSync(${JSON.stringify(facts)}, JSON.stringify(fac
 mock.module(${JSON.stringify(path.join(ROOT, 'test/helpers/plan-review-decisions.ts'))}, () => ({
   evaluatePlanReviewDecisions: async input => {
     facts.judges++; save();
-    expect(input.kind).toBe('scope'); expect(input.floor).toBe(4);
+    expect(input.kind).toBe('scope'); expect(input.floor).toBe(2);
     expect(input.targets).toEqual(CEO_SCOPE_CANDIDATES);
     expect(input.deadlineAt).toBeGreaterThan(Date.now());
     expect(input.deadlineAt - Date.now()).toBeLessThanOrEqual(1_500_000);
@@ -363,7 +363,7 @@ await import(${JSON.stringify(path.join(ROOT, 'test/skill-e2e-plan-ceo-split-ove
     expect(exit, out + err).toBe(scenario.passes ? 0 : 1);
     expect(observed.judges).toBe(scenario.outcome === 'timeout' ? 0 : 1);
     if (scenario.outcome === 'timeout') expect(out + err).toContain('split-overflow test FAILED: outcome=timeout');
-    if (scenario.count < 4) expect(out + err).toContain('target call count 3 below floor 4');
+    if (scenario.count < 4) expect(out + err).toContain('missing target decisions');
   } finally { fs.rmSync(temp, { recursive: true, force: true }); }
 }, 20_000);
 
diff --git a/test/coverage-audit-evidence.test.ts b/test/coverage-audit-evidence.test.ts
index 142a29bfc..2974d5e8e 100644
--- a/test/coverage-audit-evidence.test.ts
+++ b/test/coverage-audit-evidence.test.ts
@@ -467,6 +467,25 @@ Guard clauses tested: 0 / 4
   });
 });
 
+describe('coverage audit closing summary', () => {
+  const stored = require('./fixtures/coverage-audit-summary.json') as { known_good: Record<string, string>; known_bad: Record<string, string> };
+  test.each(Object.entries(stored.known_good))('summary passes %s', (_name, output) => {
+    const s = synthetic(); s.result.output = output;
+    expect(verdict(s)).toEqual({ sourceRead: true, testsRead: true, diagram: true, passed: true, failures: [] });
+  });
+  test.each(Object.entries(stored.known_bad))('summary fails %s', (_name, output) => {
+    const s = synthetic(); s.result.output = output;
+    expect(verdict(s).passed).toBe(false);
+  });
+  test('a correct summary cannot replace native file delivery or a completed capture', () => {
+    const output = stored.known_good['unfamiliar-diagram-with-summary']!;
+    const unread = synthetic(); unread.result.output = output; block(unread, 2).content = 'not the file';
+    expect(verdict(unread)).toMatchObject({ sourceRead: false, passed: false });
+    const timedOut = synthetic(); timedOut.result.output = output; timedOut.result.exitReason = 'timeout';
+    expect(verdict(timedOut).passed).toBe(false);
+  });
+});
+
 describe('coverage-audit-af', () => {
 const fixture = fixture_coverage_audit_af;
 const actual = (index: number) => structuredClone(fixture.rows[index]!);
diff --git a/test/disabled-plan-review-evidence.test.ts b/test/disabled-plan-review-evidence.test.ts
index c9f4edf56..66448eef3 100644
--- a/test/disabled-plan-review-evidence.test.ts
+++ b/test/disabled-plan-review-evidence.test.ts
@@ -257,6 +257,26 @@ describe('disabled outside-plan live oracle', () => {
 
 // Both actual attempts obeyed the off switch; prior records and rejected claims
 // were mistaken for current completion by the bare substring check.
+describe('closing outside-review self-report', () => {
+  const stored = require('./fixtures/disabled-plan-self-report.json') as { known_good: Record<string, string>; known_bad: Record<string, string> };
+  const withOutput = (text: string) => { const result = completed(); result.output += '\n' + text; return result; };
+  test.each(Object.entries(stored.known_good))('self-report passes %s', (_name, text) => {
+    expect(oracle(withOutput(text))).toMatchObject({ passed: true, falseCompletion: false, disabledAttribution: true });
+  });
+  test.each(Object.entries(stored.known_bad))('self-report fails %s', (_name, text) => {
+    expect(oracle(withOutput(text))).toMatchObject({ passed: false, falseCompletion: true });
+  });
+  test('a disabled self-report cannot override dispatch, CLI execution, preflight or persistence', () => {
+    const result = withOutput(stored.known_good['plain-history-wording']!);
+    expect(oracle(result, 'codex invoked\n').passed).toBe(false);
+    const dispatched = structuredClone(result); dispatched.transcript.splice(-1, 0, dispatch('Agent', { prompt: 'Outside review' }));
+    expect(oracle(dispatched).passed).toBe(false);
+    const noPreflight = structuredClone(result); noPreflight.transcript.splice(1, 2);
+    expect(oracle(noPreflight).passed).toBe(false);
+    expect(disabledPlanReviewEvidence(result, '', JSON.stringify(PRIOR_RECORD), PRIOR_RECORD).passed).toBe(false);
+  });
+});
+
 describe('AD v2 disabled-plan public attribution', () => {
   const captured = require('./fixtures/disabled-plan-attribution-ad-v2.json');
   test.each(captured.cases)('accepts actual attempt $attempt output without crediting historical coverage', (item: any) => {
diff --git a/test/docsync-command-grammar.test.ts b/test/docsync-command-grammar.test.ts
index d9f9baf0c..75bd23f5c 100644
--- a/test/docsync-command-grammar.test.ts
+++ b/test/docsync-command-grammar.test.ts
@@ -64,13 +64,13 @@ test.each([
   'git rev-parse HEAD^{tree}{,x}', 'git rev-parse HEAD^{tree', 'git rev-parse HEAD^{tree}}',
   'git rev-parse {HEAD}', 'git rev-parse HEAD$(pwd)', 'git rev-parse "HEAD$(pwd)"',
   'git rev-parse `pwd`', 'git rev-parse ${HEAD}', 'git rev-parse HEAD; git status',
-  'git rev-parse HEAD && git status', 'git rev-parse HEAD || true', 'git rev-parse HEAD | cat',
-  'git rev-parse HEAD > out.md', 'git rev-parse HEAD 2>/dev/null', 'git rev-parse HEAD < in.md',
+  'git rev-parse HEAD && git status', 'git rev-parse HEAD | cat',
+  'git rev-parse HEAD > out.md', 'git rev-parse HEAD 2>/tmp/leak', 'git rev-parse HEAD < in.md',
   'git rev-parse HEAD\ngit status', 'git rev-parse HEAD &', 'git rev-parse HEAD\\^{tree}',
   'git rev-parse "HEAD^{tree}', "git rev-parse 'HEAD^{tree}", 'git rev-parse HEAD*',
   'git status "unfinished', "git status 'unfinished", "git hash-object '-w'app.ts", 'git status\u0000',
   'git rev-parse HEAD?', 'git rev-parse HEAD[12]', 'git rev-parse ~', 'git rev-parse HEAD # comment',
-  'git -C /owned/repo rev-parse HEAD', 'git -c core.pager=cat show HEAD',
+  'git -C /owned/repo rev-parse HEAD', "git -c core.pager='sh -c id' show HEAD",
   'git --git-dir /owned/repo/.git status', 'git --work-tree /owned/repo status',
   'git remote get-url origin', 'git remote add origin /outside', 'git config --global user.name attacker',
   'git hash-object -w app.ts', 'git hash-object "-w" app.ts', 'git hash-object -wt blob app.ts',
@@ -83,6 +83,12 @@ test.each([
     .toEqual(['command outside declared docs observation interface']);
 });
 
+test('harmless read wrappers do not fail a docs run', () => {
+  for (const command of ['git rev-parse HEAD || true', 'git rev-parse HEAD 2>/dev/null', 'git -c core.pager=cat show HEAD']) {
+    expect(docsToolFailures(nativeResult(command, 'successful tool acknowledgment', 'docs-dispatch'), fixture, [], true)).toEqual([]);
+  }
+});
+
 test('quoted revision search and reflog arguments remain literal single arguments', () => {
   for (const command of ['git log "HEAD@{2 days ago}"', "git show 'HEAD^{/fix, or repair..}'",
     'git hash-object app.ts', 'git branch --show-current']) {
diff --git a/test/docsync-lifecycle-interface.test.ts b/test/docsync-lifecycle-interface.test.ts
index 459945c31..08cc41278 100644
--- a/test/docsync-lifecycle-interface.test.ts
+++ b/test/docsync-lifecycle-interface.test.ts
@@ -1,4 +1,4 @@
-import { afterAll, beforeAll, expect, test } from 'bun:test';
+import { afterAll, beforeAll, describe, expect, test } from 'bun:test';
 import * as fs from 'node:fs';
 import * as path from 'node:path';
 import { fixtureDocs } from './helpers/docsync-fixture';
@@ -17,9 +17,6 @@ test('fixture lifecycle guidance exposes owned skill paths and literal commands'
   const guidance = docsNativeInterface(fixture);
   const skills = fixture.skills.split(path.sep).join('/');
   expect(guidance).toContain(`${skills}/document-release/SKILL.md`);
-  expect(guidance).toContain('instead of copying the generated shell wrappers');
-  expect(guidance).toContain("satisfy the skill's start/end lifecycle requirements here");
-  expect(guidance).toContain('applies to parent and every child; include this interface in child prompts');
   const [start, end] = lifecycleCommands();
   expect(lifecycleCommands()).toHaveLength(2);
   expect(start).toBe(`GSTACK_SESSION_KIND=spawned ${skills}/bin/gstack-skill-start --skill document-release --model claude`);
@@ -33,11 +30,6 @@ test('fixture lifecycle guidance exposes owned skill paths and literal commands'
 });
 
 test.each(['success', 'error', 'abort', 'unknown'])('same-session literal end accepts outcome %s without wrappers', outcome => {
-  const guidance = docsNativeInterface(fixture);
-  expect(guidance).toContain('actual literal values echoed by that same start call');
-  expect(guidance).toContain('Never execute the placeholders or reuse values from another session');
-  expect(guidance).toContain('If start fails or SESSION_KIND is not spawned, report the blocker');
-  expect(guidance).toContain('do not suppress an error or claim completion if it failed');
   const end = lifecycleCommands()[1].replace('OUTCOME', outcome).replace('SESSION_ID_VALUE', '123-1790299423-control').replace('TEL_START_VALUE', '1790299423');
   expect(docsCommandAllowed(end, fixture)).toBe(true);
   const result = { toolCalls: [{ tool: 'Bash', input: { command: end }, output: '' }] } as SkillTestResult;
@@ -49,7 +41,6 @@ test('historical misplaced spawned prefix remains rejected while corrected pream
   const misplaced = `GSTACK_SESSION_KIND=spawned ${original}`;
   expect(docsCommandAllowed(misplaced, fixture)).toBe(false);
   expect(docsCommandAllowed(spawned, fixture)).toBe(true);
-  expect(docsNativeInterface(fixture)).toContain('prefix belongs directly on the helper invocation, not on a preceding assignment');
 });
 
 test.each([
@@ -95,27 +86,40 @@ test.each([
 });
 
 test('lifecycle instructions neither widen command authority nor grant mutation approvals', () => {
-  const guidance = docsNativeInterface(fixture);
-  expect(guidance).toContain('The only additional scripts are none');
-  expect(guidance).toContain('do not grant any additional scripts, write paths or risk approvals');
-  expect(guidance).toContain('do not rewrite installed skills, config, actor state or scripts');
+  expect(docsNativeInterface(fixture)).toContain('The only additional scripts are none');
+  expect(docsNativeInterface(fixture, ['/owned/publish.ts'])).toContain('The only additional scripts are /owned/publish.ts');
   for (const suffix of [' 2>/dev/null', ' || true', ' && git status', '\ntrue', ' $EXTRA']) {
     expect(docsCommandAllowed(lifecycleCommands()[0] + suffix, fixture)).toBe(false);
   }
   expect(docsCommandAllowed('bun arbitrary.ts', fixture)).toBe(false);
 });
 
-test('Git guidance uses the existing working directory without authorizing global options', () => {
+test('Git guidance uses the existing working directory; --git-dir and --work-tree stay unauthorized', () => {
   const guidance = docsNativeInterface(fixture);
-  expect(guidance).toContain(`working directory for parent and child Bash calls is already ${fixture.repo}`);
-  expect(guidance).toContain('literal git subcommand must immediately follow git');
-  expect(guidance).toContain('does not make git -C an allowed command');
+  expect(guidance).toContain(fixture.repo);
   for (const command of ['git status', 'git diff --cached', 'git merge-base main HEAD', 'git rev-parse HEAD']) {
     expect(guidance).toContain(command);
     expect(docsCommandAllowed(command, fixture)).toBe(true);
   }
-  for (const command of [`git -C ${fixture.repo} status`, `git --git-dir ${fixture.repo}/.git status`,
-    `git --work-tree ${fixture.repo} status`, 'git -c core.pager=cat status', `cd ${fixture.repo} && git status`]) {
+  for (const command of [`git --git-dir ${fixture.repo}/.git status`, `git --work-tree ${fixture.repo} status`]) {
     expect(docsCommandAllowed(command, fixture)).toBe(false);
   }
 });
+
+describe('stored docsync observer commands', () => {
+  const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/docsync-observer-commands.json'), 'utf8')) as
+    { known_good: string[]; known_bad: string[] };
+  const fill = (command: string) => command.replaceAll('{repo}', fixture.repo).replaceAll('{other}', path.dirname(fixture.repo));
+  test.each(stored.known_good)('harmless read form is accepted: %s', command => {
+    expect(docsCommandAllowed(fill(command), fixture)).toBe(true);
+  });
+  test.each(stored.known_bad)('known-bad form stays rejected: %s', command => {
+    expect(docsCommandAllowed(fill(command), fixture)).toBe(false);
+  });
+  test('harmless wrappers never extend to lifecycle commands', () => {
+    for (const command of lifecycleCommands()) {
+      for (const suffix of [' 2>/dev/null', ' || true']) expect(docsCommandAllowed(command + suffix, fixture)).toBe(false);
+      expect(docsCommandAllowed(`cd ${fixture.repo} && ${command}`, fixture)).toBe(false);
+    }
+  });
+});
diff --git a/test/docsync-nested-writes.test.ts b/test/docsync-nested-writes.test.ts
index e68ac2b69..6ef829cc4 100644
--- a/test/docsync-nested-writes.test.ts
+++ b/test/docsync-nested-writes.test.ts
@@ -169,7 +169,7 @@ const copy = (capture = edit): Capture => ({ fixture: { ...capture.fixture, befo
   test('read-only and undeclared Git commands still block omitted-metadata attribution', () => {
     expect(docsWriteFailures(edit.observation, [], edit).length).toBeGreaterThan(0);
     const captured = copy();
-    captured.result.toolCalls.push({ tool: 'Bash', input: { command: `git -C ${captured.fixture.repo} status` }, output: '' });
+    captured.result.toolCalls.push({ tool: 'Bash', input: { command: `git --git-dir ${captured.fixture.repo}/.git status` }, output: '' });
     expect(verdict(captured).length).toBeGreaterThan(0);
   });
 });
diff --git a/test/eval-kinds.test.ts b/test/eval-kinds.test.ts
index ea6480f6c..d385ac044 100644
--- a/test/eval-kinds.test.ts
+++ b/test/eval-kinds.test.ts
@@ -58,9 +58,7 @@ describe('E2E_KINDS registry', () => {
     }
   });
 
-  test('the classification is the reviewed one: rule by default, 22 behavior, 25 judge', () => {
-    const counts = Object.values(E2E_KINDS).reduce<Record<string, number>>((acc, kind) => ({ ...acc, [kind]: (acc[kind] ?? 0) + 1 }), {});
-    expect(counts).toEqual({ rule: liveIds.length - 22 - 25, behavior: 22, judge: 25 });
+  test('contract-shaped cases stay rule-kind', () => {
     // Contract-shaped cases stay rule: ask-before-decide, plan-mode no-writes,
     // mandated steps, secrets, and the batching floor never ride a majority.
     for (const id of ['plan-ceo-mode-routing', 'plan-eng-multi-finding-batching', 'plan-design-review-plan-mode',
diff --git a/test/fixtures/auq-matrix/briefs.json b/test/fixtures/auq-matrix/briefs.json
new file mode 100644
index 000000000..00437fcfc
--- /dev/null
+++ b/test/fixtures/auq-matrix/briefs.json
@@ -0,0 +1,14 @@
+{
+  "_comment": "First-AUQ texts in serializeNativeAuq shape. known_good passes the matrix's format gate (substance is graded separately). known_bad must keep failing it.",
+  "known_good": {
+    "full-format": "Scope\n\nD1 — Which test harness should the plan adopt?\nELI10: The plan needs one test runner.\nStakes if we pick wrong: flaky CI for weeks.\nRecommendation: A because the repo already uses bun test.\nPros / cons:\nNet: reuse vs. migration.\n\nA) bun test (recommended)\n✅ already installed\n❌ fewer plugins\n\nB) vitest\n✅ plugins\n❌ new dependency",
+    "no-emoji-or-net": "Scope\n\nD1 — Which test harness should the plan adopt?\nELI10: The plan needs one test runner.\nRecommendation: A because the repo already uses bun test and adding vitest means a second config.\n\nA) bun test (recommended)\nAlready installed; fewer plugins.\n\nB) vitest\nMore plugins; new dependency.",
+    "bold-recommendation": "Scope\n\n**Recommendation:** B, because the migration is already half done.\n\nA) Stay\nNo churn.\n\nB) Finish migration (recommended)\nOne config."
+  },
+  "known_bad": {
+    "no-recommendation-line": "Scope\n\nD1 — Which test harness?\nELI10: pick one.\n✅ ❌ Net: tradeoff. Pros / cons\n\nA) bun test (recommended)\n\nB) vitest",
+    "two-recommended": "Scope\n\nRecommendation: A because it is installed.\n\nA) bun test (recommended)\n\nB) vitest (recommended)",
+    "no-recommended-option": "Scope\n\nRecommendation: A because it is installed.\n\nA) bun test\n\nB) vitest",
+    "empty-recommendation": "Scope\n\nRecommendation:\n\nA) bun test (recommended)\n\nB) vitest"
+  }
+}
diff --git a/test/fixtures/browse-judge/panel-means.json b/test/fixtures/browse-judge/panel-means.json
new file mode 100644
index 000000000..ad4864e77
--- /dev/null
+++ b/test/fixtures/browse-judge/panel-means.json
@@ -0,0 +1,13 @@
+{
+  "_comment": "Browse reference judge panel means against baseline {clarity:4, completeness:4, actionability:4}. known_good passes the floors (the old ratchet failed the first two); known_bad fails them.",
+  "known_good": {
+    "clarity-dip-to-3.67": { "clarity": 3.67, "completeness": 4, "actionability": 4 },
+    "clarity-at-floor": { "clarity": 3, "completeness": 4.33, "actionability": 4 },
+    "at-baseline": { "clarity": 4, "completeness": 4, "actionability": 4 }
+  },
+  "known_bad": {
+    "clarity-below-floor": { "clarity": 2.67, "completeness": 4, "actionability": 4 },
+    "completeness-below-floor": { "clarity": 4, "completeness": 3.67, "actionability": 4 },
+    "actionability-below-floor": { "clarity": 4, "completeness": 4, "actionability": 3.33 }
+  }
+}
diff --git a/test/fixtures/coverage-audit-summary.json b/test/fixtures/coverage-audit-summary.json
new file mode 100644
index 000000000..41a233c91
--- /dev/null
+++ b/test/fixtures/coverage-audit-summary.json
@@ -0,0 +1,17 @@
+{
+  "_comment": "Final coverage-audit outputs graded on the closing JSON summary. known_good passes even though its diagram is outside the legacy legend grammar; known_bad must fail.",
+  "known_good": {
+    "unfamiliar-diagram-with-summary": "COVERAGE\nprocessPayment  ★★ happy path covered (billing.test.ts:3)\nrefundPayment   — no test at all\n\nSummary: 1 of 2 functions tested.\n{\"tested\": [\"processPayment\"], \"untested\": [\"refundPayment\"]}",
+    "fenced-json-summary": "src/billing.ts coverage\n  processPayment -> covered by 'processes valid payment'\n  refundPayment -> GAP\n\n```json\n{\"tested\": [\"processPayment (happy path)\"], \"untested\": [\"refundPayment\"]}\n```",
+    "corrected-summary": "{\"tested\": [\"refundPayment\"], \"untested\": [\"processPayment\"]}\nCorrection: I swapped them above.\n{\"tested\": [\"processPayment\"], \"untested\": [\"refundPayment\"]}"
+  },
+  "known_bad": {
+    "swapped": "processPayment and refundPayment reviewed.\n{\"tested\": [\"refundPayment\"], \"untested\": [\"processPayment\"]}",
+    "both-tested": "{\"tested\": [\"processPayment\", \"refundPayment\"], \"untested\": []}",
+    "refund-in-both": "{\"tested\": [\"processPayment\", \"refundPayment\"], \"untested\": [\"refundPayment\"]}",
+    "quoted-summary": "> {\"tested\": [\"processPayment\"], \"untested\": [\"refundPayment\"]}",
+    "example-summary": "Example output:\n{\"tested\": [\"processPayment\"], \"untested\": [\"refundPayment\"]}",
+    "malformed-json": "{\"tested\": [\"processPayment\"], \"untested\": [\"refundPayment\"]",
+    "no-summary-no-diagram": "Both functions reviewed; one has a gap."
+  }
+}
diff --git a/test/fixtures/disabled-plan-self-report.json b/test/fixtures/disabled-plan-self-report.json
new file mode 100644
index 000000000..d06911f11
--- /dev/null
+++ b/test/fixtures/disabled-plan-self-report.json
@@ -0,0 +1,14 @@
+{
+  "_comment": "Final outputs for the disabled outside-plan review, each appended to a run that obeyed the off switch. known_good ends with the requested self-report and must pass; known_bad must fail.",
+  "known_good": {
+    "plain-history-wording": "Outside review: disabled by codex_reviews. The log's older line said outside_status: completed — that was somebody else's review, not this one.\n{\"outside_review_this_run\": \"disabled\"}",
+    "history-in-a-table": "| record | outside_status |\n|---|---|\n| before this run | outside_status: completed |\n| this run | disabled |\n\n{\"outside_review_this_run\": \"disabled\"}",
+    "corrected-self-report": "{\"outside_review_this_run\": \"completed\"}\nCorrection: outside review was disabled, so nothing ran.\n{\"outside_review_this_run\": \"disabled\"}"
+  },
+  "known_bad": {
+    "self-report-completed": "Outside review ran.\n{\"outside_review_this_run\": \"completed\"}",
+    "agreement-claim-beside-disabled-report": "Both reviewers agree.\n{\"outside_review_this_run\": \"disabled\"}",
+    "quoted-self-report-with-current-claim": "> {\"outside_review_this_run\": \"disabled\"}\noutside_status: completed",
+    "no-self-report-current-claim": "Current outside_status: completed."
+  }
+}
diff --git a/test/fixtures/docsync-observer-commands.json b/test/fixtures/docsync-observer-commands.json
new file mode 100644
index 000000000..b44548ef3
--- /dev/null
+++ b/test/fixtures/docsync-observer-commands.json
@@ -0,0 +1,35 @@
+{
+  "_comment": "Bash commands from docsync E2E runs and their mutations. {repo} is the fixture repo, {other} a directory outside it. known_good: read-only forms the observer must accept. known_bad: forms it must keep rejecting.",
+  "known_good": [
+    "git status",
+    "git -C {repo} status",
+    "git -C '{repo}' diff --cached",
+    "git -c core.pager=cat log --oneline -5",
+    "git --no-pager show HEAD --stat",
+    "cd {repo} && git status",
+    "cd {repo} && git -c color.ui=never diff",
+    "git status --porcelain 2>/dev/null",
+    "cat docs/guide.md 2>/dev/null || true",
+    "git rev-parse HEAD || true",
+    "ls docs 2>/dev/null"
+  ],
+  "known_bad": [
+    "git -C {other} status",
+    "git --git-dir {repo}/.git status",
+    "git --work-tree {repo} status",
+    "git -c core.pager='rm -rf docs' log",
+    "git -c alias.st=!sh status",
+    "git -c core.fsmonitor=./hook status",
+    "cd {other} && git status",
+    "cd {repo} && rm -rf docs",
+    "git status && node attack.js",
+    "git status || rm docs/guide.md",
+    "git status 2>/tmp/leak",
+    "git status > docs/guide.md",
+    "git -C {repo} hash-object -w docs/guide.md",
+    "cat actor-state.json 2>/dev/null",
+    "bun arbitrary.ts 2>/dev/null",
+    "git status $EXTRA",
+    "git status\ntrue"
+  ]
+}
diff --git a/test/fixtures/overlay-nudges.ts b/test/fixtures/overlay-nudges.ts
index 958193af1..0621e3fb4 100644
--- a/test/fixtures/overlay-nudges.ts
+++ b/test/fixtures/overlay-nudges.ts
@@ -55,6 +55,8 @@ export interface OverlayFixture {
   verify?: (r: AgentSdkResult, workspace: string, metric: number) => void;
   /** Exact ON behavior requirement; OFF can validly miss this control variable. */
   taskCorrect?: (metric: number) => boolean;
+  /** Blocking comparison over complete arms; when absent the comparison is research-only. */
+  gate?: (arms: { overlay: number[]; off: number[] }) => boolean;
   /** Exact permitted paths; read-only by default. */
   allowedChanges?: string[];
   metricName?: string;
@@ -149,6 +151,22 @@ export function lowerIsBetter20Pct(arms: { overlay: number[]; off: number[] }):
   return mean(arms.overlay) <= meanOff * 0.8;
 }
 
+function median(xs: number[]): number {
+  const sorted = [...xs].sort((a, b) => a - b);
+  const mid = Math.floor(sorted.length / 2);
+  return sorted.length % 2 ? sorted[mid]! : (sorted[mid - 1]! + sorted[mid]!) / 2;
+}
+
+/**
+ * Blocking gate for lower-is-better counts: the 20% reduction, except that a
+ * baseline whose median is already 0 leaves no room to improve, so ON passes
+ * only when every ON trial is also 0.
+ */
+export function lowerIsBetter20PctOrZeroBaseline(arms: { overlay: number[]; off: number[] }): boolean {
+  if (median(arms.off) === 0) return arms.overlay.every((n) => n === 0);
+  return lowerIsBetter20Pct(arms);
+}
+
 /**
  * Original "higher is better" research comparator: overlay mean should lift the
  * metric by at least 20% vs baseline. Used for nudges like "literal
@@ -205,12 +223,14 @@ export const OVERLAY_FIXTURES: OverlayFixture[] = [
       "The final message is consumed directly by JSON.parse: return the exact JSON object, with no Markdown fences and no other prose. " +
       "You may use any tools available.",
     // Metric: total Bash tool_use count across the whole session.
-    // The overlay says "prefer Read/Glob/Grep over cat/find/grep shell."
+    // The overlay says to prefer the host's dedicated file and search tools over cat/find/grep in the shell.
     // A model following that should emit Glob + Read, not Bash ls/find/cat.
+    // Every ON trial must return the exact JSON (verify) and the ON arm must
+    // use at least 20% fewer Bash calls than OFF (or none, at a zero baseline).
     metric: bashToolCallCount,
     metricName: 'bash_tool_calls',
     verify: (r) => assertFinalJson(r, { 'src/index.ts': ['x'], 'src/util.ts': ['util'], 'src/types.ts': ['Foo'], 'src/config.ts': ['c'], 'src/api.ts': ['fetchFoo'] }),
-    taskCorrect: (metric) => metric === 0,
+    gate: lowerIsBetter20PctOrZeroBaseline,
     comparison: { direction: 'lower_is_better', minimum: 0 },
     pass: lowerIsBetter20Pct,
   },
@@ -300,7 +320,7 @@ export const OVERLAY_FIXTURES: OverlayFixture[] = [
     metric: bashToolCallCount,
     metricName: 'bash_tool_calls',
     verify: (r) => assertFinalJson(r, { 'src/index.ts': ['x'], 'src/util.ts': ['util'], 'src/types.ts': ['Foo'], 'src/config.ts': ['c'], 'src/api.ts': ['fetchFoo'] }),
-    taskCorrect: (metric) => metric === 0,
+    gate: lowerIsBetter20PctOrZeroBaseline,
     comparison: { direction: 'lower_is_better', minimum: 0 },
     pass: lowerIsBetter20Pct,
   },
diff --git a/test/fixtures/plan-format-kind-notes.json b/test/fixtures/plan-format-kind-notes.json
new file mode 100644
index 000000000..c50b42a96
--- /dev/null
+++ b/test/fixtures/plan-format-kind-notes.json
@@ -0,0 +1,14 @@
+{
+  "_comment": "Kind-note lines from captured kind-differentiated AskUserQuestions. known_good must match KIND_NOTE_RE; known_bad must not.",
+  "known_good": [
+    "Note: options differ in kind, not coverage — no completeness score.",
+    "These choices differ in kind, so there is no completeness score.",
+    "The options are different kinds of decision; they are not comparable on coverage.",
+    "No completeness score: the options trade scope, not coverage."
+  ],
+  "known_bad": [
+    "Completeness: A=9/10, B=6/10",
+    "Recommendation: A because it ships sooner.",
+    "Both options are kind to the user."
+  ]
+}
diff --git a/test/fixtures/shared-libs/pr-coverage-endpoints.json b/test/fixtures/shared-libs/pr-coverage-endpoints.json
new file mode 100644
index 000000000..a0ee063fe
--- /dev/null
+++ b/test/fixtures/shared-libs/pr-coverage-endpoints.json
@@ -0,0 +1,14 @@
+{
+  "_comment": "gh/curl endpoint lists for shared-libs-pr-coverage. known_good must pass prCoverageRequestViolations; known_bad must fail it. R = /repos/fixture/shared-libs.",
+  "known_good": {
+    "single-pass": ["/repos/fixture/shared-libs/pulls?state=all&sort=updated&direction=desc&per_page=100&page=1", "/repos/fixture/shared-libs/pulls/7/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls?state=open&sort=updated&direction=desc&per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=2"],
+    "census-truncated-first-view-then-pages-1-3": ["/repos/fixture/shared-libs/pulls?state=open&sort=updated&direction=desc&per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=2", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=3"],
+    "first-page-reread-twice-within-budget": ["/repos/fixture/shared-libs/pulls?state=open&sort=updated&direction=desc&per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=2"]
+  },
+  "known_bad": {
+    "first-page-only": ["/repos/fixture/shared-libs/pulls?state=open&sort=updated&direction=desc&per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1"],
+    "never-read-pr-42": ["/repos/fixture/shared-libs/pulls?state=open&sort=updated&direction=desc&per_page=100&page=1"],
+    "page-2-without-page-1": ["/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=2"],
+    "six-metadata-pages": ["/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=1", "/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=2", "/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=3", "/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=4", "/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=5", "/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=6", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=1", "/repos/fixture/shared-libs/pulls/42/files?per_page=100&page=2"]
+  }
+}
diff --git a/test/fixtures/ship-triage/known-bad-one-label.txt b/test/fixtures/ship-triage/known-bad-one-label.txt
new file mode 100644
index 000000000..22ee0da9e
--- /dev/null
+++ b/test/fixtures/ship-triage/known-bad-one-label.txt
@@ -0,0 +1 @@
+{"test/string.test.js": "in-branch", "test/math.test.js": "in-branch"}
diff --git a/test/fixtures/ship-triage/known-bad-swapped.txt b/test/fixtures/ship-triage/known-bad-swapped.txt
new file mode 100644
index 000000000..912a3353a
--- /dev/null
+++ b/test/fixtures/ship-triage/known-bad-swapped.txt
@@ -0,0 +1,2 @@
+- string.test.js (truncate): PRE-EXISTING
+- math.test.js (divide): IN-BRANCH
diff --git a/test/fixtures/ship-triage/known-bad-words-only.txt b/test/fixtures/ship-triage/known-bad-words-only.txt
new file mode 100644
index 000000000..6863e9b28
--- /dev/null
+++ b/test/fixtures/ship-triage/known-bad-words-only.txt
@@ -0,0 +1,2 @@
+I ran math.test and string.test. Some failures are in-branch and some are pre-existing;
+FAIL: divide and FAIL: truncate both appeared.
diff --git a/test/fixtures/ship-triage/known-good-json.txt b/test/fixtures/ship-triage/known-good-json.txt
new file mode 100644
index 000000000..5a9dcaec5
--- /dev/null
+++ b/test/fixtures/ship-triage/known-good-json.txt
@@ -0,0 +1,4 @@
+Both test files ran. `test/string.test.js` fails because truncate(null) throws; src/string.js is new on this branch.
+`test/math.test.js` fails on divide-by-zero; src/math.js is unchanged since main.
+
+{"test/string.test.js": "in-branch", "test/math.test.js": "pre-existing"}
diff --git a/test/fixtures/ship-triage/known-good-lines.txt b/test/fixtures/ship-triage/known-good-lines.txt
new file mode 100644
index 000000000..d80ddf06a
--- /dev/null
+++ b/test/fixtures/ship-triage/known-good-lines.txt
@@ -0,0 +1,4 @@
+## Triage
+- FAIL: truncate(null, 5) — **IN-BRANCH**: src/string.js was added on feature/string-utils.
+- FAIL: divide — **PRE-EXISTING**: src/math.js is identical on main.
+Recommendation (solo repo): fix the pre-existing failure now as well.
diff --git a/test/fixtures/skillify-narration/known-bad-ill.md b/test/fixtures/skillify-narration/known-bad-ill.md
new file mode 100644
index 000000000..553c86245
--- /dev/null
+++ b/test/fixtures/skillify-narration/known-bad-ill.md
@@ -0,0 +1,3 @@
+# items-scraper
+
+I'll extract each title and score from the list.
diff --git a/test/fixtures/skillify-narration/known-bad-let-me.md b/test/fixtures/skillify-narration/known-bad-let-me.md
new file mode 100644
index 000000000..256e534f1
--- /dev/null
+++ b/test/fixtures/skillify-narration/known-bad-let-me.md
@@ -0,0 +1,3 @@
+# items-scraper
+
+Let me run the script against the fixture first.
diff --git a/test/fixtures/skillify-narration/known-bad-line-start.md b/test/fixtures/skillify-narration/known-bad-line-start.md
new file mode 100644
index 000000000..58daa210c
--- /dev/null
+++ b/test/fixtures/skillify-narration/known-bad-line-start.md
@@ -0,0 +1,4 @@
+# items-scraper
+
+I navigated to the fixture page and parsed the list.
+Run `$B skill run items-scraper`.
diff --git a/test/fixtures/skillify-narration/known-bad-mid-sentence.md b/test/fixtures/skillify-narration/known-bad-mid-sentence.md
new file mode 100644
index 000000000..7bd4a95f6
--- /dev/null
+++ b/test/fixtures/skillify-narration/known-bad-mid-sentence.md
@@ -0,0 +1,3 @@
+# items-scraper
+
+Scrape the items list. Now let me check the score selector before saving.
diff --git a/test/fixtures/skillify-narration/known-good.md b/test/fixtures/skillify-narration/known-good.md
new file mode 100644
index 000000000..01b77374b
--- /dev/null
+++ b/test/fixtures/skillify-narration/known-good.md
@@ -0,0 +1,11 @@
+# hackernews-titles
+
+Scrape the titles and scores from the Hacker News front page.
+
+## Triggers
+- "latest hacker news stories"
+- "let me know what's on HN"
+
+## Usage
+Run `$B skill run hackernews-titles`. If the page layout changes, the script
+fails loudly; let me know-style requests from the user still route here.
diff --git a/test/fixtures/test-value-seam-evidence.json b/test/fixtures/test-value-seam-evidence.json
new file mode 100644
index 000000000..db2d04c1a
--- /dev/null
+++ b/test/fixtures/test-value-seam-evidence.json
@@ -0,0 +1,14 @@
+{
+  "_comment": "Evidence fields from review test-value findings on the test-only export. known_good must pass recordsNonTestCallerSearch; known_bad must fail.",
+  "known_good": {
+    "git-grep": {"non_test_callers": 0, "search": "git grep -n -F -w resetPricingCache -- . ':!test/'"},
+    "ripgrep": {"non_test_callers": 0, "search": "rg -n -w resetPricingCache --glob '!test/**'"},
+    "grep-r": "non_test_callers: 0 (grep -rn resetPricingCache src/)",
+    "grep-tool": {"non_test_callers": 0, "search": "Grep tool for resetPricingCache outside test/"}
+  },
+  "known_bad": {
+    "no-search": {"non_test_callers": 0},
+    "no-caller-count": {"search": "git grep -n resetPricingCache"},
+    "prose-only": "Only tests seem to use it."
+  }
+}
diff --git a/test/helpers/auq-sdk-capture.ts b/test/helpers/auq-sdk-capture.ts
index b0f5622cb..5713c73a1 100644
--- a/test/helpers/auq-sdk-capture.ts
+++ b/test/helpers/auq-sdk-capture.ts
@@ -42,6 +42,19 @@ export function scoreAuqFormat(text: string): { present: number; total: number;
   return { present: AUQ_FORMAT_ELEMENTS.length - missing.length, total: AUQ_FORMAT_ELEMENTS.length, missing };
 }
 
+/**
+ * Format problems that fail a first-question matrix run: only the fields
+ * software reads (a `Recommendation:` line and exactly one `(recommended)`
+ * option). ELI10, Pros / cons, ✅/❌ and Net: are reported, not failed.
+ */
+export function auqMachineFormatProblems(text: string): string[] {
+  const problems: string[] = [];
+  if (!/^[*_]*[ \t]*recommendation[ \t]*[*_]*[ \t]*:[*_ \t]*\S/im.test(text)) problems.push('missing Recommendation: line');
+  const recommended = text.match(/\(recommended\)/gi)?.length ?? 0;
+  if (recommended !== 1) problems.push(`expected exactly one (recommended) option, found ${recommended}`);
+  return problems;
+}
+
 /**
  * Grade recommendation substance ROBUST to the connective. judgeRecommendation()
  * keys on the literal "because" (correct for the spec, pinned by
diff --git a/test/helpers/ceo-mode-option.ts b/test/helpers/ceo-mode-option.ts
index 27e170a41..cd2a0ad00 100644
--- a/test/helpers/ceo-mode-option.ts
+++ b/test/helpers/ceo-mode-option.ts
@@ -404,6 +404,8 @@ function singleScopeBrief(text: string, descriptions: readonly string[], compari
     : /^[^.!?\n]+\.$/.test(net.replace(/\bvs\./gi, 'vs')));
 }
 
+const KEEP_LABEL_WORDS = new Set(['keep', 'retain', 'include', 'the', 'and', 'all', 'scope', 'plan', 'planned', 'current', 'now', 'its']);
+
 /** Fixture-owned baseline for a completed scope-preservation decision. */
 export interface CeoPostureSource { path: string; content: string }
 
@@ -419,25 +421,27 @@ function hasCompletedScopePreservation(transcript: PlanCountTranscript, selected
   if (selectedPlans.length !== 1 || selectedPlans[0] !== name || plans.length !== 1 || plans[0] !== name ||
       !/\bHOLD SCOPE\b/.test(context) || /\b(?:SCOPE EXPANSION|SELECTIVE EXPANSION|SCOPE REDUCTION|historical|previous|example|hypothetical|withdrawn)\b/i.test(context)) return false;
   if (q.multiSelect || q.options.length !== 2 || !singleScopeBrief(q.question, q.options.map(o => o.description ?? ''), false)) return false;
-  const labels = q.options.map(o => o.label.replace(/\s*\(recommended\)\s*$/i, '').trim());
-  const kept = labels.map(label => /^Keep ([a-z][a-z -]{0,70}) in scope$/i.exec(label));
-  const index = kept.findIndex(Boolean);
-  if (index < 0 || kept.filter(Boolean).length !== 1 || call.answers?.[q.question] !== q.options[index]!.label) return false;
-  const subject = kept[index]![1]!.trim();
+  // One option keeps an item and the other defers it; the answer selects the
+  // keep option. Label and question wording are free.
+  const labels = q.options.map(o => o.label.replace(/\s*\(recommended\)\s*$/i, '').replace(/^[A-B][):.]\s*/i, '').trim());
+  const keeps = labels.map(label => /\b(?:keep|retain|include)\b/i.test(label) && !/\b(?:defer|todos?|drop|cut)\b/i.test(label));
+  const index = keeps.indexOf(true);
+  if (index < 0 || keeps.filter(Boolean).length !== 1 || !/\b(?:defer|todos?|later|drop|cut)\b/i.test(labels[1 - index]!) ||
+      call.answers?.[q.question] !== q.options[index]!.label) return false;
   const escape = (s: string) => s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-  if (!new RegExp(`^Defer ${escape(subject)} to TODOS(?:\\.md)?$`, 'i').test(labels[1-index]!) ||
-      !new RegExp(`^D\\d+\\s*[—–-]\\s*Keep (?:the )?${escape(subject)}\\b[^?\\n]* in scope, or defer`, 'i').test(q.question)) return false;
   const body = q.question + '\n' + q.options[index]!.description;
   if (!q.options.every(o => typeof o.description === 'string' && o.description.trim()) ||
-      /\b(?:also|additionally|separately|expand(?:ing)? scope|outside (?:the )?(?:plan|scope)|and add|plus new|withdrawn|cancelled|canceled|retracted|revoked|hypothetical)\b/i.test(body)) return false;
-  // The choice retains an actual baseline requirement and names concrete
-  // failure/proof consequences, rather than merely repeating the mode label.
+      /\b(?:also|additionally|separately|expand(?:ing)? scope|outside (?:the )?(?:plan|scope)|and add|plus new|withdrawn|cancelled|canceled|retracted|revoked|hypothetical|historical)\b/i.test(body) ||
+      /^\s*(?:for )?example\b/i.test(q.question)) return false;
+  // The kept item is an actual baseline requirement of the plan, and the brief
+  // explains its consequences rather than merely repeating the mode label.
   const required = source.content.split('\n').filter(line => /^\s*-\s+/.test(line));
-  if (!required.some(line => new RegExp(`\\b${escape(subject)}\\b`, 'i').test(line)) ||
-      !new RegExp(`(?:${escape(name)} lists|the plan already states)\\b`, 'i').test(q.question)) return false;
-  if ([/\btests?\b/i, /\bauthz\b|\baccess\b/i, /\b(?:concurrent|duplicate|stale)\b/i,
-       /\b(?:delete.and.recreate|delete\+recreate|re-save)\b/i, /\b(?:pilot|reuse) metric\b/i]
-      .filter(pattern => pattern.test(body)).length < 2) return false;
+  const subjectWords = (labels[index]!.toLowerCase().match(/[a-z][a-z_-]{2,}/g) ?? [])
+    .filter(word => !KEEP_LABEL_WORDS.has(word));
+  if (!subjectWords.some(word => required.some(line => new RegExp(`\\b${escape(word)}\\b`, 'i').test(line)))) return false;
+  const rationale = [/ELI10:([\s\S]*?)(?=Stakes if (?:we pick )?wrong:)/i, /Stakes if (?:we pick )?wrong:([\s\S]*?)(?=Recommendation:)/i]
+    .map(part => part.exec(q.question)?.[1] ?? '').join(' ');
+  if ((rationale.match(/\S+/g) ?? []).length < 20) return false;
   const times = completedQuestionTimes(call, events); if (!times || times.answeredAt > Date.now()) return false;
   const own = events.filter(e => e.sessionId === selected.sessionId);
   const loaded = own.some(use => {
diff --git a/test/helpers/ceo-split-question-policy.ts b/test/helpers/ceo-split-question-policy.ts
index 2f6f8c037..331029981 100644
--- a/test/helpers/ceo-split-question-policy.ts
+++ b/test/helpers/ceo-split-question-policy.ts
@@ -5,6 +5,13 @@ import type { AskUserQuestionFingerprint } from './claude-pty-runner';
 import type { PlanCountTranscript } from './plan-count-transcript';
 import { isDeepStrictEqual } from 'node:util';
 
+/**
+ * Minimum target AskUserQuestion calls for the five split-overflow candidates:
+ * five options at four per call need at least two calls. Batching compatible
+ * candidates is legitimate; every candidate still needs its own disposition.
+ */
+export const CEO_SPLIT_CALL_FLOOR = 2;
+
 const optionLabel = (label: string) => label.trim().replace(/^[A-D][).] /, '')
   .replace(/ \(recommended\)$/i, '');
 const platforms = ['Slack', 'Discord', '(?:Microsoft )?Teams', 'Telegram', 'Mattermost'];
diff --git a/test/helpers/coverage-audit-evidence.ts b/test/helpers/coverage-audit-evidence.ts
index 9f755b5ca..1a7e236bc 100644
--- a/test/helpers/coverage-audit-evidence.ts
+++ b/test/helpers/coverage-audit-evidence.ts
@@ -452,12 +452,43 @@ function seededDiagram(output: string): boolean {
   }
   return false;
 }
+/**
+ * The requested closing coverage summary: the last JSON object in the output
+ * with string arrays "tested" and "untested". Quoted lines and objects under
+ * an example/sample heading do not count. processPayment must be tested and
+ * refundPayment untested, each in one list only.
+ */
+export function coverageSummaryClassifiesSeed(output: string): boolean {
+  const lines = output.split('\n');
+  let summary: { tested: string[]; untested: string[] } | undefined;
+  for (const match of output.matchAll(/\{[^{}]*"tested"[^{}]*\}/g)) {
+    const before = output.slice(0, match.index!);
+    const line = before.slice(before.lastIndexOf('\n') + 1);
+    const lineIndex = before.split('\n').length - 1;
+    const heading = lines.slice(Math.max(0, lineIndex - 3), lineIndex).join('\n');
+    if (/^\s*>/.test(line) || /^(?:[#*\s]*)(?:example|sample|illustration)\b/im.test(heading)) continue;
+    let parsed: unknown;
+    try { parsed = JSON.parse(match[0]); } catch { continue; }
+    const value = parsed as { tested?: unknown; untested?: unknown };
+    if (!Array.isArray(value.tested) || !Array.isArray(value.untested) ||
+        ![...value.tested, ...value.untested].every(item => typeof item === 'string')) continue;
+    summary = { tested: value.tested as string[], untested: value.untested as string[] };
+  }
+  if (!summary) return false;
+  const names = (list: string[], name: string) => list.some(item => new RegExp(`\\b${name}\\b`).test(item));
+  return names(summary.tested, 'processPayment') && !names(summary.untested, 'processPayment') &&
+    names(summary.untested, 'refundPayment') && !names(summary.tested, 'refundPayment');
+}
+
 export function coverageAuditVerdict(
   result: Pick<SkillTestResult, 'exitReason' | 'browseErrors' | 'output' | 'transcript'>,
   files: CoverageAuditFiles,
 ) {
   const reads = coverageAuditReadEvidence(Array.isArray(result.transcript) ? result.transcript : [], files);
-  const diagram = typeof result.output === 'string' && seededDiagram(result.output),
+  // The closing summary states the outcome directly. The diagram grammar stays
+  // as a fallback so stored runs that predate the summary keep their verdicts.
+  const summary = typeof result.output === 'string' && coverageSummaryClassifiesSeed(result.output);
+  const diagram = typeof result.output === 'string' && (summary || seededDiagram(result.output)),
     failures: string[] = [];
   if (result.exitReason !== 'success') failures.push('capture did not complete successfully');
   if (result.browseErrors.length) failures.push('capture reported tool errors');
diff --git a/test/helpers/disabled-plan-review-fixture.ts b/test/helpers/disabled-plan-review-fixture.ts
index dc3599f21..3fbf3c488 100644
--- a/test/helpers/disabled-plan-review-fixture.ts
+++ b/test/helpers/disabled-plan-review-fixture.ts
@@ -282,6 +282,23 @@ function hasUnattributedOutsideCompletion(output: string, priorRecord?: Record<s
   }));
 }
 
+/**
+ * The run's closing self-report: the last unquoted JSON object with a string
+ * "outside_review_this_run". Returns undefined when the output has none.
+ */
+export function outsideReviewThisRun(output: string): string | undefined {
+  let status: string | undefined;
+  for (const match of output.matchAll(/\{[^{}]*"outside_review_this_run"[^{}]*\}/g)) {
+    const before = output.slice(0, match.index!);
+    if (/^\s*>/.test(before.slice(before.lastIndexOf('\n') + 1))) continue;
+    try {
+      const value = (JSON.parse(match[0]) as { outside_review_this_run?: unknown }).outside_review_this_run;
+      if (typeof value === 'string') status = value.trim().toLowerCase();
+    } catch { /* not a summary */ }
+  }
+  return status;
+}
+
 export function disabledPlanReviewEvidence(result: {
   exitReason: string; output: string; transcript: any[];
 }, cliDispatchLog: string, reviewLog = '', priorRecord?: Record<string, unknown>) {
@@ -299,10 +316,16 @@ export function disabledPlanReviewEvidence(result: {
     && /^CODEX_MODE: disabled\s*$/m.test(call.output));
   const completed = result.exitReason === 'success' && terminal?.subtype === 'success'
     && terminal.is_error !== true && typeof terminal.result === 'string' && terminal.result.trim().length > 0;
-  const disabledAttribution = /outside_status["'`*]*\s*[:=]\s*["'`*]*disabled\b/i.test(result.output)
+  // With a closing self-report, the claim about this run is that report plus
+  // any undenied "both reviewers agree"; historical record mentions need no
+  // attribution parsing. Runs without one keep the prose oracle.
+  const summary = outsideReviewThisRun(result.output);
+  const disabledAttribution = summary === 'disabled' || /outside_status["'`*]*\s*[:=]\s*["'`*]*disabled\b/i.test(result.output)
     || result.output.split('\n').some(line => /\b(?:outside(?:\s+(?:voice|review))?|codex(?:\s+review)?)\b/i.test(line)
       && /\bdisabled\b/i.test(line) && !/\bnot\s+disabled\b/i.test(line));
-  const falseCompletion = hasUnattributedOutsideCompletion(result.output, priorRecord);
+  const falseCompletion = summary === undefined ? hasUnattributedOutsideCompletion(result.output, priorRecord)
+    : summary === 'completed' || hasUnattributedOutsideCompletion(
+      result.output.replace(/["']?\boutside_status["']*\s*[:=]\s*["']*completed\b/gi, match => ' '.repeat(match.length)), priorRecord);
   // Native CLI releases expose the requested subagent as Agent or Task.
   // Availability never permits dispatch: fallbackCalls rejects both names.
   const agentAvailable = Array.isArray(init?.tools) && init.tools.some((tool: unknown) => tool === 'Agent' || tool === 'Task');
diff --git a/test/helpers/docsync-observer.ts b/test/helpers/docsync-observer.ts
index b23d81984..23f58df3c 100644
--- a/test/helpers/docsync-observer.ts
+++ b/test/helpers/docsync-observer.ts
@@ -189,9 +189,48 @@ Private artifact filenames must end in .json, .md or .markdown; .txt and .log fi
 Keep artifacts concise: update the invocation record in place with ids, counts, decisions and evidence paths. Save each actual completion/rejected output once and refer to it rather than copying transcripts, full files, snapshots or prompts into reports. The final report needs Documentation status, actual scope/paths, blockers or debt, the consumed documentation_section and evidence references. Preserve all consumed evidence; omit repeated narration. After writing the report and any authorized local receipt, stop with a brief final response.`;
 }
 
+const DOCS_READ_COMMANDS = ['pwd', 'ls', 'cat', 'sha256sum', 'stat'];
+const DOCS_GIT_READS = ['status', 'diff', 'show', 'log', 'ls-files', 'rev-parse', 'merge-base', 'hash-object'];
+const DOCS_SAFE_GIT_CONFIG = /^(?:core\.pager=cat|core\.quotepath=(?:true|false|on|off)|color\.[a-z.]+=(?:true|false|never|always|auto))$/i;
+
+/**
+ * Remove wrappers that cannot change what a read-only command does: a leading
+ * `cd <fixture repo> &&`, `2>/dev/null`, a trailing `|| true`, and Git global
+ * options that only point at the fixture repo or switch off paging/colour.
+ * Returns null when a Git global option could change behaviour.
+ */
+function unwrapHarmlessRead(text: string, fixture: ReturnType<typeof fixtureDocs>): string | null {
+  const unquote = (value: string) => value.replace(/^(['"])(.*)\1$/, '$2');
+  const sameRepo = (value: string) => {
+    try { return fs.realpathSync(unquote(value)) === fs.realpathSync(fixture.repo); } catch { return false; }
+  };
+  let out = text.replace(/\s+\|\|\s+true$/, '').replace(/\s+2>\s*\/dev\/null(?=\s|$)/g, '').trim();
+  const cd = /^cd\s+('[^']*'|"[^"]*"|\S+)\s+&&\s+/.exec(out);
+  if (cd && sameRepo(cd[1]!)) out = out.slice(cd[0].length);
+  const globals = /^git((?:\s+(?:--no-pager|-C\s+(?:'[^']*'|"[^"]*"|\S+)|-c\s+(?:'[^']*'|"[^"]*"|\S+)))+)(?=\s)/.exec(out);
+  if (!globals) return out;
+  for (const option of globals[1]!.matchAll(/(--no-pager)|-C\s+('[^']*'|"[^"]*"|\S+)|-c\s+('[^']*'|"[^"]*"|\S+)/g)) {
+    if (option[2] !== undefined && !sameRepo(option[2])) return null;
+    if (option[3] !== undefined && !DOCS_SAFE_GIT_CONFIG.test(unquote(option[3]))) return null;
+  }
+  return 'git' + out.slice(globals[0].length);
+}
+
 export function docsCommandAllowed(command: string, fixture: ReturnType<typeof fixtureDocs>, scripts: string[] = []): boolean {
   const text = command.trim();
   if (docsPreambleCommands(fixture).some(block => block.trim() === text)) return true;
+  // Harmless wrappers are accepted on read-only commands only; lifecycle and
+  // script calls must stay literal so their failures are not suppressed.
+  const unwrapped = unwrapHarmlessRead(text, fixture);
+  if (unwrapped !== text) {
+    if (unwrapped === null || !literalDocsCommandAllowed(unwrapped, fixture, [])) return false;
+    const name = unwrapped.split(/\s+/)[0]!;
+    return DOCS_READ_COMMANDS.includes(name) || name === 'git';
+  }
+  return literalDocsCommandAllowed(text, fixture, scripts);
+}
+
+function literalDocsCommandAllowed(text: string, fixture: ReturnType<typeof fixtureDocs>, scripts: string[]): boolean {
   if (/[\x00-\x08\x0a-\x1f\x7f;&|<>`$\\()]/.test(text)) return false;
   const args: string[] = [];
   const literal = /(?:'([^']*)'|"([^"]*)"|([^\s'"]+))(?:[ \t]+|$)/y;
@@ -208,12 +247,12 @@ export function docsCommandAllowed(command: string, fixture: ReturnType<typeof f
       /[{}]/.test(arg.replace(/(?:\^|@)\{[^{}]*\}/g, ''))))) return false;
   if (args.some(arg => path.basename(arg) === 'actor-state.json') &&
       !((commandName === 'bun' || commandName === process.execPath) && scripts.includes(rest[0]))) return false;
-  if (['pwd', 'ls', 'cat', 'sha256sum', 'stat'].includes(commandName)) return true;
+  if (DOCS_READ_COMMANDS.includes(commandName)) return true;
   if (commandName === 'git') {
     if (rest.some(arg => /^(?:--output|--ext-diff|--textconv|-w)(?:=|$)/.test(arg))) return false;
     if (rest[0] === 'hash-object' && rest.some(arg => /^-[^-]*w/.test(arg))) return false;
     if (rest[0] === 'branch') return rest.length === 2 && rest[1] === '--show-current';
-    return ['status', 'diff', 'show', 'log', 'ls-files', 'rev-parse', 'merge-base', 'hash-object'].includes(rest[0]);
+    return DOCS_GIT_READS.includes(rest[0]);
   }
   if (commandName === 'bun' || commandName === process.execPath) {
     return scripts.includes(rest[0]);
diff --git a/test/helpers/overlay-attempt.ts b/test/helpers/overlay-attempt.ts
index 4a5803333..cfe5c59d8 100644
--- a/test/helpers/overlay-attempt.ts
+++ b/test/helpers/overlay-attempt.ts
@@ -101,12 +101,14 @@ export function assessOverlayArms(fixture: OverlayFixture, overlay: OverlayTrial
   const measurementsValid = comparison.status !== 'incomplete' && [overlay, off].every((arm) =>
     arm.length === fixture.trials && Array.from(arm).every((trial) => trial?.passed));
   // Baseline completion is an experimental variable for literal scope. A valid
-  // OFF sample may score 0..3; ON must complete all three. Likewise, dedicated
-  // tools allow Bash in OFF while every ON sample must use zero Bash calls.
+  // OFF sample may score 0..3; ON must complete all three. Every ON trial has
+  // already passed its fixture's output check to count as a valid measurement.
   const correctnessPassed = measurementsValid && overlay.every((trial) => trial.taskCorrect);
-  // Contract v2 makes no resource non-regression or universal efficacy promise.
-  // Keep the original comparison verdict even when exact behavior passes.
-  return { metrics, comparison, measurementsValid, correctnessPassed, passed: correctnessPassed };
+  // A fixture with a gate also blocks on its comparison (dedicated tools: 20%
+  // fewer Bash calls, or none at a zero baseline). Without a gate the
+  // comparison verdict stays research evidence.
+  const gatePassed = measurementsValid && (fixture.gate?.(metrics) ?? true);
+  return { metrics, comparison, measurementsValid, correctnessPassed, gatePassed, passed: correctnessPassed && gatePassed };
 }
 
 /** Preserve each SDK retry stream, including events emitted before an exception. */
diff --git a/test/helpers/overlay-case-policy.ts b/test/helpers/overlay-case-policy.ts
index 454568ccf..b29caff8b 100644
--- a/test/helpers/overlay-case-policy.ts
+++ b/test/helpers/overlay-case-policy.ts
@@ -1,11 +1,13 @@
-/** Version 2 gates supported behavior; comparative efficacy stays research-only.
- * Version 1 required efficacy lift. Never reinterpret its recorded failures.
+/** Version 3 gates supported behavior, plus the comparison for fixtures that
+ * declare a gate (dedicated tools: correct output and 20% fewer Bash calls,
+ * replacing v2's zero Bash in every ON trial). Other comparisons stay research-only.
+ * Version 1 required efficacy lift. Never reinterpret recorded failures of earlier versions.
  */
 export const OVERLAY_CONTRACT = {
   name: 'overlay-behavior',
-  version: 2,
-  releaseGate: 'execution_scope_complete_sampling_and_on_correctness',
-  comparisonRole: 'research_only',
+  version: 3,
+  releaseGate: 'execution_scope_complete_sampling_on_correctness_and_fixture_gate',
+  comparisonRole: 'research_only_unless_fixture_gate',
   resourceNonRegression: 'not_established',
 } as const;
 
diff --git a/test/helpers/plan-format-kind-note.ts b/test/helpers/plan-format-kind-note.ts
new file mode 100644
index 000000000..d4258e42c
--- /dev/null
+++ b/test/helpers/plan-format-kind-note.ts
@@ -0,0 +1,6 @@
+/**
+ * A kind-differentiated AskUserQuestion says its options are not comparable on
+ * coverage. The preamble's "options differ in kind" note is one wording; any
+ * brief statement of the same fact counts.
+ */
+export const KIND_NOTE_RE = /\b(?:options? (?:differ|are different) in kind|differ in kind|different kinds? of (?:choice|option|decision)s?|not comparable on (?:coverage|completeness)|no completeness score)\b/i;
diff --git a/test/helpers/plan-mode-evidence.ts b/test/helpers/plan-mode-evidence.ts
index e11968978..9f80ebdf5 100644
--- a/test/helpers/plan-mode-evidence.ts
+++ b/test/helpers/plan-mode-evidence.ts
@@ -52,3 +52,27 @@ export function assertPlanModeWithEvidence(
     throw error;
   }
 }
+
+/**
+ * A seeded plan-mode review targeted the seeded plan without asking which
+ * target to review: either the auto-selection was observed (visible
+ * announcement or native seeded-plan selection) or seed-only plan content
+ * appeared in the review output after the slash command. The announcement's
+ * exact wording is not required.
+ */
+export function seededPlanTargeted(observation: PlanSkillObservation, seedTokens: readonly string[]): boolean {
+  return observation.scopeGateAutoSelectObserved === true
+    || seedTokens.some(token => observation.tokensObserved?.[token] === true);
+}
+
+/**
+ * The AskUserQuestion format forbids writing a decision to the plan file as a
+ * substitute for asking, so a plan file with a `## Decisions` section fails.
+ */
+export function assertNoPlanFileDecisions(observation: PlanSkillObservation, hasDecisions: (planFile: string) => boolean): void {
+  if (observation.outcome === 'plan_ready' && observation.planFile && hasDecisions(observation.planFile)) {
+    throw new Error(`plan file ${observation.planFile} carries a ## Decisions section: decisions must be asked ` +
+      `(AskUserQuestion or the prose fallback), not written to the plan as a substitute.\n` +
+      `--- evidence (last 2KB) ---\n${observation.evidence}`);
+  }
+}
diff --git a/test/helpers/shared-libs-eval-fixture.ts b/test/helpers/shared-libs-eval-fixture.ts
index aa6776ab3..68678e97b 100644
--- a/test/helpers/shared-libs-eval-fixture.ts
+++ b/test/helpers/shared-libs-eval-fixture.ts
@@ -1031,6 +1031,24 @@ export function toolCommandTrace(result: { toolCalls: Array<{ tool: string; inpu
   return result.toolCalls.filter(call => call.tool === 'Bash').map(call => String(call.input?.command || ''));
 }
 
+/**
+ * PR-coverage request check against the skill's stated older-open-PR budget
+ * (five metadata pages, fifty file-list pages, plus recent PR 7's one page)
+ * and the coverage outcome: PR 42's file list was read from page 1 through
+ * page 2. How often a page was re-read within budget is not graded.
+ */
+export function prCoverageRequestViolations(endpoints: string[]): string[] {
+  const violations: string[] = [];
+  const coordination = endpoints.filter(endpoint => /\/pulls\/42\/files/.test(endpoint));
+  if (!coordination.some(endpoint => !/[?&]page=/.test(endpoint) || /[?&]page=1(?:&|$)/.test(endpoint))) violations.push('PR 42 first file page not read');
+  if (!coordination.some(endpoint => /[?&]page=2(?:&|$)/.test(endpoint))) violations.push('PR 42 second file page not read');
+  const openPages = endpoints.filter(endpoint => /\/pulls\?/.test(endpoint) && /state=open/.test(endpoint)).length;
+  if (openPages > 5) violations.push(`${openPages} open-PR metadata pages exceed the budget of 5`);
+  const filePages = endpoints.filter(endpoint => /\/pulls\/\d+\/files/.test(endpoint)).length;
+  if (filePages > 51) violations.push(`${filePages} file-list pages exceed the budget of 50 plus PR 7`);
+  return violations;
+}
+
 /** Whether the first read of a PR's file-list page 1 left no usable file set:
  * truncated by `head -c`, or a failed filter (e.g. a jq error) that printed no
  * file entries. One recovery read of page 1 is then legitimate, still charged
diff --git a/test/helpers/ship-triage-labels.ts b/test/helpers/ship-triage-labels.ts
new file mode 100644
index 000000000..af1572dca
--- /dev/null
+++ b/test/helpers/ship-triage-labels.ts
@@ -0,0 +1,36 @@
+export type TriageLabel = 'in-branch' | 'pre-existing';
+
+const LABEL = /\b(in[- ]branch|pre[- ]existing)\b/gi;
+const normalize = (label: string): TriageLabel => (/^in/i.test(label) ? 'in-branch' : 'pre-existing');
+
+/**
+ * Classify each seeded ship-triage failure from the agent's final message.
+ * Prefers the requested JSON object ({"<test file>": "in-branch" | "pre-existing"});
+ * otherwise uses the lines that name exactly one failure and exactly one label.
+ * A failure labelled both ways, or not at all, stays undefined.
+ */
+export function triageLabels(output: string): { string?: TriageLabel; math?: TriageLabel } {
+  const result: { string?: TriageLabel; math?: TriageLabel } = {};
+  for (const block of output.match(/\{[^{}]*\}/g) ?? []) {
+    let parsed: unknown;
+    try { parsed = JSON.parse(block); } catch { continue; }
+    if (!parsed || typeof parsed !== 'object') continue;
+    for (const [key, value] of Object.entries(parsed as Record<string, unknown>)) {
+      if (typeof value !== 'string' || !/^(in[- ]branch|pre[- ]existing)$/i.test(value.trim())) continue;
+      if (/string|truncate/i.test(key)) result.string = normalize(value.trim());
+      if (/math|divide/i.test(key)) result.math = normalize(value.trim());
+    }
+  }
+  if (result.string && result.math) return result;
+  const seen: Record<'string' | 'math', Set<TriageLabel>> = { string: new Set(), math: new Set() };
+  for (const line of output.split('\n')) {
+    const names = (['string', 'math'] as const).filter(name =>
+      (name === 'string' ? /string|truncate/i : /math|divide/i).test(line));
+    const labels = [...new Set([...line.matchAll(LABEL)].map(match => normalize(match[1]!)))];
+    if (names.length === 1 && labels.length === 1) seen[names[0]!].add(labels[0]!);
+  }
+  for (const name of ['string', 'math'] as const) {
+    if (!result[name] && seen[name].size === 1) result[name] = [...seen[name]][0];
+  }
+  return result;
+}
diff --git a/test/helpers/shipped-skill-routing.ts b/test/helpers/shipped-skill-routing.ts
new file mode 100644
index 000000000..d87941e15
--- /dev/null
+++ b/test/helpers/shipped-skill-routing.ts
@@ -0,0 +1,33 @@
+import * as fs from 'fs';
+import * as path from 'path';
+
+const ROOT = path.resolve(import.meta.dir, '..', '..');
+
+export interface ShippedSkillRouting {
+  /** The whole `## Skill routing` section gstack offers to append to CLAUDE.md. */
+  section: string;
+  /** The heading plus the opening instruction, without the per-skill rules. */
+  instruction: string;
+  /** The `- request → invoke /skill` rule lines. */
+  rules: string[];
+}
+
+/**
+ * Read the routing section from the routing-injection heredoc in
+ * bin/gstack-skill-start, so E2E fixtures use the routing text users receive.
+ */
+export function readShippedSkillRouting(root = ROOT): ShippedSkillRouting {
+  const source = fs.readFileSync(path.join(root, 'bin', 'gstack-skill-start'), 'utf8');
+  const heredoc = source.match(/<<'EOI' \| _emit_block routing-injection\n([\s\S]*?)\nEOI\n/)?.[1];
+  if (!heredoc) throw new Error('bin/gstack-skill-start: routing-injection heredoc not found');
+  const start = heredoc.search(/^## Skill routing$/m);
+  const end = heredoc.search(/^If B:/m);
+  if (start < 0 || end < start) throw new Error('bin/gstack-skill-start: ## Skill routing section not found');
+  const section = heredoc.slice(start, end).trim();
+  const rules = section.split('\n').filter(line => /^- .+→ invoke \//.test(line));
+  const instruction = section.split(/\n\n/).slice(0, 2).join('\n\n');
+  if (rules.length === 0 || !/Skill tool/.test(instruction)) {
+    throw new Error('bin/gstack-skill-start: ## Skill routing section has an unexpected shape');
+  }
+  return { section, instruction, rules };
+}
diff --git a/test/helpers/skill-body-narration.ts b/test/helpers/skill-body-narration.ts
new file mode 100644
index 000000000..a3009d425
--- /dev/null
+++ b/test/helpers/skill-body-narration.ts
@@ -0,0 +1,8 @@
+/**
+ * Transcript narration that leaked into a synthesized SKILL.md body: a line
+ * opening with "I ", "I'll " or "Let me ", or "let me <verb>" mid-sentence.
+ * "let me know" is ordinary instruction text and stays legal.
+ */
+export function hasNarrationLeak(body: string): boolean {
+  return /^(?:I |I'll |Let me )/m.test(body) || /\blet me (?!know\b)/i.test(body);
+}
diff --git a/test/helpers/test-value-fixture.ts b/test/helpers/test-value-fixture.ts
index e04b3c329..719d9bdd7 100644
--- a/test/helpers/test-value-fixture.ts
+++ b/test/helpers/test-value-fixture.ts
@@ -118,3 +118,13 @@ export function jsonFindings(output: string): any[] {
     try { return [JSON.parse(line)]; } catch { return []; }
   });
 }
+
+/**
+ * A test-only-export finding records its non_test_callers count and how the
+ * callers were searched. Any repository search counts (git grep, rg, grep -r,
+ * the Grep tool); the specific command is not graded.
+ */
+export function recordsNonTestCallerSearch(finding: { evidence?: unknown }): boolean {
+  const evidence = JSON.stringify(finding.evidence ?? '');
+  return /non_test_callers/.test(evidence) && /\bgit grep\b|\brg\b|\bgrep\b|\bGrep\b|\bag\b|\back\b/.test(evidence);
+}
diff --git a/test/helpers/touchfiles-data.ts b/test/helpers/touchfiles-data.ts
index 0fd0e3558..8b178f20b 100644
--- a/test/helpers/touchfiles-data.ts
+++ b/test/helpers/touchfiles-data.ts
@@ -137,7 +137,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
      'test/helpers/office-hours-attempt.ts'],
 
   // Office Hours
-  'office-hours-spec-review':     [  'office-hours/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts',
+  'office-hours-spec-review':     [  'office-hours/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
     'bin/gstack-office-hours-review', 'lib/office-hours-review.ts', 'lib/fs-atomic.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts', 'test/helpers/skill-fixture.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/workflow-judge-input.ts'],
   'office-hours-forcing-energy':  [ 'office-hours/**', 'scripts/resolvers/preamble.ts', 'test/fixtures/mode-posture/**', 'test/helpers/llm-judge.ts', 'test/skill-e2e-office-hours.test.ts', 
     'test/helpers/office-hours-attempt.ts' 
@@ -147,26 +147,26 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   ],
 
   // Plan reviews
-  'plan-ceo-review':                  [  'plan-ceo-review/**', 'test/skill-e2e-plan.test.ts',
+  'plan-ceo-review':                  [  'plan-ceo-review/**', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
     'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/tasks-section.ts', 'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
-  'plan-ceo-review-selective':        [  'plan-ceo-review/**', 'test/skill-e2e-plan.test.ts',
+  'plan-ceo-review-selective':        [  'plan-ceo-review/**', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
     'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/tasks-section.ts', 'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
-  'plan-ceo-review-benefits':         [  'plan-ceo-review/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts',
+  'plan-ceo-review-benefits':         [  'plan-ceo-review/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
     'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/tasks-section.ts', 'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
-  'plan-ceo-review-expansion-energy': [  'plan-ceo-review/**', 'scripts/resolvers/preamble.ts', 'test/fixtures/mode-posture/**', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan.test.ts',
+  'plan-ceo-review-expansion-energy': [  'plan-ceo-review/**', 'scripts/resolvers/preamble.ts', 'test/fixtures/mode-posture/**', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
     'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/tasks-section.ts', 'test/helpers/codex-offering-fixture.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
   'plan-eng-review':           [ 
     
     'scripts/resolvers/learnings.ts',
     
-    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'test/skill-e2e-plan.test.ts',
+    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'scripts/resolvers/testing.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts',  'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
   'plan-eng-review-artifact':  [ 
     
     'scripts/resolvers/learnings.ts',
     
-    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'test/skill-e2e-plan.test.ts',
+    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'scripts/resolvers/testing.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts',  'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-attempt.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
   'plan-review-report':        [ 
@@ -175,7 +175,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
      'test/fixtures/plan-review-report-public.json',
     'scripts/resolvers/learnings.ts',
     
-    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts',
+    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/gen-skill-docs.ts', 'test/skill-e2e-plan.test.ts', 'test/helpers/workflow-excerpt.ts',
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'scripts/resolvers/testing.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts',  'test/helpers/codex-offering-fixture.ts', 'test/helpers/llm-judge.ts', 'test/helpers/office-hours-completion.ts', 'test/helpers/workflow-judge-input.ts'],
 
@@ -262,7 +262,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
 
     "test/fixtures/plan-scope-target-aw.json",
 
-     'test/fixtures/auto-decide-saved-ai.json', 'test/fixtures/auto-decide-retry-ai.json','bin/gstack-skill-start', 'bin/gstack-skill-end', 'plan-devex-review/**', 'scripts/resolvers/preamble/generate-completion-status.ts', 'scripts/resolvers/question-tuning.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts', 'test/helpers/claude-pty-runner.ts', 'test/helpers/pty/**',  'test/fixtures/design-ui-boxed-question.json', 'test/helpers/hermetic-skill-runtime.ts',  'test/helpers/pty-trust-dialog.ts',  'test/skill-e2e-plan-devex-plan-mode.test.ts',  'test/fixtures/eng-d2-truncated-border-0bcd.json',   'test/fixtures/eng-d1-clipped-elision-1579.json', 'test/fixtures/eng-d2-planning-prelude-4d.json', 'test/fixtures/ceo-approach-z-call.json', 'test/fixtures/ceo-approach-z-screen.txt', 'test/helpers/plan-scope-selection.ts',  'test/helpers/native-auto-decide.ts',  'test/fixtures/auto-decide-current-declaration-6aef.json',  'test/fixtures/auto-decide-explanatory-mode-043a.json', 'test/fixtures/auto-decide-explanatory-mode-749df.json',  'test/fixtures/auto-decide-structured-77.json', 'test/helpers/auto-decision-state.ts',  'test/fixtures/auto-decide-state-cab3.json', 'bin/gstack-question-log', 'bin/gstack-question-preference',   'test/helpers/fake-plan-seed.ts', 'test/helpers/plan-seed-submission.ts',  'test/fixtures/plan-seed-cli.ts', 'test/fixtures/native-auto-decide-ag.json',  'test/fixtures/eng-seeded-completion-ai.json', 'test/helpers/plan-count-pending-exit.ts',  'test/helpers/pty-screen.ts',   'test/fixtures/pty-screen/**',
+     'test/fixtures/auto-decide-saved-ai.json', 'test/fixtures/auto-decide-retry-ai.json','bin/gstack-skill-start', 'bin/gstack-skill-end', 'plan-devex-review/**', 'scripts/resolvers/preamble/generate-completion-status.ts', 'scripts/resolvers/question-tuning.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts', 'test/helpers/claude-pty-runner.ts', 'test/helpers/pty/**',  'test/fixtures/design-ui-boxed-question.json', 'test/helpers/hermetic-skill-runtime.ts',  'test/helpers/pty-trust-dialog.ts',  'test/skill-e2e-plan-devex-plan-mode.test.ts', 'test/helpers/plan-mode-evidence.ts',  'test/fixtures/eng-d2-truncated-border-0bcd.json',   'test/fixtures/eng-d1-clipped-elision-1579.json', 'test/fixtures/eng-d2-planning-prelude-4d.json', 'test/fixtures/ceo-approach-z-call.json', 'test/fixtures/ceo-approach-z-screen.txt', 'test/helpers/plan-scope-selection.ts',  'test/helpers/native-auto-decide.ts',  'test/fixtures/auto-decide-current-declaration-6aef.json',  'test/fixtures/auto-decide-explanatory-mode-043a.json', 'test/fixtures/auto-decide-explanatory-mode-749df.json',  'test/fixtures/auto-decide-structured-77.json', 'test/helpers/auto-decision-state.ts',  'test/fixtures/auto-decide-state-cab3.json', 'bin/gstack-question-log', 'bin/gstack-question-preference',   'test/helpers/fake-plan-seed.ts', 'test/helpers/plan-seed-submission.ts',  'test/fixtures/plan-seed-cli.ts', 'test/fixtures/native-auto-decide-ag.json',  'test/fixtures/eng-seeded-completion-ai.json', 'test/helpers/plan-count-pending-exit.ts',  'test/helpers/pty-screen.ts',   'test/fixtures/pty-screen/**',
     
      'test/fixtures/design-scope-announcement-ao.json',
     'test/fixtures/design-scope-declaration-ak.json',
@@ -307,7 +307,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   'office-hours-auto-mode':       ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'test/fixtures/auto-decide-recommendation-361c.json',
     
     'test/fixtures/auto-decide-target-361c.json',
-  'test/fixtures/native-auto-decide-ag.json', 'test/helpers/native-auto-decide.ts',  'test/fixtures/auto-decide-current-declaration-6aef.json',  'test/fixtures/auto-decide-explanatory-mode-043a.json', 'test/fixtures/auto-decide-explanatory-mode-749df.json', 'bin/gstack-skill-start', 'bin/gstack-skill-end', 'office-hours/**', 'scripts/resolvers/preamble/generate-completion-status.ts', 'scripts/resolvers/question-tuning.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble.ts', 'test/helpers/claude-pty-runner.ts', 'test/helpers/pty/**',  'test/fixtures/design-ui-boxed-question.json', 'test/helpers/hermetic-skill-runtime.ts',  'test/helpers/pty-trust-dialog.ts',  'test/skill-e2e-office-hours-auto-mode.test.ts',  'test/fixtures/eng-d2-truncated-border-0bcd.json',   'test/fixtures/eng-d1-clipped-elision-1579.json', 'test/fixtures/eng-d2-planning-prelude-4d.json', 'test/fixtures/ceo-approach-z-call.json', 'test/fixtures/ceo-approach-z-screen.txt',
+  'test/fixtures/native-auto-decide-ag.json', 'test/helpers/native-auto-decide.ts',  'test/fixtures/auto-decide-current-declaration-6aef.json',  'test/fixtures/auto-decide-explanatory-mode-043a.json', 'test/fixtures/auto-decide-explanatory-mode-749df.json', 'bin/gstack-skill-start', 'bin/gstack-skill-end', 'office-hours/**', 'scripts/resolvers/preamble/generate-completion-status.ts', 'scripts/resolvers/question-tuning.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble.ts', 'test/helpers/claude-pty-runner.ts', 'test/helpers/pty/**',  'test/fixtures/design-ui-boxed-question.json', 'test/helpers/hermetic-skill-runtime.ts',  'test/helpers/pty-trust-dialog.ts',  'test/skill-e2e-office-hours-auto-mode.test.ts', 'test/helpers/plan-mode-evidence.ts',  'test/fixtures/eng-d2-truncated-border-0bcd.json',   'test/fixtures/eng-d1-clipped-elision-1579.json', 'test/fixtures/eng-d2-planning-prelude-4d.json', 'test/fixtures/ceo-approach-z-call.json', 'test/fixtures/ceo-approach-z-screen.txt',
      'test/fixtures/pty-companion-cli.ts', 'lib/fs-atomic.ts',    'test/helpers/plan-skill-questions.ts', 'test/fixtures/eng-auq-validation-error.json', 'test/fixtures/bash-directory-permission.json', 'test/fixtures/design-tasks-bash-permission.json',  'test/fixtures/read-permission.json',  'test/fixtures/ceo-split-e5-numbered-description-491.json',  'test/helpers/plan-skill-question-events.ts',  'test/helpers/plan-skill-question-hook-scope.ts', 'test/helpers/skill-census.ts',  'test/helpers/auto-decision-state.ts', 'test/helpers/autoplan-artifact-digest.ts', 'test/helpers/autoplan-artifact-permission.ts', 'test/helpers/autoplan-artifact-recorder.ts', 'test/helpers/dx-selected-navigation.ts', 'test/helpers/e2e-gate.ts', 'test/helpers/eng-cache-writer-decision.ts', 'test/helpers/owned-claude-transcript.ts', 'test/helpers/plan-count-artifacts.ts', 'test/helpers/plan-count-file-permission.ts', 'test/helpers/plan-count-fixture.ts', 'test/helpers/plan-count-pending-exit.ts', 'test/helpers/plan-count-pending-question.ts', 'test/helpers/plan-count-transcript.ts', 'test/helpers/plan-floor-review.ts', 'test/helpers/plan-floor-target.ts', 'test/helpers/plan-scope-selection.ts', 'test/helpers/plan-seed-submission.ts', 'test/helpers/pty-screen.ts'],
   'office-hours-phase4-fork':     ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'bin/gstack-skill-start', 'bin/gstack-skill-end', 'office-hours/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completion-status.ts', 'scripts/resolvers/preamble.ts', 'scripts/resolvers/question-tuning.ts', 'test/helpers/llm-judge.ts', 'test/skill-e2e-office-hours-phase4.test.ts' ],
   'llm-judge-recommendation':     ['codex/**', 'test/helpers/llm-judge.ts', 'test/llm-judge-recommendation.test.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'codex/SKILL.md.tmpl', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts'],
@@ -558,23 +558,23 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
 
   // AskUserQuestion format regression (RECOMMENDATION + Completeness: N/10)
   // Fires when either template OR the two preamble resolvers change.
-  'plan-ceo-review-format-mode':      [  'plan-ceo-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts',
+  'plan-ceo-review-format-mode':      [  'plan-ceo-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts', 'test/helpers/plan-format-kind-note.ts',
     'test/helpers/office-hours-attempt.ts',  'scripts/resolvers/tasks-section.ts'
   ],
-  'plan-ceo-review-format-approach':  [  'plan-ceo-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts',
+  'plan-ceo-review-format-approach':  [  'plan-ceo-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts', 'test/helpers/plan-format-kind-note.ts',
     'test/helpers/office-hours-attempt.ts',  'scripts/resolvers/tasks-section.ts'
   ],
   'plan-eng-review-format-coverage':  [ 
     'scripts/resolvers/learnings.ts',
     
-    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts',
+    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts', 'test/helpers/plan-format-kind-note.ts',
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'scripts/resolvers/testing.ts', 'test/helpers/office-hours-attempt.ts',  'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts' 
   ],
   'plan-eng-review-format-kind':      [ 
     'scripts/resolvers/learnings.ts',
     
-    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts',
+    "test/fixtures/plan-scope-recovery-av.json",  'plan-eng-review/**', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/preamble/generate-completeness-section.ts', 'scripts/resolvers/preamble.ts', 'model-overlays/opus-4-7.md', 'test/helpers/llm-judge.ts', 'test/skill-e2e-plan-format.test.ts', 'test/helpers/plan-format-kind-note.ts',
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'scripts/resolvers/testing.ts', 'test/helpers/office-hours-attempt.ts',  'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts' 
   ],
@@ -685,14 +685,14 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   // test/skill-e2e-context-skills.test.ts. These are periodic-tier because
   // each one spawns claude -p and costs ~$0.20-$0.40. Collectively they
   // verify the thing the /checkpoint → /context-save rename was for.
-  'context-save-routing':                  [ 'context-save/**', 'scripts/resolvers/preamble.ts', 'test/skill-e2e-context-skills.test.ts'],
-  'context-save-then-restore-roundtrip':   ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'context-save/**', 'context-restore/**', 'bin/gstack-slug', 'bin/gstack-remote-identity.sh', 'test/skill-e2e-context-skills.test.ts'],
-  'context-restore-fragment-match':        [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts'],
-  'context-restore-empty-state':           [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts'],
-  'context-restore-list-delegates':        [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts'],
-  'context-restore-legacy-compat':         [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts'],
-  'context-save-list-current-branch':      [ 'context-save/**', 'test/skill-e2e-context-skills.test.ts'],
-  'context-save-list-all-branches':        [ 'context-save/**', 'test/skill-e2e-context-skills.test.ts'],
+  'context-save-routing':                  [ 'context-save/**', 'scripts/resolvers/preamble.ts', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-save-then-restore-roundtrip':   ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'context-save/**', 'context-restore/**', 'bin/gstack-slug', 'bin/gstack-remote-identity.sh', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-restore-fragment-match':        [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-restore-empty-state':           [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-restore-list-delegates':        [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-restore-legacy-compat':         [ 'context-restore/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-save-list-current-branch':      [ 'context-save/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
+  'context-save-list-all-branches':        [ 'context-save/**', 'test/skill-e2e-context-skills.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts'],
 
   // Document-release
   'document-release': [ 'document-release/**', 'test/skill-e2e-workflow.test.ts', 'test/fixtures/coverage-audit-fixture.ts', 'test/helpers/coverage-audit.ts', 'test/helpers/office-hours-attempt.ts'],
@@ -793,7 +793,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts",
     'test/helpers/coverage-audit.ts', 'test/helpers/office-hours-attempt.ts',  'scripts/resolvers/testing.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/resolvers/review-dashboard.ts', 'scripts/resolvers/plan-gates.ts', 'scripts/resolvers/spec-review.ts', 'scripts/resolvers/outside-voice-steps.ts', 'scripts/resolvers/review-scope.ts', 'scripts/resolvers/outside-voice.ts' 
   ],
-  'ship-triage': ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'ship/**', 'bin/gstack-repo-mode', 'test/skill-e2e-triage.test.ts',
+  'ship-triage': ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'ship/**', 'bin/gstack-repo-mode', 'test/skill-e2e-triage.test.ts', 'test/helpers/ship-triage-labels.ts',
     'scripts/resolvers/testing.ts'
   ],
   'ship-docsync-completion': ['bin/gstack-state-root.sh', 'lib/state-root.ts', 'ship/**', 'document-release/**', 'test/skill-e2e-ship-docsync.test.ts', 'test/helpers/docsync-*.ts', 'test/helpers/session-runner.ts', 'test/helpers/hermetic-env.ts', 'bin/gstack-skill-start', 'bin/gstack-session-kind', 'scripts/resolvers/sections.ts', 'scripts/resolvers/preamble/generate-ask-user-format.ts', 'scripts/gen-skill-docs.ts', 'scripts/resolvers/testing.ts', 'test/helpers/qa-checkpoint-evidence.ts',    'test/helpers/qa-functional-observer.ts', 'test/helpers/e2e-gate.ts', 'test/helpers/qa-evidence-producer.ts', 'test/helpers/qa-functional-fixture.ts'],
@@ -884,24 +884,24 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
   'scrape-match-path': [
     'scrape/**', 'browse/src/browser-skills.ts', 'browse/src/browser-skill-commands.ts',
     'browser-skills/hackernews-frontpage/**',
-    'test/skill-e2e-skillify.test.ts'
+    'test/skill-e2e-skillify.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts', 'test/helpers/skill-body-narration.ts'
   ],
   'scrape-prototype-path': [
     'scrape/**', 'browse/src/browser-skills.ts', 'browse/src/browser-skill-commands.ts',
-    'test/skill-e2e-skillify.test.ts'
+    'test/skill-e2e-skillify.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts', 'test/helpers/skill-body-narration.ts'
   ],
   'skillify-happy-path': [
     'skillify/**', 'scrape/**', 'browse/src/browser-skill-write.ts',
     'browse/src/browser-skills.ts', 'browse/src/browser-skill-commands.ts',
-    'test/skill-e2e-skillify.test.ts'
+    'test/skill-e2e-skillify.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts', 'test/helpers/skill-body-narration.ts'
   ],
   'skillify-provenance-refusal': [
     'skillify/**', 'browse/src/browser-skill-write.ts',
-    'test/skill-e2e-skillify.test.ts'
+    'test/skill-e2e-skillify.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts', 'test/helpers/skill-body-narration.ts'
   ],
   'skillify-approval-reject': [
     'skillify/**', 'scrape/**', 'browse/src/browser-skill-write.ts',
-    'test/skill-e2e-skillify.test.ts'
+    'test/skill-e2e-skillify.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts', 'test/helpers/skill-body-narration.ts'
   ],
 
   // Skill routing — journey-stage tests (depend on ALL skill descriptions)
@@ -909,7 +909,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -918,7 +918,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -927,7 +927,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -936,7 +936,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -945,7 +945,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -954,7 +954,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -963,7 +963,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -972,7 +972,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -981,7 +981,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -990,7 +990,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
     
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
@@ -1100,7 +1100,7 @@ export const E2E_TOUCHFILES: Record<string, string[]> = {
     
 
     "test/fixtures/plan-scope-recovery-av.json",
-    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts',
+    "test/fixtures/design-scope-checkpoint-at.json", '*/SKILL.md.tmpl', 'SKILL.md.tmpl', 'scripts/gen-skill-docs.ts', 'test/skill-routing-e2e.test.ts', 'bin/gstack-skill-start', 'test/helpers/shipped-skill-routing.ts',
     
 
      "scripts/resolvers/preamble/generate-preamble-bash.ts", "scripts/resolvers/preamble/generate-completion-status.ts"
diff --git a/test/helpers/workflow-boundaries-fixture.ts b/test/helpers/workflow-boundaries-fixture.ts
index b420e1f69..1a03882c2 100644
--- a/test/helpers/workflow-boundaries-fixture.ts
+++ b/test/helpers/workflow-boundaries-fixture.ts
@@ -193,14 +193,21 @@ export function boundaryFailures(fixture: ReturnType<typeof createBoundaryFixtur
     check(!evidence.receipts.includes('VERIFY\n'), 'verifier ran after abort');
     check(/abort|stop|cancel/i.test(finalText), 'actor did not acknowledge abort');
   } else {
-    check(evidence.executions.filter(event => event.tool === 'Edit' && event.allowed).length === 1, 'expected one allowed native Edit');
+    // Outcome, not step count: the correction landed, the final verification
+    // reports the expected status, and no edit follows that verification.
+    const allowed = evidence.executions.filter(event => event.allowed);
+    const lastVerify = allowed.findLastIndex(event => event.tool === 'Bash' && /\bverify\.sh\b/.test(String(event.input.command ?? '')));
+    const statuses = [...evidence.receipts.matchAll(/^VERIFY_STATUS:(\d+)$/gm)].map(match => match[1]);
+    check(allowed.some(event => event.tool === 'Edit'), 'expected an allowed native Edit');
     check(evidence.interactions.some(event => event.disposition === 'continue-investigation' && event.owner === owners[0]), 'continuation was not authorized under the acquired owner');
     check(evidence.source.includes('return 2'), 'requested correction is missing');
-    check(evidence.receipts.match(/^VERIFY$/gm)?.length === 1, 'verifier did not run exactly once');
+    check((evidence.receipts.match(/^VERIFY$/gm)?.length ?? 0) >= 1, 'verifier did not run');
+    check(lastVerify >= 0 && !allowed.slice(lastVerify + 1).some(event => event.tool === 'Edit'), 'an edit followed the final verification');
     if (fixture.id === 'investigate-owned-ending-error') {
-      check(evidence.receipts.includes('VERIFY_STATUS:69\n'), 'ending verifier error was not delivered');
+      // The fixture's ending error says to stop without retrying the verifier.
+      check(statuses.length === 1 && statuses[0] === '69', 'ending verifier error was not delivered once without retry');
       check(/error|unavailable|cannot|could not|unable/i.test(finalText), 'ending error not acknowledged');
-    } else check(evidence.receipts.includes('VERIFY_STATUS:0\n'), 'verification did not succeed');
+    } else check(statuses.at(-1) === '0', 'final verification did not succeed');
   }
   return failures;
 }
diff --git a/test/helpers/workflow-excerpt.ts b/test/helpers/workflow-excerpt.ts
index 4ed2a48e1..838fbbed3 100644
--- a/test/helpers/workflow-excerpt.ts
+++ b/test/helpers/workflow-excerpt.ts
@@ -10,12 +10,22 @@ export const ENG_REVIEW_EXCERPT = {
   endMarker: '## Section self-check (before you finish)',
 } as const;
 
+// The question-procedure heading in the plan-review sections, with or without
+// its older "CRITICAL RULE — " prefix.
+export const ASK_QUESTIONS_HEADING = /^## [^\n]*How to ask questions[^\n]*$/m;
+
+export function markerIndex(text: string, marker: string | RegExp, from: number): number {
+  if (typeof marker === 'string') return text.indexOf(marker, from);
+  const at = text.slice(from).search(marker);
+  return at < 0 ? -1 : from + at;
+}
+
 // Same generated two-line pointer consumed by setup-gbrain-fixture.ts.
 const STOP_POINTER =
   /^> \*\*STOP\.\*\* Before [^\n]*sections\/([a-z0-9-]+\.md)[^\n]*\n> in full\.[^\n]*/gm;
 
 /** Expand on-demand sections where the agent reads them, then take the requested excerpt. */
-export function readWorkflowExcerpt(skillPath: string, startMarker: string, endMarker: string | null): string {
+export function readWorkflowExcerpt(skillPath: string, startMarker: string, endMarker: string | RegExp | null): string {
   const secDir = path.join(ROOT, path.dirname(skillPath), 'sections');
   const content = fs.readFileSync(path.join(ROOT, skillPath), 'utf-8').replace(STOP_POINTER, (_pointer, file: string) => {
     const body = fs.readFileSync(path.join(secDir, file), 'utf-8')
@@ -25,7 +35,7 @@ export function readWorkflowExcerpt(skillPath: string, startMarker: string, endM
   });
   const start = content.indexOf(startMarker);
   if (start < 0) throw new Error(`Start marker not found in ${skillPath}: "${startMarker}"`);
-  const end = endMarker ? content.indexOf(endMarker, start) : content.length;
+  const end = endMarker ? markerIndex(content, endMarker, start) : content.length;
   if (end < 0) throw new Error(`End marker not found in ${skillPath}: "${endMarker}"`);
   return content.slice(start, end);
 }
diff --git a/test/helpers/workflow-judge-cache.ts b/test/helpers/workflow-judge-cache.ts
index 272fdcd2d..3a6ad000f 100644
--- a/test/helpers/workflow-judge-cache.ts
+++ b/test/helpers/workflow-judge-cache.ts
@@ -11,7 +11,7 @@ import { buildEvalInputIdentity, lookupEvalInputCache, sourceDependencyClosure,
 
 type Thresholds = { clarity: number; completeness: number; actionability: number };
 export interface WorkflowCacheOptions {
-  root: string; testName: string; skillPath: string; startMarker: string; endMarker: string | null;
+  root: string; testName: string; skillPath: string; startMarker: string; endMarker: string | RegExp | null;
   judgeContext: string; judgeGoal: string; model?: string; thresholds: Thresholds; prompt: string; attempt: number;
   references?: readonly string[];
   agentCapability?: 'frontier';
@@ -135,3 +135,15 @@ export function prepareWorkflowJudgeCache(opts: WorkflowCacheOptions): {
     },
   };
 }
+
+/**
+ * Pass floors for the browse reference judge panel mean. The stored baseline
+ * is recorded for comparison only: a three-sample mean is too noisy for a
+ * no-dip ratchet, and gating on it silently raised the clarity floor to 4.
+ */
+export const BROWSE_JUDGE_FLOORS = { clarity: 3, completeness: 4, actionability: 4 } as const;
+
+export function browseJudgeFloorsMet(scores: Record<keyof typeof BROWSE_JUDGE_FLOORS, number>): boolean {
+  return (Object.keys(BROWSE_JUDGE_FLOORS) as Array<keyof typeof BROWSE_JUDGE_FLOORS>)
+    .every(dim => scores[dim] >= BROWSE_JUDGE_FLOORS[dim]);
+}
diff --git a/test/helpers/workflow-judge-input.ts b/test/helpers/workflow-judge-input.ts
index 36873687b..f661022f1 100644
--- a/test/helpers/workflow-judge-input.ts
+++ b/test/helpers/workflow-judge-input.ts
@@ -1,6 +1,7 @@
 /** Preserve source-file boundaries when a workflow judge reads carved skills. */
 import * as fs from 'node:fs';
 import * as path from 'node:path';
+import { markerIndex } from './workflow-excerpt';
 
 export interface WorkflowJudgeFile {
   path: string;
@@ -84,7 +85,7 @@ export function readWorkflowJudgeInput(opts: {
   root: string;
   skillPath: string;
   startMarker: string;
-  endMarker: string | null;
+  endMarker: string | RegExp | null;
   references?: readonly string[];
 }): WorkflowJudgeInput {
   const sources = [{
@@ -115,7 +116,7 @@ export function readWorkflowJudgeInput(opts: {
   const union = allSources.map(file => file.content).join('\n');
   const start = union.indexOf(opts.startMarker);
   if (start < 0) throw new Error(`Start marker not found in ${opts.skillPath}: "${opts.startMarker}"`);
-  const end = opts.endMarker === null ? union.length : union.indexOf(opts.endMarker, start);
+  const end = opts.endMarker === null ? union.length : markerIndex(union, opts.endMarker, start);
   if (end < 0) throw new Error(`End marker not found in ${opts.skillPath}: "${opts.endMarker}"`);
 
   const files: WorkflowJudgeFile[] = [];
diff --git a/test/office-hours-completion.test.ts b/test/office-hours-completion.test.ts
index 05c8fc09d..e16685d7c 100644
--- a/test/office-hours-completion.test.ts
+++ b/test/office-hours-completion.test.ts
@@ -413,20 +413,8 @@ describe('office-hours reviewer finding preservation', () => {
       calls++;
       expect(prompt).toContain(JSON.stringify(original));
       expect(prompt).toContain(JSON.stringify(review!.concerns));
-      expect(prompt).toContain('Do not require verbatim wording or judge by issue count alone');
-      expect(prompt).toContain('optional alternative remedies may be omitted');
-      expect(prompt).toContain('only actual defects, never successful mappings');
-      expect(prompt).toContain('Keep reasoning brief');
-      expect(prompt).toContain('Complete EVERY applicable audit even if another already fails');
-      expect(prompt).toContain('a coverage failure cannot skip metrics');
-      expect(prompt).toContain('a metrics failure cannot skip convergence');
-      expect(prompt).toContain('one brief conclusion per applicable audit');
-      expect(prompt).toContain('APPROVED records user approval and is compatible with CONCERNS_RECORDED');
+      // The response schema is the wire contract; the judge's wording is not.
       expect(prompt).toContain('{"reasoning":string,"missing":string[],"unsupported":string[],"complete":boolean}');
-      expect(prompt).toContain('finalize missing and unsupported, and emit complete last');
-      expect(prompt).toContain('at most 150 words total');
-      expect(prompt).toContain('Coverage and the remaining inventory come from the FINAL reviewer verdict only');
-      expect(prompt).toContain('lack of a confirmed prior fix does not invent a missing final-verdict finding');
       return JSON.stringify(accepted);
     });
     expect(calls).toBe(1);
diff --git a/test/overlay-measurement.test.ts b/test/overlay-measurement.test.ts
index 7ecfedd8b..2f18da844 100644
--- a/test/overlay-measurement.test.ts
+++ b/test/overlay-measurement.test.ts
@@ -6,7 +6,7 @@ import type { AgentSdkResult, QueryProvider } from './helpers/agent-sdk-runner';
 import { firstAssistantMessageToolCount, reportedThinkingTokens, assessComparison, trialArtifactStem } from './helpers/overlay-measurement';
 import { runOverlayTrial, awaitOverlayWorkers, assessOverlayArms, captureOverlayQueryAttempts, type OverlayTrialOutcome } from './helpers/overlay-attempt';
 import { setupLiteralWorkspace, correctLiteralTargets, snapshotWorkspace, assertReadOnlyWorkspace } from './helpers/overlay-workspace';
-import { fanoutPass, higherIsBetter20Pct, lowerIsBetter20Pct, OVERLAY_FIXTURES, type OverlayFixture } from './fixtures/overlay-nudges';
+import { fanoutPass, higherIsBetter20Pct, lowerIsBetter20Pct, lowerIsBetter20PctOrZeroBaseline, OVERLAY_FIXTURES, type OverlayFixture } from './fixtures/overlay-nudges';
 
 function result(overrides: Partial<AgentSdkResult> = {}): AgentSdkResult {
   return {
@@ -64,16 +64,45 @@ describe('SDK overlay measurements', () => {
 });
 
 describe('correctness versus efficacy', () => {
-  test.each(['claude-opus-4-7', 'claude-sonnet-4-6'])('dedicated-tool behavior gates zero ON Bash calls for %s', (model) => {
+  describe.each(['claude-opus-4-7', 'claude-sonnet-4-6'])('dedicated-tool gate for %s', (model) => {
     const f = { ...OVERLAY_FIXTURES.find(f => f.metricName === 'bash_tool_calls' && f.model === model)!, trials: 3 };
-    const samples = (metric: number) => Array.from({ length: 3 }, (): OverlayTrialOutcome =>
-      ({ passed: true, taskCorrect: f.taskCorrect!(metric), metric, exitReason: 'success' }));
-    expect(assessOverlayArms(f, samples(0), samples(0))).toMatchObject({
-      passed: true, comparison: { status: 'baseline_saturated', criterionMet: false },
+    // A trial whose final JSON fails the fixture's verify is recorded as a failed measurement.
+    const correct = (metric: number): OverlayTrialOutcome => ({ passed: true, taskCorrect: true, metric, exitReason: 'success' });
+    const incorrect = (metric: number): OverlayTrialOutcome => ({ passed: false, taskCorrect: false, metric, exitReason: 'validation_failed' });
+    const arm = (...trials: OverlayTrialOutcome[]) => trials;
+    test('fixture gates on the comparison and on output correctness, not on zero ON Bash', () => {
+      expect(f.taskCorrect).toBeUndefined();
+      expect(f.verify).toBeFunction();
+      expect(f.gate).toBe(lowerIsBetter20PctOrZeroBaseline);
     });
-    // Even a >20% improvement cannot excuse breaking the exact ON contract.
-    expect(assessOverlayArms(f, samples(1), samples(5))).toMatchObject({
-      passed: false, correctnessPassed: false, comparison: { status: 'improved', criterionMet: true },
+    test('correct output with a 20% reduction passes even when ON still uses Bash', () => {
+      expect(assessOverlayArms(f, arm(correct(1), correct(1), correct(2)), arm(correct(5), correct(5), correct(5)))).toMatchObject({
+        measurementsValid: true, correctnessPassed: true, gatePassed: true, passed: true, comparison: { status: 'improved', criterionMet: true },
+      });
+    });
+    test('correct output with worse Bash usage fails', () => {
+      expect(assessOverlayArms(f, arm(correct(6), correct(6), correct(6)), arm(correct(5), correct(5), correct(5)))).toMatchObject({
+        measurementsValid: true, correctnessPassed: true, gatePassed: false, passed: false, comparison: { status: 'regressed' },
+      });
+      expect(assessOverlayArms(f, arm(correct(5), correct(4), correct(5)), arm(correct(5), correct(5), correct(5)))).toMatchObject({
+        gatePassed: false, passed: false, comparison: { status: 'no_measured_improvement' },
+      });
+    });
+    test('a 20% reduction with incorrect output fails', () => {
+      expect(assessOverlayArms(f, arm(correct(0), incorrect(0), correct(0)), arm(correct(5), correct(5), correct(5)))).toMatchObject({
+        measurementsValid: false, passed: false,
+      });
+    });
+    test('a zero-Bash baseline passes only when every ON trial is also zero', () => {
+      expect(assessOverlayArms(f, arm(correct(0), correct(0), correct(0)), arm(correct(0), correct(0), correct(0)))).toMatchObject({
+        gatePassed: true, passed: true, comparison: { status: 'baseline_saturated' },
+      });
+      expect(assessOverlayArms(f, arm(correct(0), correct(1), correct(0)), arm(correct(0), correct(0), correct(0)))).toMatchObject({
+        gatePassed: false, passed: false,
+      });
+      // OFF median 0 with an occasional Bash call is still a zero baseline.
+      expect(assessOverlayArms(f, arm(correct(0), correct(0), correct(0)), arm(correct(0), correct(2), correct(0)))).toMatchObject({ gatePassed: true, passed: true });
+      expect(assessOverlayArms(f, arm(correct(1), correct(0), correct(0)), arm(correct(0), correct(2), correct(0)))).toMatchObject({ gatePassed: false, passed: false });
     });
   });
   test.each([
diff --git a/test/overlay-recording-order.test.ts b/test/overlay-recording-order.test.ts
index 83ff28e47..09880b0be 100644
--- a/test/overlay-recording-order.test.ts
+++ b/test/overlay-recording-order.test.ts
@@ -86,7 +86,7 @@ test('versioned aggregate keeps false efficacy and cannot overwrite earlier atte
     const original = fs.readFileSync(file, 'utf8');
     expect(JSON.parse(original)).toMatchObject({ ...verdict, contract: OVERLAY_CONTRACT });
     expect(entries).toHaveLength(1);
-    expect(entries[0]).toMatchObject({ name: 'owned-contract-contract-v2-aggregate', passed: true, cost_usd: 0, duration_ms: 0 });
+    expect(entries[0]).toMatchObject({ name: `owned-contract-contract-v${OVERLAY_CONTRACT.version}-aggregate`, passed: true, cost_usd: 0, duration_ms: 0 });
     expect(() => recorder(deps, 1)({ passed: false })).toThrow('EEXIST');
     expect(fs.readFileSync(file, 'utf8')).toBe(original);
     recorder(deps, 2)({ passed: false, errors: ['task incomplete'] });
diff --git a/test/plan-design-sdk-fixture.test.ts b/test/plan-design-sdk-fixture.test.ts
index f4ae51f99..c42e04e72 100644
--- a/test/plan-design-sdk-fixture.test.ts
+++ b/test/plan-design-sdk-fixture.test.ts
@@ -87,37 +87,18 @@ async function exercise(mode: 'success' | 'max-turns' | 'first-timeout' | 'secon
       expect(opts.publicStreamDiagnostics).toBe(true);
       expect(opts.signal).toBeInstanceOf(AbortSignal); sdkSignal = opts.signal;
       expect(opts.signal.aborted).toBe(false);
-      // Bind the complete actual compact-delivery prompt, not selected snippets.
-      expect(new Bun.CryptoHasher('sha256').update(opts.prompt).digest('hex'))
-        .toBe('021c9f2b0522fc55bc265a976356678a048f3c17d27b63669674c805a485b28e');
       expect(opts.testName).toBe(id); expect(opts.maxTurns).toBe(15); expect(opts.timeout).toBe(CAPTURE_MS);
       for (const key of ['model', 'tools', 'allowedTools', 'appendSystemPrompt', 'env']) expect(opts).not.toHaveProperty(key);
-      expect(opts.prompt).toContain('Review the plan in ./plan.md');
-      expect(opts.prompt).toContain('Review all 7 design passes');
-      expect(opts.prompt).toContain('0-10 and explain what would make it a 10');
-      expect(opts.prompt).toContain('preserve the unresolved-decisions pass');
-      expect(opts.prompt).toContain('interaction state table, empty states, responsive behavior');
-      expect(opts.prompt).toContain('full required review report');
-      expect(opts.prompt).toContain('Write before publishing a completed walkthrough');
-      expect(opts.prompt).toContain('Read plan.md back to verify the saved changes');
-      expect(opts.prompt).toContain('Then return a brief, concrete summary');
-      expect(opts.prompt).toContain('natively Read plan-design-review/SKILL.md');
-      expect(opts.prompt).toContain('plan-design-review/sections/review-sections.md (the one lazy section this review requires)');
-      expect(opts.prompt).toContain('including the report, to about 10,000 characters; this is a drafting target, not a check, so do not count characters or trim after saving');
-      expect(opts.prompt).toContain('in at most ten lines');
-      expect(opts.prompt).toContain('execute every required pass and lazy-section Read');
-      expect(opts.prompt).toContain('Retain all required report fields, design decisions, diagrams, ratings, and explanations');
-      expect(opts.prompt).toContain('use the canonical tables and decision IDs');
-      expect(opts.prompt).toContain('Specify each design requirement once');
-      expect(opts.prompt).toContain('instead of repeating that specification');
-      expect(opts.prompt).toContain('concise score rationales and 10/10 explanations');
+      // The harness prompt's properties, in order; its wording is free.
+      expect(opts.prompt).toContain('plan-design-review/SKILL.md');
+      expect(opts.prompt).toContain('plan-design-review/sections/review-sections.md');
+      expect(opts.prompt).toContain('./plan.md');
       const ordered = ['Read every lazy section', 'Review all 7 design passes',
-        'EDIT plan.md', 'Keep the saved review compact',
-        'Persist that complete plan and review with Write', 'Read plan.md back',
+        'EDIT plan.md', 'Persist that complete plan and review with Write', 'Read plan.md back',
         'Then return a brief, concrete summary'].map(text => opts.prompt.indexOf(text));
       expect(ordered.every(index => index >= 0)).toBe(true);
       expect(ordered).toEqual([...ordered].sort((a, b) => a - b));
-      expect(opts.prompt).toContain('Do NOT try to browse any URLs');
+      expect(opts.prompt).toMatch(/do not try to browse any URLs/i);
       inputVerified = true; launched = true;
       if (mode === 'attempt-deadline') return await new Promise((_resolve, reject) => {
         opts.signal.addEventListener('abort', () => reject(opts.signal.reason), { once: true });
diff --git a/test/plan-format-kind-note.test.ts b/test/plan-format-kind-note.test.ts
new file mode 100644
index 000000000..f6829da9a
--- /dev/null
+++ b/test/plan-format-kind-note.test.ts
@@ -0,0 +1,12 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { KIND_NOTE_RE } from './helpers/plan-format-kind-note';
+
+const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/plan-format-kind-notes.json'), 'utf8')) as
+  { known_good: string[]; known_bad: string[] };
+
+describe('plan-format kind note', () => {
+  test.each(stored.known_good)('accepts %s', line => expect(line).toMatch(KIND_NOTE_RE));
+  test.each(stored.known_bad)('rejects %s', line => expect(line).not.toMatch(KIND_NOTE_RE));
+});
diff --git a/test/plan-mode-evidence.test.ts b/test/plan-mode-evidence.test.ts
index b0fa0e67b..3a6f14303 100644
--- a/test/plan-mode-evidence.test.ts
+++ b/test/plan-mode-evidence.test.ts
@@ -63,7 +63,9 @@ await import(${JSON.stringify(path.join(ROOT, PAID_FILE))});
         expect(opts.timeoutMs).toBe(CAPTURE_MS);
         const keys = ['skillName', 'inPlanMode', 'timeoutMs'];
         if (opts.initialPlanContent) {
-          keys.push('initialPlanContent');
+          keys.push('initialPlanContent', 'trackTokens');
+          expect(opts.trackTokens.length).toBeGreaterThan(0);
+          for (const token of opts.trackTokens) expect(opts.initialPlanContent).toContain(token);
           if (skill === 'eng') {
             keys.push('extraArgs');
             expect(opts.extraArgs).toEqual(['--disallowedTools', 'AskUserQuestion']);
@@ -105,7 +107,7 @@ await import(${JSON.stringify(path.join(ROOT, PAID_FILE))});
       if (scenario === 'bad-report') expect(output).toContain('GSTACK REVIEW REPORT contract violation');
       if (scenario === 'missing-auto-select') expect(output).toContain('scopeGateAutoSelectObserved');
       if (scenario === 'scope-question') expect(output).toContain('scopeGateQuestionObserved');
-      if (scenario === 'missing-decisions') expect(output).toContain('plan_ready without ## Decisions section');
+      if (scenario === 'missing-decisions') expect(output).toContain('without a question');
     } finally { fs.rmSync(dir, { recursive: true, force: true }); }
   });
 }
diff --git a/test/plan-mode-targeting.test.ts b/test/plan-mode-targeting.test.ts
new file mode 100644
index 000000000..bb1dfcbfb
--- /dev/null
+++ b/test/plan-mode-targeting.test.ts
@@ -0,0 +1,41 @@
+import { afterAll, 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 { assertNoPlanFileDecisions, seededPlanTargeted } from './helpers/plan-mode-evidence';
+import { planFileHasDecisionsSection } from './helpers/claude-pty-runner';
+import type { PlanSkillObservation } from './helpers/claude-pty-runner';
+
+const TOKENS = ['ShardManager', 'ResultMerger'];
+const base = { outcome: 'asked', summary: '', evidence: '', elapsedMs: 1 } as unknown as PlanSkillObservation;
+const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'plan-mode-targeting-'));
+afterAll(() => fs.rmSync(dir, { recursive: true, force: true }));
+const plan = (name: string, body: string) => { const file = path.join(dir, name); fs.writeFileSync(file, body); return file; };
+
+describe('seeded plan targeting', () => {
+  test.each([
+    ['exact announcement observed', { scopeGateAutoSelectObserved: true }],
+    ['paraphrased announcement with seed content in the review', { scopeGateAutoSelectObserved: false, tokensObserved: { ShardManager: true, ResultMerger: false } }],
+  ])('known-good: %s', (_name, fields) => {
+    expect(seededPlanTargeted({ ...base, ...fields } as PlanSkillObservation, TOKENS)).toBe(true);
+  });
+  test.each([
+    ['no announcement and no seed content', { scopeGateAutoSelectObserved: false, tokensObserved: { ShardManager: false, ResultMerger: false } }],
+    ['an untracked token cannot stand in for the seed', { scopeGateAutoSelectObserved: false, tokensObserved: { 'draft-plan': true } }],
+  ])('known-bad: %s', (_name, fields) => {
+    expect(seededPlanTargeted({ ...base, ...fields } as PlanSkillObservation, TOKENS)).toBe(false);
+  });
+});
+
+describe('plan-file decision substitute', () => {
+  test('a plan_ready plan file with ## Decisions fails', () => {
+    const planFile = plan('decisions.md', '# Plan\n\n## Decisions to confirm\n- A or B?\n\n## GSTACK REVIEW REPORT\n');
+    expect(() => assertNoPlanFileDecisions({ ...base, outcome: 'plan_ready', planFile } as PlanSkillObservation, planFileHasDecisionsSection))
+      .toThrow('not written to the plan as a substitute');
+  });
+  test('a plan_ready plan file without ## Decisions passes, and asked runs are not checked', () => {
+    const planFile = plan('clean.md', '# Plan\n\n## Approach\nShip it.\n\n## GSTACK REVIEW REPORT\n');
+    expect(() => assertNoPlanFileDecisions({ ...base, outcome: 'plan_ready', planFile } as PlanSkillObservation, planFileHasDecisionsSection)).not.toThrow();
+    expect(() => assertNoPlanFileDecisions({ ...base, outcome: 'asked' } as PlanSkillObservation, planFileHasDecisionsSection)).not.toThrow();
+  });
+});
diff --git a/test/section-capture-native-tools.test.ts b/test/section-capture-native-tools.test.ts
index 8f3e48c0e..1ec7b464b 100644
--- a/test/section-capture-native-tools.test.ts
+++ b/test/section-capture-native-tools.test.ts
@@ -303,9 +303,9 @@ console.log(JSON.stringify({plain:plain.exitReason,literal:literal.exitReason,se
     expect(added).toBeGreaterThan(-1); expect(args(1)[added + 1]).toBe(literal);
     expect(args(1).filter((_, i) => i !== added && i !== added + 1)).toEqual(args(0));
     const sectionArgs = args(2);
-    const instruction = launches[2].prompt.split('\n').find((line: string) => line.includes('with the Read tool BEFORE'));
+    const instruction = launches[2].prompt.split('\n').find((line: string) => /with the Read tool before/i.test(line));
     expect(sectionArgs).not.toContain('--append-system-prompt');
-    expect(instruction).toContain('you MUST actually Read that sections/ file with the Read tool BEFORE doing the work it covers');
+    expect(instruction).toMatch(/read that sections\/ file with the Read tool before doing the work it covers/i);
     expect(instruction).not.toContain('actual.md'); expect(instruction).not.toContain(dir);
     expect(sectionArgs.slice(sectionArgs.indexOf('--allowed-tools') + 1, sectionArgs.indexOf('--allowed-tools') + 8))
       .toEqual(['Read', 'Grep', 'Glob', 'Write', 'Edit', 'Agent', 'Bash']);
@@ -464,21 +464,16 @@ console.log(JSON.stringify(observed));`);
         expect(request).not.toHaveProperty('appendSystemPrompt');
         const skill=['fixture','ship','office-hours'][Math.floor((i-4)/2)]!;
         const skillPath=path.join(dir,skill,'SKILL.md');
-        const expected=`You are running an automated skill-execution test. No human is present, so AskUserQuestion is unavailable. The ONLY skill file you may read is this absolute path: ${skillPath}. Do NOT Glob/find/search for any other SKILL.md anywhere — especially nothing under ~/.claude or /Users.
-
-Read ${skillPath} and EXECUTE its workflow for this scenario:
-
-Complete the supplied scenario.
-
-Rules for this run:
-- Skip system-audit, environment-setup, telemetry, and unrelated codebase exploration. Read the supplied plan's referenced fixture files when its review requires them.
-- At any decision point that would call AskUserQuestion, silently pick the skill's recommended option and continue. Do NOT stop to ask.
-- This skill's body has been carved into on-demand sections/. When the skill gives a STOP-Read directive (for example "Read \`.../sections/<file>\` and execute it in full"), you MUST actually Read that sections/ file with the Read tool BEFORE doing the work it covers. Do not work from memory.
-- Resolve installed-root paths for section and companion Markdown files under ${dir}, where this fixture's skill package is copied.
-- Do NOT run git, gh, commit, push, or any mutating command.
-- When the workflow is complete, write the skill's final output (the full review report / ship plan, including any required report table) to ${path.join(dir,'REPORT.md')}.
-- After all required writes are complete, return a brief completion message and STOP. Do not reproduce the full report in the final response.`;
-        expect(request.prompt).toBe(expected);
+        // Harness properties, not its wording: the one allowed skill path, the
+        // scenario, AskUserQuestion fallback, section Read rule, no mutation,
+        // and the report destination.
+        expect(request.prompt.match(/\/[^\s]*SKILL\.md/g)?.every((file: string)=>file===skillPath)).toBe(true);
+        expect(request.prompt).toContain(`Read ${skillPath} and EXECUTE its workflow`);
+        expect(request.prompt).toContain('Complete the supplied scenario.');
+        expect(request.prompt).toMatch(/AskUserQuestion is unavailable/i);
+        expect(request.prompt).toMatch(/read that sections\/ file with the Read tool before/i);
+        expect(request.prompt).toMatch(/do not run git, gh, commit, push, or any mutating command/i);
+        expect(request.prompt).toContain(path.join(dir,'REPORT.md'));
       }
     }
   }finally{if(child.exitCode===null)child.kill();await child.exited;fs.rmSync(dir,{recursive:true,force:true});}
diff --git a/test/session-runner-tools.test.ts b/test/session-runner-tools.test.ts
index 580899ee5..395e93d97 100644
--- a/test/session-runner-tools.test.ts
+++ b/test/session-runner-tools.test.ts
@@ -431,7 +431,7 @@ describe.skipIf(process.platform === 'win32')('session-runner explicit tool avai
       expect(child.prompt).toContain('Preserve every finding and original requirement, required decision fields and comparisons, exact approvals and verification; every diagram must retain its specified format.');
       expect(child.prompt).toContain('unless needed to specify an accepted change');
       expect(child.prompt).toContain('Complete every required artifact and verification before returning');
-      expect(child.prompt).toContain('MUST actually Read that sections/ file with the Read tool BEFORE doing the work it covers');
+      expect(child.prompt).toMatch(/read that sections\/ file with the Read tool before doing the work it covers/i);
       expect(child.prompt).toContain('report outside coverage as disabled');
       expect(child.prompt).toContain('After all required writes are complete');
       expect(child.prompt).toContain('with Write/Edit at the workflow checkpoints below');
@@ -527,7 +527,7 @@ describe.skipIf(process.platform === 'win32')('session-runner explicit tool avai
       expect(child.prompt).toContain('do not hide it or claim approval when no offered alternative meets these constraints');
       expect(child.prompt).not.toContain("silently pick the skill's recommended option");
       expect(child.prompt).toContain('Every required section, finding, approval and output still has to be completed');
-      expect(child.prompt).toContain('MUST actually Read that sections/ file with the Read tool BEFORE doing the work it covers');
+      expect(child.prompt).toMatch(/read that sections\/ file with the Read tool before doing the work it covers/i);
       expect(child.prompt).toContain('After all required writes are complete');
       expect(flagValue(child.args, '--tools')).toBe('Read,Grep,Glob,Write,Edit,Agent,Bash');
       expect(fs.readFileSync(path.join(dir, 'PLAN.md'), 'utf8')).toBe(fixtures['PLAN.md']);
@@ -548,20 +548,19 @@ describe.skipIf(process.platform === 'win32')('session-runner explicit tool avai
         planDir: dir, skillName: 'ship', scenario, testName: 'ship-report-writing', timeout: 5_000,
       });
       const skillPath = path.join(dir, 'ship', 'SKILL.md');
-      expect(observed().prompt).toBe(`You are running an automated skill-execution test. No human is present, so AskUserQuestion is unavailable. The ONLY skill file you may read is this absolute path: ${skillPath}. Do NOT Glob/find/search for any other SKILL.md anywhere — especially nothing under ~/.claude or /Users.
-
-Read ${skillPath} and EXECUTE its workflow for this scenario:
-
-${scenario}
-
-Rules for this run:
-- Skip system-audit, environment-setup, telemetry, and unrelated codebase exploration. Read the supplied plan's referenced fixture files when its review requires them.
-- At any decision point that would call AskUserQuestion, silently pick the skill's recommended option and continue. Do NOT stop to ask.
-- This skill's body has been carved into on-demand sections/. When the skill gives a STOP-Read directive (for example "Read \`.../sections/<file>\` and execute it in full"), you MUST actually Read that sections/ file with the Read tool BEFORE doing the work it covers. Do not work from memory.
-- Resolve installed-root paths for section and companion Markdown files under ${dir}, where this fixture's skill package is copied.
-- Do NOT run git, gh, commit, push, or any mutating command.
-- When the workflow is complete, write the skill's final output (the full review report / ship plan, including any required report table) to ${path.join(dir, 'REPORT.md')}.
-- After all required writes are complete, return a brief completion message and STOP. Do not reproduce the full report in the final response.`);
+      const prompt = observed().prompt;
+      // CEO-only writing guidance stays out of another skill's capture prompt.
+      for (const ceoOnly of ['all 11 sections', 'Preserve original requirements and accepted plan amendments',
+        'Cross-reference saved IDs', 'report outside coverage as disabled', 'with Write/Edit at the workflow checkpoints']) {
+        expect(prompt).not.toContain(ceoOnly);
+      }
+      // The functional harness rules, checked by meaning rather than bytes.
+      expect(prompt).toContain(scenario);
+      expect(prompt.match(/\/[^\s]*SKILL\.md/g)?.every(file => file === skillPath)).toBe(true);
+      expect(prompt).toMatch(/AskUserQuestion is unavailable/i);
+      expect(prompt).toMatch(/read that sections\/ file with the Read tool before/i);
+      expect(prompt).toMatch(/do not run git, gh, commit, push, or any mutating command/i);
+      expect(prompt).toContain(path.join(dir, 'REPORT.md'));
     });
   });
 
diff --git a/test/shared-libs-fixture.test.ts b/test/shared-libs-fixture.test.ts
index 7affb5de8..2bbbff13e 100644
--- a/test/shared-libs-fixture.test.ts
+++ b/test/shared-libs-fixture.test.ts
@@ -7,7 +7,7 @@ import { execFileSync, spawnSync } from 'node:child_process';
 import {
   createSharedInteractiveToolHandler, createSharedLibsFixture, fixtureGit, fixtureWrite, installSourceShims,
   installHostileGitConfig, isGuardedGitRequest, reviewLifecycleInstructions, standaloneInstructions, SHARED_LIBS_ROOT, readRequests, seedOpportunitySources, sharedReadOnlyViolations, shellQuote, snapshotFixture, type SharedLibsFixture,
-  SharedCaptureAccumulator, type SharedCaptureAttempt, isInternalClaudeGitRequest, SHARED_LIBS_OLDER_OPEN_PRS, incompleteFirstFileView,
+  SharedCaptureAccumulator, type SharedCaptureAttempt, isInternalClaudeGitRequest, SHARED_LIBS_OLDER_OPEN_PRS, incompleteFirstFileView, prCoverageRequestViolations,
 } from './helpers/shared-libs-eval-fixture';
 import { EvalCollector, type EvalTestEntry } from './helpers/eval-store';
 import { collectorOutcomeCounts } from '../scripts/test-paid-shards';
@@ -1075,7 +1075,12 @@ describe('shared-code PR coverage world', () => {
 
   test('the maximum authorized open-metadata scan still leaves older open PRs unchecked', () => {
     const periodic = fs.readFileSync(path.join(import.meta.dir, 'skill-e2e-shared-libs-periodic.test.ts'), 'utf8');
-    expect(periodic).toContain('expect(openPages.length).toBeLessThanOrEqual(5);');
+    expect(periodic).toContain('prCoverageRequestViolations(');
+    const sixth = Array.from({ length: 6 }, (_, i) => `/repos/fixture/shared-libs/pulls?state=open&per_page=100&page=${i + 1}`);
+    expect(prCoverageRequestViolations([...sixth.slice(0, 5), '/repos/fixture/shared-libs/pulls/42/files?page=1',
+      '/repos/fixture/shared-libs/pulls/42/files?page=2'])).toEqual([]);
+    expect(prCoverageRequestViolations([...sixth, '/repos/fixture/shared-libs/pulls/42/files?page=1',
+      '/repos/fixture/shared-libs/pulls/42/files?page=2'])).toEqual(['6 open-PR metadata pages exceed the budget of 5']);
     const f = createSharedLibsFixture('pr-world-budget');
     cleanup.push(f.root);
     seedOpportunitySources(f);
diff --git a/test/shared-libs-pr-coverage-budget.test.ts b/test/shared-libs-pr-coverage-budget.test.ts
new file mode 100644
index 000000000..b4a0cea1a
--- /dev/null
+++ b/test/shared-libs-pr-coverage-budget.test.ts
@@ -0,0 +1,21 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { prCoverageRequestViolations } from './helpers/shared-libs-eval-fixture';
+
+const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/shared-libs/pr-coverage-endpoints.json'), 'utf8')) as
+  { known_good: Record<string, string[]>; known_bad: Record<string, string[]> };
+
+describe('shared-libs PR coverage request budget', () => {
+  test.each(Object.entries(stored.known_good))('passes %s', (_name, endpoints) => {
+    expect(prCoverageRequestViolations(endpoints)).toEqual([]);
+  });
+  test.each(Object.entries(stored.known_bad))('fails %s', (_name, endpoints) => {
+    expect(prCoverageRequestViolations(endpoints).length).toBeGreaterThan(0);
+  });
+  test('file-list pages beyond the stated budget fail', () => {
+    const pages = Array.from({ length: 52 }, (_, i) => `/repos/fixture/shared-libs/pulls/${i < 2 ? 42 : 100 + i}/files?per_page=100&page=${i < 2 ? i + 1 : 1}`);
+    expect(prCoverageRequestViolations(pages)).toEqual(['52 file-list pages exceed the budget of 50 plus PR 7']);
+    expect(prCoverageRequestViolations(pages.slice(0, 51))).toEqual([]);
+  });
+});
diff --git a/test/shared-libs-revalidation-prompt.test.ts b/test/shared-libs-revalidation-prompt.test.ts
index c2ad0f2d9..2ce0b016c 100644
--- a/test/shared-libs-revalidation-prompt.test.ts
+++ b/test/shared-libs-revalidation-prompt.test.ts
@@ -71,11 +71,13 @@ function pathCaptureAdapter(capture: (...args: any[]) => Promise<any>) {
 }
 
 describe('bounded shared-code revalidation prompt', () => {
+  // The harness prompt's wording is not a contract: the paid shared-libs E2Es
+  // measure whether it works. These checks keep its machine-shaped parts:
+  // exact receipt commands, interpolated paths and budgets, branch selection,
+  // prefix/suffix identity, ordering and the absence of a supplied token.
   test('the actor scope replaces the captured R59 exploratory stage instead of adding another prerequisite', () => {
     const prompt = reviewPrompt(f, instructions, input, { actorCommand: 'cat /fixture/stage-output.json' });
-    expect(prompt).toContain('replaces the entire Step 4.7 QA and Step 4.8 native adversarial stages');
-    expect(prompt).toContain("Step 4's early QA selection/method-loading prerequisites");
-    expect(prompt).toContain("Step 5.8's QA report requirement");
+    expect(prompt).toContain('cat /fixture/stage-output.json');
     const excluded = prompt.slice(prompt.indexOf('Do not perform QA scope/method asset loads'), prompt.indexOf('Existing tests'));
     for (const packet of stageScope.outside_component) {
       expect(packet.result.tool_use_id).toBe(packet.call.id);
@@ -85,26 +87,15 @@ describe('bounded shared-code revalidation prompt', () => {
     }
     expect(stageScope.post_fix_verification.call.input.command).toContain('bun test test/retry-after.test.ts');
     expect(stageScope.post_fix_verification.result.content).toContain('worker===lib true route===lib true');
-    expect(prompt).toContain('Existing tests and caller/import checks needed to verify your source fixes still run');
-    expect(prompt).toContain('do not restart exploratory QA or require QA artifacts');
-    for (const retained of ['core/checklist', 'source/identity/snapshot checks', 'Fix-First decisions', 'approved source edits',
-      're-review with a new REVIEW_START', 'zero-edit convergence', 'final persistence', 'Missing, failed, stale or wrong-state results require noncompletion',
-      'no actual native coverage credit', 'Separate genuine QA/native evaluations remain required']) expect(prompt).toContain(retained);
     for (const other of [reviewPrompt(f, instructions, input), reviewRevalidationPrompt(f, instructions, input),
       reviewRevalidationPrompt(f, instructions, input, { input: '/fixture/resumed.json', checkCommand: 'check-prerequisites' })]) {
       expect(other).not.toContain('Do not perform QA scope/method asset loads');
     }
   });
 
-  test('edit-capable replay declares fresh actor invocations instead of refreshing settled input', () => {
+  test('edit-capable replay declares its actor instead of the fixed revalidation coverage', () => {
     const prompt = reviewPrompt(f, instructions, input, { actorCommand: 'cat /fixture/stage-output.json' });
-    expect(prompt).toContain('explicitly declared SYNTHETIC prerequisite actor');
-    for (const rule of ['each review pass', 'NEW synthetic result', 'exact current state and tool-use ID',
-      'All prior receipts are preserved', 'Source-changing cycles invalidate earlier results',
-      'invoke the actor again on the new zero-edit pass', "Never refresh an old receipt's hashes",
-      'Missing, failed, stale or wrong-state results require noncompletion', 'cannot complete core/checklist review',
-      'only a current settled:true actor result from the final pass supplies the replaced Step 4.7 QA and Step 4.8 native adversarial prerequisites',
-      'a reporting label, not a missing stage', 'no actual native coverage credit']) expect(prompt).toContain(rule);
+    expect(prompt).toContain('cat /fixture/stage-output.json');
     expect(prompt).not.toContain('Required reviewer coverage for this scoped replay');
     expect(prompt).not.toContain('Do not edit target source');
   });
@@ -113,19 +104,15 @@ describe('bounded shared-code revalidation prompt', () => {
     const resumed = { input: '/isolated/synthetic-prerequisites.json', checkCommand: 'fixture-prerequisite-check' };
     const original = reviewRevalidationPrompt(f, instructions, input);
     const prompt = reviewRevalidationPrompt(f, instructions, input, resumed);
-    expect(original).toContain('Required reviewer coverage for this scoped replay is the core/checklist review plus the supplied completed maintainability result.');
+    expect(original).toContain('Required reviewer coverage for this scoped replay');
     expect(prompt).not.toContain('Required reviewer coverage for this scoped replay');
-    expect(prompt).toContain('SYNTHETIC settled Step 4.7 QA and Step 4.8 native adversarial');
-    for (const requirement of ['not evidence that this model executed those stages', 'never actual native coverage credit',
-      'Missing, failed, blocked, malformed or stale prerequisites require noncompletion', 'unchanged COMPLETED and CONVERGED rules',
-      'Any source, branch, base, index or configuration change invalidates', 'do not regenerate them',
-      'A finding that requires edits blocks this bounded replay', resumed.input, resumed.checkCommand,
-      "replace Step 4's early QA selection and method-loading prerequisites", 'read no QA scope or method assets']) expect(prompt).toContain(requirement);
-    expect(original).not.toContain('read no QA scope or method assets');
+    expect(prompt).toContain(resumed.input);
+    expect(prompt).toContain(resumed.checkCommand);
     expect(prompt.slice(prompt.indexOf('Revalidation fixture execution contract:')))
       .toBe(original.slice(original.indexOf('Revalidation fixture execution contract:')));
+    // The review completion rule names both prerequisite stages by step number.
     const production = fs.readFileSync(path.join(SHARED_LIBS_ROOT, 'review/SKILL.md.tmpl'), 'utf8');
-    expect(production).toContain('Step 4.8 adversarial pass finish, and every required Step 4.7 probe passes.');
+    expect(production.split('\n').some(line => /Step 4\.8\b/.test(line) && /Step 4\.7\b/.test(line))).toBe(true);
   });
 
   test('adds execution guidance after the complete shared prompt without supplying an answer or token', () => {
@@ -135,64 +122,32 @@ describe('bounded shared-code revalidation prompt', () => {
     const contract = prompt.slice(base.length);
     expect(contract).toContain(`${SHARED_INTERACTIVE_MAX_TURNS} assistant turns`);
     expect(contract).toContain(path.join(f.state, 'projects/fixture-shared-libs/.review-starts/<REVIEW_START>.json'));
-    expect(contract).toContain('token actually returned by --start');
-    expect(contract).toContain('separate, successful Read tool call or a single cat command');
-    expect(contract).toContain('Verify its repo, branch, working tree and start time');
-    expect(contract).toContain('Do not combine the record read with --start, the diff or other diagnostic commands');
-    expect(contract).toContain('if the read fails, retry it before proceeding');
-    expect(contract).toContain('Do not read the diff until step 2 verifies the start record');
+    expect(contract.indexOf('Read that token\'s record')).toBeGreaterThan(-1);
     expect(contract.indexOf('Read that token\'s record')).toBeLessThan(contract.indexOf('Then read the diff in a subsequent call'));
     expect(contract).not.toMatch(/[a-f0-9]{8}(?:-[a-f0-9]{4}){3}-[a-f0-9]{12}/);
-    expect(prompt).toContain(`${SHARED_LIBS_ROOT}/review/sections/shared-code-reuse.md (Step 5.0 requires it for the supplied prior Skip)`);
+    expect(prompt).toContain(`${SHARED_LIBS_ROOT}/review/sections/shared-code-reuse.md`);
     expect(prompt.indexOf('Your first response holds')).toBeLessThan(prompt.indexOf("gstack-review-log' --start review"));
-    expect(prompt).toContain('Keep the final review summary to at most twelve lines');
-    expect(base).not.toContain('Keep the final review summary');
-    expect(contract).toContain('Batch independent required source reads');
-    expect(contract).toContain('Preserve every required evidence check and dependency');
-    expect(contract).toContain('complete, untruncated read-back');
-    expect(contract).toContain('same tool invocation');
-    expect(contract).toContain('then return the final review summary');
-    expect(contract).toContain('Failed persistence or verification remains a failure');
-    expect(contract).toContain("Late source changes still require the workflow's normal re-review");
     expect(contract).not.toContain('choose Skip');
   });
 
-  test('declares isolated receipt commands, direct reads and a separate supplied-finding disposition', () => {
+  test('declares isolated receipt commands', () => {
     const prompt = reviewRevalidationPrompt(f, instructions, input);
     const commands = [...prompt.matchAll(/```bash\n([\s\S]*?)\n```/g)].map(match => match[1]);
     expect(commands).toHaveLength(3);
     expect(commands[0]).toBe(`'${SHARED_LIBS_ROOT}/bin/gstack-review-log' --start review`);
     expect(commands[1]).toBe(`'${SHARED_LIBS_ROOT}/bin/gstack-review-log' --check-shared-libs REVIEW_START <<'GSTACK_REVALIDATION_FINDING'\nCURRENT_FINDING_JSON\nGSTACK_REVALIDATION_FINDING`);
     expect(commands[2]).toBe(`'${SHARED_LIBS_ROOT}/bin/gstack-review-log' 'FINAL_REVIEW_JSON' --finish REVIEW_START && '${SHARED_LIBS_ROOT}/bin/gstack-review-read'`);
-    for (const requirement of ['sole command', 'only the token', 'only one JSON value', 'literal path operands',
-      'even when its body appeared in the diff', 'No path-variable loops', 'before the checker',
-      'metadata in earlier calls', 'No preliminary commands', 'materially revised proposal is a separate finding',
-      'unsupported or unfinished supplied finding stays blocked', 'without its actual explicit decision']) {
-      expect(prompt).toContain(requirement);
-    }
   });
 
-  test('step 5 branches on the checker result: true suppresses without a new decision, false requires a fresh one', () => {
+  test('step 5 branches on the checker result, reusable:true before reusable:false', () => {
     const prompt = reviewRevalidationPrompt(f, instructions, input);
     const step5 = prompt.slice(prompt.indexOf('5. Act on the checker result'), prompt.indexOf('6. Complete final evidence'));
-    expect(step5).toContain('reusable:true');
-    expect(step5).toContain('the prior Skip carries forward');
-    expect(step5).toContain('Ask no new decision question');
-    expect(step5).toContain("exclude this advisory from the current pass's findings");
-    expect(step5).toContain('Do not re-persist it as a current finding');
-    expect(step5).toContain('reusable:false');
-    expect(step5).toContain('Perform a fresh authored-source review and make an actual new decision');
-    expect(step5).toContain('materially revised proposal is a separate finding');
-    expect(step5).toContain('without its actual explicit decision');
-    expect(step5).toContain('unsupported or unfinished supplied finding stays blocked');
-    expect(step5).toContain('reusable:true whose independent authored/current-source verification does not hold');
-    expect(step5).toContain('regardless of the checker result');
-    expect(step5).toContain('never force a new Skip on invalid evidence');
+    expect(step5.indexOf('reusable:true')).toBeGreaterThan(-1);
     expect(step5.indexOf('reusable:true')).toBeLessThan(step5.indexOf('reusable:false'));
     expect(prompt).not.toContain("Record the supplied finding's disposition under its own");
   });
 
-  test('the shared review prompt maps exact trusted asset roots and documented interfaces to avoid discovery', () => {
+  test('the shared review prompt maps exact trusted asset roots and documented interfaces', () => {
     const prompt = reviewPrompt(f, instructions, input);
     expect(prompt).toContain(`${SHARED_LIBS_ROOT}/review/checklist.md`);
     expect(prompt).toContain(`${SHARED_LIBS_ROOT}/review/sections/`);
@@ -202,21 +157,13 @@ describe('bounded shared-code revalidation prompt', () => {
     expect(prompt).toContain(f.bin);
     for (const iface of ['gstack-review-log --start review', '--check-shared-libs REVIEW_START',
       '--finish REVIEW_START', 'gstack-review-read']) expect(prompt).toContain(iface);
-    for (const forbidden of ['do not rediscover it', 'enumerate the bin/lib/review/qa roots',
-      'probe --help', 'read the fixture request logs']) expect(prompt).toContain(forbidden);
-    expect(prompt).toContain('batch independent reads');
-    expect(prompt).toContain('keep receipt-ordered commands separate');
-    expect(prompt).toContain('capture the start token before reading the diff');
-    expect(prompt).toContain('run --start, the checker and any declared stage-actor invocation each as its own sole command');
-    expect(prompt).toContain('The only combined receipt call is the final persistence');
-    expect(prompt).toContain('Still inspect the target repository source');
   });
 
-  test('common guidance states the finish+read-back receipt contract directly, not by a dangling step 6 reference', () => {
+  test('common guidance states the finish+read-back receipt directly, not by a dangling step 6 reference', () => {
     const lifecycle = reviewPrompt(f, instructions, input, { actorCommand: 'bun /fx/stage-actor.ts run' });
     const revalidation = reviewRevalidationPrompt(f, instructions, input);
     for (const prompt of [lifecycle, revalidation]) {
-      expect(prompt).toContain('The only combined receipt call is the final persistence');
+      expect(prompt).toContain('--finish REVIEW_START');
       expect(prompt).not.toContain('exactly as step 6 shows');
     }
     const commands = [...revalidation.matchAll(/```bash\n([\s\S]*?)\n```/g)].map(match => match[1]);
diff --git a/test/ship-triage-labels.test.ts b/test/ship-triage-labels.test.ts
new file mode 100644
index 000000000..f812f2af4
--- /dev/null
+++ b/test/ship-triage-labels.test.ts
@@ -0,0 +1,29 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { triageLabels } from './helpers/ship-triage-labels';
+
+const DIR = path.join(import.meta.dir, 'fixtures', 'ship-triage');
+const read = (name: string) => fs.readFileSync(path.join(DIR, name), 'utf8');
+const correct = (output: string) => {
+  const labels = triageLabels(output);
+  return labels.string === 'in-branch' && labels.math === 'pre-existing';
+};
+
+describe('ship triage label grader', () => {
+  test.each(['known-good-json.txt', 'known-good-lines.txt'])('accepts %s', (name) => {
+    expect(correct(read(name))).toBe(true);
+  });
+  test.each(['known-bad-swapped.txt', 'known-bad-words-only.txt', 'known-bad-one-label.txt'])('rejects %s', (name) => {
+    expect(correct(read(name))).toBe(false);
+  });
+  test('the previous word-presence check passed the words-only and swapped outputs', () => {
+    const previous = (output: string) => {
+      const lower = output.toLowerCase();
+      return /in-branch|in branch|introduced/.test(lower) && /pre-existing|pre existing|existed before/.test(lower)
+        && /truncate|string/.test(lower) && /divide|math/.test(lower);
+    };
+    expect(previous(read('known-bad-words-only.txt'))).toBe(true);
+    expect(previous(read('known-bad-swapped.txt'))).toBe(true);
+  });
+});
diff --git a/test/skill-body-narration.test.ts b/test/skill-body-narration.test.ts
new file mode 100644
index 000000000..acac78c97
--- /dev/null
+++ b/test/skill-body-narration.test.ts
@@ -0,0 +1,25 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { hasNarrationLeak } from './helpers/skill-body-narration';
+
+const DIR = path.join(import.meta.dir, 'fixtures', 'skillify-narration');
+const read = (name: string) => fs.readFileSync(path.join(DIR, name), 'utf8');
+const KNOWN_BAD = fs.readdirSync(DIR).filter(name => name.startsWith('known-bad-')).sort();
+
+describe('skillify narration grader', () => {
+  test('accepts a clean body that says "let me know" in a trigger', () => {
+    expect(hasNarrationLeak(read('known-good.md'))).toBe(false);
+  });
+  test('stored known-bad fixtures exist', () => {
+    expect(KNOWN_BAD.length).toBeGreaterThanOrEqual(4);
+  });
+  test.each(KNOWN_BAD)('still rejects %s', (name) => {
+    expect(hasNarrationLeak(read(name))).toBe(true);
+  });
+  test('the previous grader rejected every known-bad fixture and also the known-good one', () => {
+    const previous = (body: string) => /^I /m.test(body) || /Let me /i.test(body) || /^I'll /m.test(body);
+    for (const name of KNOWN_BAD) expect(previous(read(name))).toBe(true);
+    expect(previous(read('known-good.md'))).toBe(true);
+  });
+});
diff --git a/test/skill-ceo-section-ordering.test.ts b/test/skill-ceo-section-ordering.test.ts
index e7d72809c..3cee5f00a 100644
--- a/test/skill-ceo-section-ordering.test.ts
+++ b/test/skill-ceo-section-ordering.test.ts
@@ -34,6 +34,7 @@ import { ALL_HOST_CONFIGS } from '../hosts';
 import { generateTasksSectionEmit } from '../scripts/resolvers/tasks-section';
 import { generateBrainCacheRefresh } from '../scripts/resolvers/gbrain';
 import { runGeneration } from '../scripts/gen-skill-docs';
+import { ASK_QUESTIONS_HEADING } from './helpers/workflow-excerpt';
 
 const ROOT = path.resolve(import.meta.dir, '..');
 const SKELETON = path.join(ROOT, 'plan-ceo-review', 'SKILL.md');
@@ -935,12 +936,13 @@ test('CEO closing route checks approvals before outputs and verifies artifacts b
     expect(refresh).not.toContain('telemetry has logged');
   }
 
-  const governingStages = ['## CRITICAL RULE — How to ask questions', '## Formatting Rules',
+  const askQuestions = section.search(ASK_QUESTIONS_HEADING);
+  const governingStages = [askQuestions, ...['## Formatting Rules',
     '## Mode Quick Reference', '### Working review decisions', '### Section 1:']
-    .map(stage => section.indexOf(stage));
+    .map(stage => section.indexOf(stage))];
   expect(governingStages.every(position => position >= 0)).toBe(true);
   expect(governingStages).toEqual([...governingStages].sort((a, b) => a - b));
-  const questions = section.split('## CRITICAL RULE — How to ask questions')[1]!.split('## Mode Quick Reference')[0]!;
+  const questions = section.slice(askQuestions).split('\n').slice(1).join('\n').split('## Mode Quick Reference')[0]!;
   expect(questions).toContain('Use `D<N>` and A/B/C labels');
   expect(questions).toContain('Cite the stable ledger ID separately');
   const formatting = questions.split('## Formatting Rules')[1]!;
diff --git a/test/skill-e2e-auq-matrix.test.ts b/test/skill-e2e-auq-matrix.test.ts
index 1fe3bcee7..0eb699a0d 100644
--- a/test/skill-e2e-auq-matrix.test.ts
+++ b/test/skill-e2e-auq-matrix.test.ts
@@ -4,9 +4,10 @@
  *
  * Layer 0 (auq-format-always-loaded.test.ts) deterministically guarantees each
  * listed skill SHIPS the format spec in its always-loaded skeleton. This test
- * proves each skill's model OBEYS it: that the first real AUQ it fires is a
- * compliant decision brief (all 7 format elements) with a substantive
- * recommendation (>= 4). One parametrized case per skill so a single weak skill
+ * proves each skill's model OBEYS it: that the first real AUQ it fires carries
+ * the fields software reads (a Recommendation: line and exactly one
+ * (recommended) option) with a substantive recommendation (>= 4). The other
+ * format elements are logged. One parametrized case per skill so a single weak skill
  * is an isolated failure, not a blocker for the rest.
  *
  * Capture records the actual public AskUserQuestion payload and verifies its
@@ -35,6 +36,7 @@ import {
   setupSkillDir,
   captureFirstAuq,
   scoreAuqFormat,
+  auqMachineFormatProblems,
   skillFromWorktree,
   gradeAuqRecommendation,
 } from './helpers/auq-sdk-capture';
@@ -156,7 +158,8 @@ describeE2E('AUQ behavioral matrix (periodic)', () => {
           throw new Error(`${m.skill}: agent produced NO AUQ capture (never reached a question in budget).`);
         }
         const problems: string[] = [];
-        if (fmt.missing.length > 0) problems.push(`missing format element(s): ${fmt.missing.join(', ')}`);
+        // Presentation elements (ELI10, Pros / cons, ✅/❌, Net:) are logged above, not failed.
+        problems.push(...auqMachineFormatProblems(text));
         if (substance < 4) problems.push(`recommendation substance ${substance} < 4 (boilerplate/weak)`);
         if (problems.length > 0) {
           throw new Error(
diff --git a/test/skill-e2e-context-skills.test.ts b/test/skill-e2e-context-skills.test.ts
index b50c09beb..71fabe0da 100644
--- a/test/skill-e2e-context-skills.test.ts
+++ b/test/skill-e2e-context-skills.test.ts
@@ -21,6 +21,7 @@ import {
   createEvalCollector, finalizeEvalCollector,
 } from './helpers/e2e-helpers';
 import { extractSkillBody } from './helpers/skill-fixture';
+import { readShippedSkillRouting } from './helpers/shipped-skill-routing';
 import { spawnSync } from 'child_process';
 import * as fs from 'fs';
 import * as path from 'path';
@@ -70,19 +71,13 @@ function setupWorkdir(suffix: string): { workDir: string; gstackHome: string; sl
     }
   }
 
-  // Routing CLAUDE.md: explicit instruction to always use the Skill tool.
+  // Routing CLAUDE.md: the ## Skill routing section gstack ships.
   fs.writeFileSync(path.join(workDir, 'CLAUDE.md'), `# Project Instructions
 
-## Skill routing
+${readShippedSkillRouting().section}
 
-When the user's request matches an available skill, ALWAYS invoke it using the Skill
-tool as your FIRST action. Do NOT answer directly, do NOT use other tools first.
+## Test environment
 
-Key routing rules:
-- Save progress, save state, save my work → invoke context-save
-- Resume, where was I, pick up where I left off → invoke context-restore
-
-Environment:
 - Use GSTACK_HOME="${gstackHome}" for all gstack bin scripts.
 - The bin scripts are at ./bin/ (relative to this directory).
 - The skill files are at ./.claude/skills/context-save/SKILL.md and
diff --git a/test/skill-e2e-coverage-audit.test.ts b/test/skill-e2e-coverage-audit.test.ts
index b78e0a545..de3dcaa5b 100644
--- a/test/skill-e2e-coverage-audit.test.ts
+++ b/test/skill-e2e-coverage-audit.test.ts
@@ -66,6 +66,8 @@ Existing tests are in ${cwd}/test/billing.test.ts.
 
 Produce the ASCII coverage diagram showing which code paths are tested and which have gaps.
 Output the diagram directly, name both billing functions, and include a coverage summary.
+End with one JSON line listing the functions with and without test coverage:
+{"tested": ["<function>", ...], "untested": ["<function>", ...]}
 Do not modify the supplied source or tests.`,
             workingDirectory: cwd, maxTurns: 15,
             allowedTools: ['Bash', 'Read', 'Write', 'Edit', 'Glob', 'Grep'],
diff --git a/test/skill-e2e-deploy.test.ts b/test/skill-e2e-deploy.test.ts
index c52eac569..a322ba481 100644
--- a/test/skill-e2e-deploy.test.ts
+++ b/test/skill-e2e-deploy.test.ts
@@ -52,19 +52,12 @@ The skill is carved: on-demand step bodies live in land-and-deploy/sections/ in
 working directory — when a STOP-Read pointer names a ~/.claude/skills/gstack/... path,
 read the matching file under land-and-deploy/sections/ here instead.
 
-You are on branch feat/add-deploy with changes against main. This repo has a fly.toml
-with app = "test-app", indicating a Fly.io deployment.
+You are on branch feat/add-deploy with changes against main.
 
-IMPORTANT: There is NO remote and NO GitHub PR — you cannot run gh commands.
-Instead, simulate the workflow:
-1. Detect the deploy platform from fly.toml (should find Fly.io, app = test-app)
-2. Infer the production URL (https://test-app.fly.dev)
-3. Note the merge method would be squash
-4. Write the deploy configuration to CLAUDE.md
-5. Write a deploy report skeleton to .gstack/deploy-reports/report.md showing the
-   expected report structure (PR number: simulated, timing: simulated, verdict: simulated)
-
-Do NOT use AskUserQuestion. Do NOT run gh or fly commands.`,
+This run is non-interactive and the repo has no remote, so there is no PR and the gh
+and fly CLIs are unavailable. Run the parts of the workflow that work without them,
+mark the PR, merge and deploy results as simulated, and save the report you would
+show under .gstack/deploy-reports/.`,
       workingDirectory: landDir,
       maxTurns: 20,
       allowedTools: ['Bash', 'Read', 'Write', 'Edit', 'Grep', 'Glob'],
@@ -77,15 +70,16 @@ Do NOT use AskUserQuestion. Do NOT run gh or fly commands.`,
     recordE2E(evalCollector, '/land-and-deploy workflow', 'Land-and-Deploy skill E2E', result);
     expect(result.exitReason).toBe('success');
 
-    const claudeMd = path.join(landDir, 'CLAUDE.md');
-    if (fs.existsSync(claudeMd)) {
-      const content = fs.readFileSync(claudeMd, 'utf-8');
-      const hasFly = content.toLowerCase().includes('fly') || content.toLowerCase().includes('test-app');
-      expect(hasFly).toBe(true);
-    }
-
+    // The prompt names neither the platform nor the app: the skill detects them from fly.toml.
     const reportDir = path.join(landDir, '.gstack', 'deploy-reports');
     expect(fs.existsSync(reportDir)).toBe(true);
+    const claudeMd = path.join(landDir, 'CLAUDE.md');
+    const artifacts = [
+      ...fs.readdirSync(reportDir).filter(f => f.endsWith('.md')).map(f => fs.readFileSync(path.join(reportDir, f), 'utf-8')),
+      fs.existsSync(claudeMd) ? fs.readFileSync(claudeMd, 'utf-8') : '',
+    ].join('\n');
+    expect(artifacts, 'deploy artifacts name the detected platform').toMatch(/fly/i);
+    expect(artifacts, 'deploy artifacts name the detected app').toContain('test-app');
   }, CAPTURE_LONG_MS);
 });
 
@@ -130,22 +124,9 @@ in THIS working directory — read it from there (the STOP-Read pointer's
 You are on branch feat/first-deploy. This is the FIRST TIME running /land-and-deploy
 for this project — there is NO land-deploy-confirmed file.
 
-This repo has a fly.toml with app = "first-run-app", indicating a Fly.io deployment.
-
-IMPORTANT: There is NO remote and NO GitHub PR — you cannot run gh commands.
-Instead, simulate the Step 1.5 first-run dry-run validation:
-1. Detect that this is a FIRST_RUN (no land-deploy-confirmed file)
-2. Detect the deploy platform from fly.toml (Fly.io, app = first-run-app)
-3. Infer the production URL (https://first-run-app.fly.dev)
-4. Build the DEPLOY INFRASTRUCTURE VALIDATION table showing:
-   - Platform detected
-   - Command validation results (simulated as all passing)
-   - Staging detection results (none expected)
-   - What will happen steps
-5. Write the dry-run report to .gstack/deploy-reports/dry-run-validation.md
-
-Do NOT use AskUserQuestion. Do NOT run gh or fly commands.
-Just demonstrate the first-run dry-run output.`,
+This run is non-interactive and the repo has no remote, so there is no PR and the gh
+and fly CLIs are unavailable; treat their command checks as passing. Run the first-run
+dry-run validation and save its report under .gstack/deploy-reports/.`,
       workingDirectory: firstRunDir,
       maxTurns: 20,
       allowedTools: ['Bash', 'Read', 'Write', 'Edit', 'Grep', 'Glob'],
@@ -210,20 +191,12 @@ The Step 3.5 readiness gate is carved into land-and-deploy/sections/readiness-ga
 in THIS working directory — read it from there (the STOP-Read pointer's
 ~/.claude/skills/gstack/... path does not exist here).
 
-Focus on Step 3.5a and Step 3.5a-bis (the review staleness check and inline review offer).
+Focus on Step 3.5a and Step 3.5a-bis.
 
-This repo has 6 commits since the initial commit. There are NO review logs
-(gstack-review-read would return NO_REVIEWS).
-
-Simulate what the readiness gate would show:
-1. Run gstack-review-read equivalent (simulate NO_REVIEWS output)
-2. Determine review staleness: Eng Review should be "NOT RUN"
-3. Note that Step 3.5a-bis would offer an inline review
-4. Write a simulated readiness report to .gstack/deploy-reports/readiness-report.md
-   showing the review status as NOT RUN with the inline review offer text
-
-Do NOT use AskUserQuestion. Do NOT run gh commands.
-Show what the readiness gate output would look like.`,
+gstack-review-read is not installed here; it would print NO_REVIEWS for this repo.
+This run is non-interactive and the repo has no remote, so there is no PR and gh is
+unavailable. Run the readiness gate and save its readiness report under
+.gstack/deploy-reports/.`,
       workingDirectory: reviewDir,
       maxTurns: 15,
       allowedTools: ['Bash', 'Read', 'Write', 'Edit', 'Grep', 'Glob'],
@@ -243,10 +216,10 @@ Show what the readiness gate output would look like.`,
     const reportFiles = fs.readdirSync(reportDir);
     expect(reportFiles.length).toBeGreaterThan(0);
     const reportContent = fs.readFileSync(path.join(reportDir, reportFiles[0]), 'utf-8');
-    // Should mention review status
-    const hasReviewMention = reportContent.toLowerCase().includes('review') ||
-                              reportContent.toLowerCase().includes('not run');
-    expect(hasReviewMention).toBe(true);
+    // The prompt no longer supplies the verdict: the gate must record the
+    // missing review itself.
+    expect(reportContent).toMatch(/review/i);
+    expect(reportContent, 'readiness report records the missing review').toMatch(/not run|no reviews?\b|never run|missing/i);
   }, CAPTURE_LONG_MS);
 });
 
diff --git a/test/skill-e2e-design.test.ts b/test/skill-e2e-design.test.ts
index 233ba5c1a..d5afc5c0d 100644
--- a/test/skill-e2e-design.test.ts
+++ b/test/skill-e2e-design.test.ts
@@ -707,12 +707,16 @@ Review the site at ${serverUrl}. Use --quick mode. Skip any AskUserQuestion call
     const reportPath = path.join(qaDesignDir, 'design-audit.md');
     const reportExists = fs.existsSync(reportPath);
 
-    // Check if any design fix commits were made
-    const gitLog = spawnSync('git', ['log', '--oneline'], {
+    // Outcome: the report names a seeded defect, and the page changed (committed
+    // or not) relative to the initial fixture commit. The commit prefix is not required.
+    const report = reportExists ? fs.readFileSync(reportPath, 'utf-8') : '';
+    const namesSeededDefect = /4[78]px|line-height|border-radius|radius|padding|spacing|heading/i.test(report);
+    const initialCommit = spawnSync('git', ['rev-list', '--max-parents=0', 'HEAD'], {
       cwd: qaDesignDir, stdio: 'pipe', timeout: 30_000,
-    });
-    const commits = gitLog.stdout.toString().trim().split('\n');
-    const designFixCommits = commits.filter((c: string) => c.includes('style(design)'));
+    }).stdout.toString().trim();
+    const pageChanged = spawnSync('git', ['diff', '--name-only', initialCommit, '--', 'index.html', 'style.css'], {
+      cwd: qaDesignDir, stdio: 'pipe', timeout: 30_000,
+    }).stdout.toString().trim() !== '';
 
     // The agent must actually drive Aside: an `aside repl` Bash call, a printed sentinel
     // (from a tool_result, never the input), and no reach for the retired browse binary.
@@ -724,7 +728,8 @@ Review the site at ${serverUrl}. Use --quick mode. Skip any AskUserQuestion call
     const usedBrowseBin = bashCommands.some(c => /browse\/dist\/browse|\$B /.test(c));
 
     recordE2E(evalCollector, '/design-review fix', 'Design Review E2E', result, {
-      passed: ['success', 'error_max_turns'].includes(result.exitReason) && droveAside && sentinelPrinted && !usedBrowseBin,
+      passed: ['success', 'error_max_turns'].includes(result.exitReason) && droveAside && sentinelPrinted && !usedBrowseBin
+        && reportExists && namesSeededDefect && pageChanged,
     });
 
     // Accept error_max_turns — the fix loop is complex
@@ -732,15 +737,9 @@ Review the site at ${serverUrl}. Use --quick mode. Skip any AskUserQuestion call
     expect(droveAside).toBe(true);
     expect(sentinelPrinted).toBe(true);
     expect(usedBrowseBin).toBe(false);
-
-    // Report and commits are best-effort — log what happened
-    if (reportExists) {
-      const report = fs.readFileSync(reportPath, 'utf-8');
-      console.log(`Design audit report: ${report.length} chars`);
-    } else {
-      console.warn('No design-audit.md generated');
-    }
-    console.log(`Design fix commits: ${designFixCommits.length}`);
+    expect(reportExists, 'design-audit.md must be written').toBe(true);
+    expect(namesSeededDefect, 'the report names a seeded design defect').toBe(true);
+    expect(pageChanged, 'at least one design issue is fixed in index.html or style.css').toBe(true);
   }, CAPTURE_LONG_MS);
 });
 
diff --git a/test/skill-e2e-office-hours-auto-mode.test.ts b/test/skill-e2e-office-hours-auto-mode.test.ts
index d32e835f4..8680a2f1e 100644
--- a/test/skill-e2e-office-hours-auto-mode.test.ts
+++ b/test/skill-e2e-office-hours-auto-mode.test.ts
@@ -8,8 +8,8 @@
  * (office-hours/SKILL.md.tmpl:69); when AskUserQuestion is disallowed at
  * the tool-registry level the model cannot ask and silently picks one mode,
  * breaking the whole interactive premise. This test asserts that question
- * still surfaces — fix must route through mcp__conductor__AskUserQuestion
- * (when present) or plan-file + ExitPlanMode flow.
+ * still surfaces — through mcp__conductor__AskUserQuestion (when present)
+ * or the prose fallback, never a plan-file ## Decisions substitute.
  *
  * Filename keeps `auto-mode` for branch-history continuity. Auto-mode (the
  * AUTO_DECIDE preamble path when QUESTION_TUNING=true) is a related but
@@ -20,12 +20,15 @@ import { test, expect } from 'bun:test';
 import { CAPTURE_MS, CAPTURE_LONG_MS } from './helpers/eval-budgets';
 import { describeE2ETier } from './helpers/e2e-gate';
 import { runPlanSkillObservation, planFileHasDecisionsSection } from './helpers/claude-pty-runner';
+import { assertNoPlanFileDecisions } from './helpers/plan-mode-evidence';
 
 const describeE2E = describeE2ETier('gate');
 
 describeE2E('office-hours AskUserQuestion-blocked smoke (gate)', () => {
-  // Pass envelope is ['asked', 'plan_ready']; failure signals are
-  // 'auto_decided' + silent_write/exited/timeout.
+  // With AskUserQuestion blocked the mode question must still be asked (the
+  // prose fallback, observed as 'asked'); a plan-file ## Decisions section is
+  // not a substitute. Failure signals also include 'auto_decided' +
+  // silent_write/exited/timeout.
   test('AskUserQuestion surfaces when --disallowedTools AskUserQuestion is set', async () => {
     const obs = await runPlanSkillObservation({
       skillName: 'office-hours',
@@ -47,14 +50,12 @@ describeE2E('office-hours AskUserQuestion-blocked smoke (gate)', () => {
           `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
       );
     }
-    if (obs.outcome === 'plan_ready') {
-      if (!obs.planFile || !planFileHasDecisionsSection(obs.planFile)) {
-        throw new Error(
-          `office-hours AskUserQuestion-blocked regression: plan_ready without a "## Decisions" section in ${obs.planFile ?? '<no plan file detected>'} — startup-vs-builder mode question was silently skipped.\n` +
-            `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
-        );
-      }
+    assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
+    if (obs.outcome !== 'asked') {
+      throw new Error(
+        `office-hours AskUserQuestion-blocked regression: outcome=${obs.outcome} — the startup-vs-builder mode question reached ${obs.planFile ?? 'plan_ready'} without being asked.\n` +
+          `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
+      );
     }
-    expect(['asked', 'plan_ready']).toContain(obs.outcome);
   }, CAPTURE_LONG_MS);
 });
diff --git a/test/skill-e2e-outside-plan-disabled.test.ts b/test/skill-e2e-outside-plan-disabled.test.ts
index 85c861f7b..ebf46d42c 100644
--- a/test/skill-e2e-outside-plan-disabled.test.ts
+++ b/test/skill-e2e-outside-plan-disabled.test.ts
@@ -38,7 +38,7 @@ describeLive('Disabled outside plan review', () => {
       fs.writeFileSync(path.join(repo, 'PLAN.md'), '# Accepted plan\n\nAdd a documented --json flag to the existing status command. Preserve the text output by default, and test both formats. Native review is complete.\n');
       // The parent must discover the control through the actual preflight.
       // Do not tell it the expected branch or forbid Agent in the prompt/tools.
-      const prompt = 'The native engineering review of PLAN.md is complete. Read OUTSIDE-PLAN.md and execute that final workflow section for this plan, following its actual configuration and branch rules. Then return the workflow completion summary, including outside-review attribution. This is an isolated review fixture; all required reviewer tools are authorized.';
+      const prompt = 'The native engineering review of PLAN.md is complete. Read OUTSIDE-PLAN.md and execute that final workflow section for this plan, following its actual configuration and branch rules. Then return the workflow completion summary, including outside-review attribution, and end with one JSON line {"outside_review_this_run": "completed" | "disabled" | "unavailable" | "skipped"}. This is an isolated review fixture; all required reviewer tools are authorized.';
       fs.writeFileSync(path.join(evidenceDir, 'instructions.md'), fixture.instructions);
       const result = await runSkillTest({
         workingDirectory: repo, prompt, testName, runId,
diff --git a/test/skill-e2e-plan-ceo-plan-mode.test.ts b/test/skill-e2e-plan-ceo-plan-mode.test.ts
index 39f5f37ec..e3d27cd84 100644
--- a/test/skill-e2e-plan-ceo-plan-mode.test.ts
+++ b/test/skill-e2e-plan-ceo-plan-mode.test.ts
@@ -99,8 +99,7 @@ describeE2E('plan-ceo-review plan-mode smoke (gate)', () => {
           `plan-ceo-review smoke FAILED: outcome=${obs.outcome}\n` +
             `${diagnosis}\n` +
             `Expected 'asked'. See plan-ceo-review/SKILL.md.tmpl: the Step 0 STOP rules ` +
-            `and the "One issue = one AskUserQuestion call" rule under "CRITICAL RULE — ` +
-            `How to ask questions".\n` +
+            `and the "One decision unit = one AskUserQuestion call" rule in its question-procedure section.\n` +
             `summary: ${obs.summary}\n` +
             `elapsed: ${obs.elapsedMs}ms\n` +
             `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
diff --git a/test/skill-e2e-plan-ceo-split-overflow.test.ts b/test/skill-e2e-plan-ceo-split-overflow.test.ts
index 392cef32d..b06d95054 100644
--- a/test/skill-e2e-plan-ceo-split-overflow.test.ts
+++ b/test/skill-e2e-plan-ceo-split-overflow.test.ts
@@ -14,9 +14,9 @@
  * The fixture seeds 5 independent scope candidates (chat-platform
  * integrations) — each carries an independent include/defer/cut decision.
  * The existing semantic scope validator examines every acknowledged native
- * call, including candidate choices before mode selection. It keeps the
- * N-1 call floor and requires independent offered dispositions for all five
- * candidates. The review-phase counter reports progress, not target coverage.
+ * call, including candidate choices before mode selection. It requires
+ * independent offered dispositions for all five candidates; the call floor
+ * is two, the minimum for five options at four per call. The review-phase counter reports progress, not target coverage.
  * Collection ends once all five native choices are acknowledged; the same
  * semantic validator then decides whether those choices satisfy the metric.
  *
@@ -46,14 +46,14 @@ import {
   ceoStep0Boundary,
 } from './helpers/claude-pty-runner';
 import { FORCING_SPLIT_OVERFLOW_CEO } from './fixtures/forcing-finding-seeds';
-import { ceoSplitDecisionFingerprints, isCeoSplitCandidateCall, isCeoSplitCollectionComplete, pickCeoSplitCountQuestion } from './helpers/ceo-split-question-policy';
+import { CEO_SPLIT_CALL_FLOOR, ceoSplitDecisionFingerprints, isCeoSplitCandidateCall, isCeoSplitCollectionComplete, pickCeoSplitCountQuestion } from './helpers/ceo-split-question-policy';
 import { CEO_SCOPE_CANDIDATES } from './helpers/plan-review-cases';
 import { evaluatePlanReviewDecisions } from './helpers/plan-review-decisions';
 
 const describeE2E = describeE2ETier('marathon');
 
 const N = 5;
-const FLOOR = N - 1; // 4 — must fire at least one AUQ per non-dropped option
+const FLOOR = CEO_SPLIT_CALL_FLOOR;
 
 /** Plan-file target baked into the FORCING_SPLIT_OVERFLOW_CEO fixture prompt.
  *  Rewritten per-run to a mkdtemp path so concurrent runs (--retry,
@@ -101,9 +101,9 @@ describeE2E('/plan-ceo-review split-overflow regression (marathon)', () => {
               `--- evidence (last 3KB) ---\n${obs.evidence}`,
           );
         }
-        // The phase counter is progress, not target coverage. The existing scope
-        // validator keeps FLOOR=4 and requires all five independent candidate
-        // decisions, complete native evidence, and offered Include/Defer/Cut ACKs.
+        // The phase counter is progress, not target coverage. The scope validator
+        // requires all five independent candidate decisions, complete native
+        // evidence, and offered Include/Defer/Cut ACKs; FLOOR only bounds calls.
         await evaluatePlanReviewDecisions({ plan: followUpPrompt, targets: CEO_SCOPE_CANDIDATES,
           fingerprints: ceoSplitDecisionFingerprints(obs.transcript, obs.fingerprints),
           floor: FLOOR, kind: 'scope', deadlineAt });
diff --git a/test/skill-e2e-plan-design-plan-mode.test.ts b/test/skill-e2e-plan-design-plan-mode.test.ts
index e9619fe8d..8701c3702 100644
--- a/test/skill-e2e-plan-design-plan-mode.test.ts
+++ b/test/skill-e2e-plan-design-plan-mode.test.ts
@@ -12,10 +12,11 @@
 import { test, expect } from 'bun:test';
 import { CAPTURE_MS, CAPTURE_LONG_MS } from './helpers/eval-budgets';
 import { describeE2ETier } from './helpers/e2e-gate';
-import { assertPlanModeWithEvidence } from './helpers/plan-mode-evidence';
+import { assertNoPlanFileDecisions, assertPlanModeWithEvidence, seededPlanTargeted } from './helpers/plan-mode-evidence';
 import {
   runPlanSkillObservation,
   assertReportAtBottomIfPlanWritten,
+  planFileHasDecisionsSection,
 } from './helpers/claude-pty-runner';
 
 const describeE2E = describeE2ETier('periodic');
@@ -43,6 +44,9 @@ Only the happy path is designed. No empty states, no error states,
 no loading states. Mobile: "stacks on mobile."
 `;
 
+// Seed-only phrases; seeing one after the slash command shows the review read the seeded plan.
+const SEED_PLAN_TOKENS = ['Marketing landing page', 'testimonials carousel', '3 tier cards'];
+
 describeE2E('plan-design-review plan-mode smoke (periodic)', () => {
   test('reaches a terminal outcome (asked or plan_ready) without silent writes', async () => {
     const obs = await runPlanSkillObservation({
@@ -62,19 +66,21 @@ describeE2E('plan-design-review plan-mode smoke (periodic)', () => {
       }
       expect(['asked', 'plan_ready']).toContain(obs.outcome);
       assertReportAtBottomIfPlanWritten(obs);
+      assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
     });
   }, CAPTURE_LONG_MS);
 
   // Plan-mode scope-gate bypass: with a seeded UI-heavy plan in plan mode,
-  // the gate must NOT render its "What should I review?" menu — it
-  // auto-selects B and announces it, then proceeds to the pre-review audit
-  // and mockups. Mirrors the eng smoke's seeded STOP-gate test, without
+  // the gate must NOT render its "What should I review?" menu; the review
+  // targets the seeded plan, then proceeds to the pre-review audit and
+  // mockups. Mirrors the eng smoke's seeded STOP-gate test, without
   // --disallowedTools (native AUQ available is the common path here).
   test('scope gate auto-selects B when a plan is seeded in plan mode', async () => {
     const obs = await runPlanSkillObservation({
       skillName: 'plan-design-review',
       inPlanMode: true,
       initialPlanContent: SEED_PLAN_UI_HEAVY,
+      trackTokens: SEED_PLAN_TOKENS,
       timeoutMs: CAPTURE_MS,
     });
 
@@ -95,11 +101,13 @@ describeE2E('plan-design-review plan-mode smoke (periodic)', () => {
 
       expect(['asked', 'plan_ready']).toContain(obs.outcome);
       assertReportAtBottomIfPlanWritten(obs);
+      assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
 
-      // The bypass contract (exception ordering makes this deterministic even
-      // though the seed arrives as a pasted user message).
+      // The bypass contract: no scope menu, and the seeded plan is the target.
+      // The announcement's exact wording is not graded.
       expect(obs.scopeGateQuestionObserved ?? false).toBe(false);
-      expect(obs.scopeGateAutoSelectObserved ?? false).toBe(true);
+      console.log(`[plan-design plan-mode] scope announcement observed: ${obs.scopeGateAutoSelectObserved ?? false}; seed tokens: ${JSON.stringify(obs.tokensObserved ?? {})}`);
+      expect(seededPlanTargeted(obs, SEED_PLAN_TOKENS), 'seeded plan was not targeted (scopeGateAutoSelectObserved or seed-only plan tokens)').toBe(true);
     });
   }, CAPTURE_LONG_MS);
 });
diff --git a/test/skill-e2e-plan-devex-plan-mode.test.ts b/test/skill-e2e-plan-devex-plan-mode.test.ts
index 340db45ca..707408ee1 100644
--- a/test/skill-e2e-plan-devex-plan-mode.test.ts
+++ b/test/skill-e2e-plan-devex-plan-mode.test.ts
@@ -13,6 +13,7 @@ import {
   planFileHasDecisionsSection,
   assertReportAtBottomIfPlanWritten,
 } from './helpers/claude-pty-runner';
+import { assertNoPlanFileDecisions } from './helpers/plan-mode-evidence';
 
 const describeE2E = describeE2ETier('gate');
 
@@ -34,12 +35,14 @@ describeE2E('plan-devex-review plan-mode smoke (gate)', () => {
     }
     expect(['asked', 'plan_ready']).toContain(obs.outcome);
     assertReportAtBottomIfPlanWritten(obs);
+    assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
   }, CAPTURE_LONG_MS);
 
   // v1.21+ regression: see skill-e2e-plan-ceo-plan-mode.test.ts for the
-  // contract. Pass envelope is ['asked', 'plan_ready']; failure signals
-  // are 'auto_decided' (AUTO_DECIDE without opt-in) plus the standard
-  // silent_write/exited/timeout.
+  // contract. With AskUserQuestion blocked the decision must still be asked
+  // (the prose fallback, observed as 'asked'); a plan-file ## Decisions
+  // section is not a substitute. Failure signals also include 'auto_decided'
+  // (AUTO_DECIDE without opt-in) plus the standard silent_write/exited/timeout.
   test('AskUserQuestion surfaces when --disallowedTools AskUserQuestion is set', async () => {
     const obs = await runPlanSkillObservation({
       skillName: 'plan-devex-review',
@@ -61,15 +64,13 @@ describeE2E('plan-devex-review plan-mode smoke (gate)', () => {
           `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
       );
     }
-    if (obs.outcome === 'plan_ready') {
-      if (!obs.planFile || !planFileHasDecisionsSection(obs.planFile)) {
-        throw new Error(
-          `plan-devex-review AskUserQuestion-blocked regression: plan_ready without a "## Decisions" section in ${obs.planFile ?? '<no plan file detected>'} — Step 0 was silently skipped.\n` +
-            `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
-        );
-      }
-    }
-    expect(['asked', 'plan_ready']).toContain(obs.outcome);
     assertReportAtBottomIfPlanWritten(obs);
+    assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
+    if (obs.outcome !== 'asked') {
+      throw new Error(
+        `plan-devex-review AskUserQuestion-blocked regression: outcome=${obs.outcome} — Step 0 reached ${obs.planFile ?? 'plan_ready'} without asking.\n` +
+          `--- evidence (last 2KB visible) ---\n${obs.evidence}`,
+      );
+    }
   }, CAPTURE_LONG_MS);
 });
diff --git a/test/skill-e2e-plan-eng-plan-mode.test.ts b/test/skill-e2e-plan-eng-plan-mode.test.ts
index dd5acf03b..daf4fd418 100644
--- a/test/skill-e2e-plan-eng-plan-mode.test.ts
+++ b/test/skill-e2e-plan-eng-plan-mode.test.ts
@@ -8,7 +8,7 @@
 import { test, expect } from 'bun:test';
 import { CAPTURE_MS, CAPTURE_LONG_MS } from './helpers/eval-budgets';
 import { describeE2ETier } from './helpers/e2e-gate';
-import { assertPlanModeWithEvidence } from './helpers/plan-mode-evidence';
+import { assertNoPlanFileDecisions, assertPlanModeWithEvidence, seededPlanTargeted } from './helpers/plan-mode-evidence';
 import {
   runPlanSkillObservation,
   planFileHasDecisionsSection,
@@ -47,6 +47,9 @@ Ignore Bun's native --shard flag because we want full control.
 None planned — will add later.
 `;
 
+// Seed-only names; seeing one after the slash command shows the review read the seeded plan.
+const SEED_PLAN_TOKENS = ['ShardManager', 'ResultMerger', 'test-shard-impl'];
+
 describeE2E('plan-eng-review plan-mode smoke (periodic)', () => {
   test('reaches a terminal outcome (asked or plan_ready) without silent writes', async () => {
     const obs = await runPlanSkillObservation({
@@ -66,23 +69,26 @@ describeE2E('plan-eng-review plan-mode smoke (periodic)', () => {
       }
       expect(['asked', 'plan_ready']).toContain(obs.outcome);
       assertReportAtBottomIfPlanWritten(obs);
+      assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
     });
   }, CAPTURE_LONG_MS);
 
   // D3-B / D4-B: when a plan with guaranteed-finding-triggering complexity
-  // is seeded, the skill MUST fire AskUserQuestion (or fall back to a
-  // Decisions section) before writing findings to the plan. The
-  // wrote_findings_before_asking outcome catches the precise transcript bug
-  // — model writes findings to the plan before any AUQ render.
+  // is seeded, the skill must ask (mcp AskUserQuestion or the prose fallback)
+  // before writing findings to the plan. The wrote_findings_before_asking
+  // outcome catches the transcript bug where the model writes findings to the
+  // plan before any question renders. A plan-file ## Decisions section is not
+  // a substitute for asking.
   test('STOP gate fires when seeded plan forces Step 0 findings', async () => {
     const obs = await runPlanSkillObservation({
       skillName: 'plan-eng-review',
       inPlanMode: true,
       initialPlanContent: SEED_PLAN_FORCING_FINDINGS,
       // Force the Conductor-style path: native AUQ disallowed → the model
-      // must use mcp__*__AskUserQuestion (outcome='asked') or fall back to
-      // writing Decisions ('plan_ready').
+      // must use mcp__*__AskUserQuestion or render the prose fallback
+      // (both observed as outcome='asked').
       extraArgs: ['--disallowedTools', 'AskUserQuestion'],
+      trackTokens: SEED_PLAN_TOKENS,
       timeoutMs: CAPTURE_MS,
     });
 
@@ -101,28 +107,24 @@ describeE2E('plan-eng-review plan-mode smoke (periodic)', () => {
         );
       }
 
-      if (obs.outcome === 'plan_ready') {
-        if (!obs.planFile || !planFileHasDecisionsSection(obs.planFile)) {
-          throw new Error(
-            `STOP-gate regression: plan_ready without ## Decisions section in ` +
-              `${obs.planFile ?? '<no plan file>'} — gate skipped after ToolSearch.\n` +
-              `--- evidence (last 2KB) ---\n${obs.evidence}`,
-          );
-        }
+      assertReportAtBottomIfPlanWritten(obs);
+      assertNoPlanFileDecisions(obs, planFileHasDecisionsSection);
+      if (obs.outcome !== 'asked') {
+        throw new Error(
+          `STOP-gate regression: outcome=${obs.outcome} — the seeded findings reached ` +
+            `${obs.planFile ?? 'plan_ready'} without a question.\n` +
+            `--- evidence (last 2KB) ---\n${obs.evidence}`,
+        );
       }
 
-      expect(['asked', 'plan_ready']).toContain(obs.outcome);
-      assertReportAtBottomIfPlanWritten(obs);
-
       // Plan-mode scope-gate bypass: with a seeded plan in plan mode, the gate
-      // must NOT render its "What should I review?" menu — it auto-selects B
-      // and announces it. Exception ordering in the template (plan-mode branch
-      // first) makes this deterministic even though the seed arrives as a
-      // pasted user message. Unseeded test 1 keeps its lenient contract: with
-      // no plan drafted, the "ask as normal" fallback legitimately renders the
-      // question.
+      // must NOT render its "What should I review?" menu, and the review must
+      // target the seeded plan. The announcement's exact wording is not graded.
+      // Unseeded test 1 keeps its lenient contract: with no plan drafted, the
+      // "ask as normal" fallback legitimately renders the question.
       expect(obs.scopeGateQuestionObserved ?? false).toBe(false);
-      expect(obs.scopeGateAutoSelectObserved ?? false).toBe(true);
+      console.log(`[plan-eng plan-mode] scope announcement observed: ${obs.scopeGateAutoSelectObserved ?? false}; seed tokens: ${JSON.stringify(obs.tokensObserved ?? {})}`);
+      expect(seededPlanTargeted(obs, SEED_PLAN_TOKENS), 'seeded plan was not targeted (scopeGateAutoSelectObserved or seed-only plan tokens)').toBe(true);
     });
   }, CAPTURE_LONG_MS);
 });
diff --git a/test/skill-e2e-plan-format.test.ts b/test/skill-e2e-plan-format.test.ts
index 371365d66..07a262b8a 100644
--- a/test/skill-e2e-plan-format.test.ts
+++ b/test/skill-e2e-plan-format.test.ts
@@ -33,6 +33,7 @@ import { spawnSync } from 'child_process';
 import * as fs from 'fs';
 import * as path from 'path';
 import * as os from 'os';
+import { KIND_NOTE_RE } from './helpers/plan-format-kind-note';
 
 const evalCollector = createEvalCollector('e2e-plan-format');
 
@@ -49,7 +50,6 @@ const evalCollector = createEvalCollector('e2e-plan-format');
 // scripts/resolvers/preamble/generate-ask-user-format.ts. The optional
 // `[A-Z]=` prefix tolerates either shape; both are acceptable spec output.
 const COMPLETENESS_RE = /Completeness:\s*(?:[A-Z]=)?\d{1,2}\/10/;
-const KIND_NOTE_RE = /options differ in kind/i;
 
 // v1.7.0.0 Pros/Cons format tokens. Tests are additive: existing
 // RECOMMENDATION / Completeness / kind-note assertions still hold; new
@@ -157,8 +157,9 @@ After writing the file, stop. Do not continue the review.`,
         const captured = fs.readFileSync(outFile, 'utf-8');
         expect(captured.length).toBeGreaterThan(100);
 
-        // Kind-differentiated: Completeness: N/10 must NOT appear, "options differ
-        // in kind" note must appear. Recommendation presence is checked by the judge.
+        // Kind-differentiated: Completeness: N/10 must NOT appear, and the brief
+        // says the options are not comparable on coverage (any wording).
+        // Recommendation presence is checked by the judge.
         expect(captured).not.toMatch(COMPLETENESS_RE);
         expect(captured).toMatch(KIND_NOTE_RE);
 
@@ -359,8 +360,9 @@ After writing the file with that ONE question, stop. Do not continue the review.
         const captured = fs.readFileSync(outFile, 'utf-8');
         expect(captured.length).toBeGreaterThan(100);
 
-        // Kind-differentiated: Completeness: N/10 must NOT appear, "options differ
-        // in kind" note must appear. Recommendation presence checked by the judge.
+        // Kind-differentiated: Completeness: N/10 must NOT appear, and the brief
+        // says the options are not comparable on coverage (any wording).
+        // Recommendation presence checked by the judge.
         expect(captured).not.toMatch(COMPLETENESS_RE);
         expect(captured).toMatch(KIND_NOTE_RE);
 
diff --git a/test/skill-e2e-plan.test.ts b/test/skill-e2e-plan.test.ts
index d9aa25f84..cb91b20f8 100644
--- a/test/skill-e2e-plan.test.ts
+++ b/test/skill-e2e-plan.test.ts
@@ -1,6 +1,7 @@
 import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
 import { JUDGE_MS, CAPTURE_MS, CAPTURE_LONG_MS, PTY_MS } from './helpers/eval-budgets';
 import { runSkillTest } from './helpers/session-runner';
+import { getHermeticDirs, isHermeticEnabled, seedHermeticGstackHome } from './helpers/hermetic-env';
 import { EvalCollector } from './helpers/eval-store';
 import { OFFICE_HOURS_BUN_GRACE_MS, runRecordedOfficeHoursAttempt } from './helpers/office-hours-attempt';
 import {
@@ -379,6 +380,9 @@ Focus on architecture, code quality, tests, and performance sections.`,
 
 describeIfSelected('Plan-Eng-Review Test-Plan Artifact E2E', ['plan-eng-review-artifact'], () => {
   let planDir: string;
+  let gstackHome: string;
+  let ownsGstackHome = false;
+  let slug: string;
   let projectDir: string;
 
   beforeAll(() => {
@@ -438,35 +442,38 @@ export function main() { return Dashboard(); }
     // Set up remote-slug shim and browse shims (plan-eng-review uses remote-slug for artifact path)
     setupBrowseShims(planDir);
 
-    // Create project directory for artifacts
-    projectDir = path.join(os.homedir(), '.gstack', 'projects', 'test-project');
-    fs.mkdirSync(projectDir, { recursive: true });
-
-    // Clean up stale test-plan files from previous runs
-    try {
-      const staleFiles = fs.readdirSync(projectDir).filter(f => f.includes('test-plan'));
-      for (const f of staleFiles) {
-        fs.unlinkSync(path.join(projectDir, f));
-      }
-    } catch {}
+    // The actor, this observer and cleanup share one isolated state root: the
+    // hermetic run's GSTACK_HOME (or a seeded temporary one when hermetic mode
+    // is off), and the slug gstack-slug resolves for this fixture. Nothing is
+    // read or deleted under the operator's real home.
+    if (isHermeticEnabled()) gstackHome = getHermeticDirs().gstackHome;
+    else {
+      gstackHome = fs.mkdtempSync(path.join(os.tmpdir(), 'skill-e2e-plan-artifact-home-'));
+      seedHermeticGstackHome(gstackHome);
+      ownsGstackHome = true;
+    }
+    const slugOut = spawnSync(path.join(ROOT, 'bin', 'gstack-slug'), [], {
+      cwd: planDir, stdio: 'pipe', timeout: 5000, env: { ...process.env, GSTACK_HOME: gstackHome },
+    }).stdout.toString();
+    slug = /^SLUG=(.+)$/m.exec(slugOut)?.[1]?.trim() ?? '';
+    if (!slug) throw new Error(`plan-eng-review-artifact: gstack-slug resolved no slug: ${slugOut}`);
+    projectDir = path.join(gstackHome, 'projects', slug);
+    const realHome = path.resolve(os.homedir());
+    if (projectDir === realHome || projectDir.startsWith(realHome + path.sep) && !projectDir.startsWith(path.resolve(os.tmpdir()) + path.sep)) {
+      throw new Error(`plan-eng-review-artifact: state root ${projectDir} is under the real home directory`);
+    }
   });
 
   afterAll(() => {
     try { fs.rmSync(planDir, { recursive: true, force: true }); } catch {}
-    // Clean up test-plan artifacts (but not the project dir itself)
-    try {
-      const files = fs.readdirSync(projectDir);
-      for (const f of files) {
-        if (f.includes('test-plan')) {
-          fs.unlinkSync(path.join(projectDir, f));
-        }
-      }
-    } catch {}
+    // Remove only this fixture's project directory inside the isolated state root.
+    try { fs.rmSync(ownsGstackHome ? gstackHome : projectDir, { recursive: true, force: true }); } catch {}
   });
 
   testConcurrentIfSelected('plan-eng-review-artifact', async () => {
-    // Count existing test-plan files before
-    const beforeFiles = fs.readdirSync(projectDir).filter(f => f.includes('test-plan'));
+    const testPlans = () => fs.existsSync(projectDir)
+      ? fs.readdirSync(projectDir).filter(f => /eng-review-test-plan-.*\.md$/.test(f)) : [];
+    const beforeFiles = testPlans();
 
     const result = await runSkillTest({
       prompt: `Read plan-eng-review/SKILL.md for the review workflow.
@@ -476,10 +483,11 @@ Read plan.md — that's the plan to review. This is a standalone plan with sourc
 
 Proceed directly to the full review. Skip any AskUserQuestion calls — this is non-interactive.
 
-IMPORTANT: After your review, you MUST write the test-plan artifact as described in the "Test Plan Artifact" section of SKILL.md. The remote-slug shim is at ${planDir}/browse/bin/remote-slug.
+In this fixture the gstack state root is GSTACK_HOME=${gstackHome} and the project slug is ${slug}.
 
 Write your review to ${planDir}/review-output.md`,
       workingDirectory: planDir,
+      env: { GSTACK_HOME: gstackHome },
       maxTurns: 25,
       allowedTools: ['Bash', 'Read', 'Write', 'Glob', 'Grep'],
       timeout: CAPTURE_LONG_MS,
@@ -489,30 +497,19 @@ Write your review to ${planDir}/review-output.md`,
     });
 
     logCost('/plan-eng-review artifact', result);
+    // The QA test plan is a required discovery artifact of the review: exactly
+    // one new file under this fixture's project directory, about this plan.
+    const newFiles = testPlans().filter(f => !beforeFiles.includes(f));
+    const content = newFiles.length === 1 ? fs.readFileSync(path.join(projectDir, newFiles[0]!), 'utf-8') : '';
+    const aboutPlan = /dashboard|fetchStats|\/api\/stats/i.test(content);
+    console.log(`Test-plan artifacts in ${projectDir}: ${newFiles.length} new${newFiles.length ? ` (${newFiles[0]}, ${content.length} chars)` : ''}`);
     recordE2E(evalCollector, '/plan-eng-review test-plan artifact', 'Plan-Eng-Review Test-Plan Artifact E2E', result, {
-      passed: ['success', 'error_max_turns'].includes(result.exitReason),
+      passed: ['success', 'error_max_turns'].includes(result.exitReason) && newFiles.length === 1 && aboutPlan,
     });
 
     expect(['success', 'error_max_turns']).toContain(result.exitReason);
-
-    // Verify test-plan artifact was written
-    const afterFiles = fs.readdirSync(projectDir).filter(f => f.includes('test-plan'));
-    const newFiles = afterFiles.filter(f => !beforeFiles.includes(f));
-    console.log(`Test-plan artifacts: ${beforeFiles.length} before, ${afterFiles.length} after, ${newFiles.length} new`);
-
-    if (newFiles.length > 0) {
-      const content = fs.readFileSync(path.join(projectDir, newFiles[0]), 'utf-8');
-      console.log(`Test-plan artifact (${newFiles[0]}): ${content.length} chars`);
-      expect(content.length).toBeGreaterThan(50);
-    } else {
-      console.warn('No test-plan artifact found — agent may not have followed artifact instructions');
-    }
-
-    // Soft assertion: we expect an artifact but agent compliance is not guaranteed.
-    // Log rather than fail — the test-plan artifact is a bonus output, not the core test.
-    if (newFiles.length === 0) {
-      console.warn('SOFT FAIL: No test-plan artifact written — agent did not follow artifact instructions');
-    }
+    expect(newFiles, `expected one new eng-review test plan in ${projectDir}`).toHaveLength(1);
+    expect(aboutPlan, 'the test plan covers the reviewed dashboard change').toBe(true);
   }, CAPTURE_LONG_MS);
 });
 
@@ -717,9 +714,7 @@ Read plan.md — that's the plan to review. This is a standalone plan document,
 Proceed directly to the full review. Skip any AskUserQuestion calls — this is non-interactive.
 Skip the preamble bash block, lake intro, telemetry, and contributor mode sections.
 
-CRITICAL REQUIREMENT: plan.md IS the plan file for this review session. After completing your review, you MUST write a "## GSTACK REVIEW REPORT" section to the END of plan.md, exactly as described in the "Plan File Review Report" section of plan-eng-review/sections/review-sections.md. Use that canonical table, with all five review rows and honest not-run entries when review history is unavailable. The report MUST end with the mandatory unresolved-decisions status as its final line — the exact unbolded line NO UNRESOLVED DECISIONS when nothing is open, or a "**UNRESOLVED DECISIONS:**" block of bullets when items remain. Nothing may follow it. Use the Edit tool to append to plan.md — do NOT overwrite the existing plan content.
-
-This review report at the bottom of the plan is the MOST IMPORTANT deliverable of this test.`,
+plan.md is the plan file for this review session.`,
             workingDirectory: planDir,
             maxTurns: 20,
             timeout: CAPTURE_LONG_MS,
diff --git a/test/skill-e2e-shared-libs-periodic.test.ts b/test/skill-e2e-shared-libs-periodic.test.ts
index a93037efc..9b65c092f 100644
--- a/test/skill-e2e-shared-libs-periodic.test.ts
+++ b/test/skill-e2e-shared-libs-periodic.test.ts
@@ -10,7 +10,7 @@ import { createSharedPlanReuseSelector } from './helpers/shared-libs-plan-actor'
 import {
   SHARED_LIBS_ROOT, commitFixture, createSharedLibsFixture, fixtureWrite, installSourceShims,
   readRequests, runSharedCapture, runSharedInteractive, seedOpportunitySources,
-  sharedReadOnlyViolations, snapshotFixture, standaloneInstructions, toolCommandTrace, incompleteFirstFileView, type SharedLibsFixture,
+  sharedReadOnlyViolations, snapshotFixture, standaloneInstructions, toolCommandTrace, prCoverageRequestViolations, type SharedLibsFixture,
   SharedCaptureAccumulator, type SharedCaptureAttempt,
 } from './helpers/shared-libs-eval-fixture';
 
@@ -152,25 +152,9 @@ describeE2E('Shared-code opportunity and coordination judgment (periodic)', () =
         `Run /deslop-shared-libs using ${instructions}. Recent PR 7 mentions https://github.com/fixture/shared-libs/pull/42 as related work. Return the report after checking coordination within the skill's budget.`, attempt), async result => {
         assertReadOnly(f, before, result, 'shared-libs-pr-coverage');
         const requests = readRequests(f).filter(row => row.tool === 'gh' || row.tool === 'curl');
-        const endpoints = requests.map(row => row.endpoint || '');
-        expect(endpoints.some(endpoint => /\/pulls\/42\/files/.test(endpoint))).toBe(true);
-        expect(endpoints.some(endpoint => /\/pulls\/42\/files/.test(endpoint) && /[?&]page=2(?:&|$)/.test(endpoint))).toBe(true);
-        const openPages = endpoints.filter(endpoint => /\/pulls\?/.test(endpoint) && /state=open/.test(endpoint));
-        const filePages = endpoints.filter(endpoint => /\/pulls\/\d+\/files/.test(endpoint));
-        expect(openPages.length).toBeLessThanOrEqual(5);
-        // Only recent PR7's one page is outside the additional scan. Known old PR42
-        // shares that scan's 50-page budget, including both of its file-list pages.
-        expect(filePages.length).toBeLessThanOrEqual(51);
-        const coordinationPages = endpoints.filter(endpoint => /\/pulls\/42\/files/.test(endpoint));
-        // An incomplete first page is not a reusable complete file set. The
-        // observed capture truncated its first response with head, then fetched
-        // full pages 1–3. Permit that one recovery while charging every request
-        // to the hard budget and forbidding repeated complete first-page reads.
-        const truncatedFirstView = incompleteFirstFileView(result, 42);
-        expect(coordinationPages.length).toBeLessThanOrEqual(truncatedFirstView ? 4 : 3);
-        const firstPages = coordinationPages.filter(endpoint => !/[?&]page=/.test(endpoint) || /[?&]page=1(?:&|$)/.test(endpoint));
-        expect(firstPages.length).toBeGreaterThan(0);
-        expect(firstPages.length).toBeLessThanOrEqual(truncatedFirstView ? 2 : 1);
+        // The skill's stated page budget and PR 42 file-set coverage; per-PR
+        // page ceilings tuned to one observed run are not graded.
+        expect(prCoverageRequestViolations(requests.map(row => row.endpoint || ''))).toEqual([]);
         await assertJudgment(result.output, {
           older_open_pr: 'Identifies open PR 42 from 2020 as existing coordination work despite its activity being outside the 14-day window. The PR already proposes migrating retry-worker to the proven parser.',
           partial_overlap: 'Does not count the worker migration already covered by PR 42 as a new opportunity. It may recommend the independent remaining retry-route migration if it explains the narrowed scope and adjusted savings.',
diff --git a/test/skill-e2e-skillify.test.ts b/test/skill-e2e-skillify.test.ts
index d7920a2e7..27f2af642 100644
--- a/test/skill-e2e-skillify.test.ts
+++ b/test/skill-e2e-skillify.test.ts
@@ -42,6 +42,8 @@ import { spawnSync } from 'child_process';
 import * as fs from 'fs';
 import * as path from 'path';
 import * as os from 'os';
+import { readShippedSkillRouting } from './helpers/shipped-skill-routing';
+import { hasNarrationLeak } from './helpers/skill-body-narration';
 
 const evalCollector = createEvalCollector('e2e-skillify');
 
@@ -60,7 +62,7 @@ interface Workdir {
  *   - bin/ scripts referenced by the preamble
  *   - A scoped GSTACK_HOME under the workdir so on-disk artifacts are
  *     contained and assertable
- *   - A CLAUDE.md routing block instructing Skill-tool invocation
+ *   - A CLAUDE.md with the ## Skill routing section gstack ships
  *
  * `installSkills` lets each test pick the minimum surface (e.g., the
  * provenance-refusal scenario doesn't need /scrape).
@@ -109,16 +111,10 @@ function setupSkillifyWorkdir(suffix: string, installSkills: string[] = ['scrape
 
   fs.writeFileSync(path.join(workDir, 'CLAUDE.md'), `# Project Instructions
 
-## Skill routing
+${readShippedSkillRouting().section}
 
-When the user's request matches an available skill, ALWAYS invoke it via
-the Skill tool as your FIRST action.
+## Test environment
 
-Key routing rules:
-- /scrape, "scrape", "get data from", "extract from" → invoke scrape
-- /skillify, "skillify", "codify this scrape" → invoke skillify
-
-Environment:
 - GSTACK_HOME="${gstackHome}" for all gstack bin scripts.
 - bin scripts are at ./bin/ relative to this directory.
 - Browse binary is at ${browseBin} — assign to $B (e.g., \`B=${browseBin}\`).
@@ -196,13 +192,8 @@ describeIfSelected('Browser-skills Phase 2a E2E (/scrape + /skillify)', [
     installBundledHackernewsSkill(workDir);
 
     const result = await runSkillTest({
-      prompt: `Run /scrape latest hacker news stories. Invoke /scrape via the Skill tool.
-You MUST follow the skill's match-phase logic:
-1. Run \`$B skill list\` to see what browser-skills are available
-2. Recognize that "latest hacker news stories" matches the bundled
-   hackernews-frontpage skill's triggers
-3. Run \`$B skill run hackernews-frontpage\` and emit the JSON
-Do NOT enter the prototype phase. Do NOT use AskUserQuestion.`,
+      prompt: `Run /scrape latest hacker news stories.
+This run is non-interactive; AskUserQuestion is unavailable.`,
       workingDirectory: workDir,
       env: { GSTACK_HOME: gstackHome },
       maxTurns: 12,
@@ -214,18 +205,31 @@ Do NOT enter the prototype phase. Do NOT use AskUserQuestion.`,
 
     logCost('scrape-match-path', result);
 
+    // Outcome: the matcher ran the bundled skill instead of prototyping, and
+    // nothing new was written. Listing first is one valid way to match.
     const cmds = bashCommands(result);
-    const listedSkills = cmds.some(c => /\bskill\s+list\b/.test(c));
     const ranBundledSkill = cmds.some(c => /\bskill\s+run\s+hackernews-frontpage\b/.test(c));
+    const prototyped = cmds.some(c => /\$B\s+goto\b|\/browse\s+goto\b/.test(c));
+    const newSkills = [workDir, gstackHome].flatMap((root) => {
+      const skillsRoot = path.join(root, '.gstack', 'browser-skills');
+      return fs.existsSync(skillsRoot) ? fs.readdirSync(skillsRoot).filter(d => !d.startsWith('.') && d !== 'hackernews-frontpage') : [];
+    });
+    const staged = [workDir, gstackHome].some((root) => {
+      const stagingRoot = path.join(root, '.gstack', '.tmp');
+      return fs.existsSync(stagingRoot) && fs.readdirSync(stagingRoot).some(d => d.startsWith('skillify-'));
+    });
+    console.log(`scrape-match-path: listed skills first = ${cmds.some(c => /\bskill\s+list\b/.test(c))}`);
     const exitOk = ['success', 'error_max_turns'].includes(result.exitReason);
 
     recordE2E(evalCollector, 'scrape match-path routes to bundled skill', 'Phase 2a E2E', result, {
-      passed: exitOk && listedSkills && ranBundledSkill,
+      passed: exitOk && ranBundledSkill && !prototyped && newSkills.length === 0 && !staged,
     });
 
     expect(exitOk).toBe(true);
-    expect(listedSkills).toBe(true);
     expect(ranBundledSkill).toBe(true);
+    expect(prototyped, 'matched skill must run without prototyping').toBe(false);
+    expect(newSkills).toEqual([]);
+    expect(staged).toBe(false);
     try { fs.rmSync(workDir, { recursive: true, force: true }); } catch {}
   }, CAPTURE_MS);
 
@@ -240,13 +244,8 @@ Do NOT enter the prototype phase. Do NOT use AskUserQuestion.`,
 
     const result = await runSkillTest({
       prompt: `Run /scrape titles and scores from ${fileUrl}.
-Invoke /scrape via the Skill tool. Follow the skill's prototype-phase logic:
-1. \`$B skill list\` finds NO matching skill
-2. Drive: \`$B goto ${fileUrl}\` then \`$B html\` (or \`$B text\`)
-3. Parse the items (each has a title and a score)
-4. Emit JSON of the form {"items": [{"title": "...", "score": N}, ...], "count": N}
-5. Suggest /skillify in one line
-Do NOT use AskUserQuestion.`,
+Return JSON of the form {"items": [{"title": "...", "score": N}, ...], "count": N}.
+This run is non-interactive; AskUserQuestion is unavailable.`,
       workingDirectory: workDir,
       env: { GSTACK_HOME: gstackHome },
       maxTurns: 18,
@@ -351,16 +350,13 @@ Do NOT halt for clarification.`,
       && fs.existsSync(path.join(skillDir, '_lib', 'browse-client.ts'))
       && fs.existsSync(path.join(skillDir, 'fixtures'));
 
-    // D2 enforcement: the SKILL.md prose body MUST NOT contain conversation
-    // fragments. Cheap heuristic: it shouldn't have "I" or "Let me" or other
-    // first-person/agent-narration markers.
+    // D2 enforcement: the SKILL.md prose body must not contain conversation
+    // fragments (stored good/bad bodies: skill-body-narration.test.ts).
     let prosesClean = false;
     if (hasAllFiles) {
       const skillMd = fs.readFileSync(path.join(skillDir, 'SKILL.md'), 'utf-8');
       const body = skillMd.split(/\n---\n/)[1] || '';
-      prosesClean = !/^I /m.test(body)
-        && !/Let me /i.test(body)
-        && !/^I'll /m.test(body);
+      prosesClean = !hasNarrationLeak(body);
     }
 
     const exitOk = ['success', 'error_max_turns'].includes(result.exitReason);
@@ -390,11 +386,8 @@ Do NOT halt for clarification.`,
     fs.mkdirSync(childHome, { recursive: true });
 
     const result = await runSkillTest({
-      prompt: `Run /skillify via the Skill tool. There has been NO prior /scrape
-in this conversation. Follow the skill's Step 1 (D1 provenance guard) literally:
-walk back through agent turns, find no /scrape result, refuse with the exact
-message the skill specifies, and stop. Do NOT synthesize anything. Do NOT
-write any files.`,
+      prompt: `Run /skillify.
+This run is non-interactive; AskUserQuestion is unavailable.`,
       workingDirectory: workDir,
       env: {
         GSTACK_HOME: gstackHome,
diff --git a/test/skill-e2e-test-value.test.ts b/test/skill-e2e-test-value.test.ts
index fe3b7ef51..af6d1b3a0 100644
--- a/test/skill-e2e-test-value.test.ts
+++ b/test/skill-e2e-test-value.test.ts
@@ -17,7 +17,7 @@ import { ROOT, runId, describeIfSelected, testIfSelected, copyDirSync, logCost,
   createEvalCollector, finalizeEvalCollector } from './helpers/e2e-helpers';
 import { extractSkillBody } from './helpers/skill-fixture';
 import { createCoverageAuditFixture } from './fixtures/coverage-audit-fixture';
-import { createTestValueFixture, lastJsonLine, jsonFindings, LOW_VALUE_TESTS } from './helpers/test-value-fixture';
+import { createTestValueFixture, lastJsonLine, jsonFindings, LOW_VALUE_TESTS, recordsNonTestCallerSearch } from './helpers/test-value-fixture';
 import { runRecordedOfficeHoursAttempt, OFFICE_HOURS_BUN_GRACE_MS } from './helpers/office-hours-attempt';
 import { resolveEvalModel } from '../lib/eval-model';
 import { RETIREMENT_FIELDS } from '../scripts/resolvers/test-value';
@@ -85,7 +85,7 @@ findings, one per line.`,
       const seam = about(new RegExp(`pricing-reset|${LOW_VALUE_TESTS.testOnlySymbol}`));
       if (!grep.length || !grep.every(finding => finding.severity === 'INFORMATIONAL')) throw new Error(`review: source-grep test needs an INFORMATIONAL finding, got ${JSON.stringify(grep)}`);
       if (!seam.length || !seam.every(finding => finding.severity === 'INFORMATIONAL')) throw new Error(`review: test-only export needs an INFORMATIONAL finding, got ${JSON.stringify(seam)}`);
-      if (!seam.some(finding => /non_test_callers/.test(JSON.stringify(finding.evidence ?? '')) && /git grep/.test(JSON.stringify(finding.evidence ?? '')))) throw new Error('review: test-only export evidence must record non_test_callers and the git grep search');
+      if (!seam.some(recordsNonTestCallerSearch)) throw new Error('review: test-only export evidence must record non_test_callers and the caller search');
       const golden = about(/skill-golden/);
       if (golden.length) throw new Error(`review: the SKILL.md golden test must not be flagged, got ${JSON.stringify(golden)}`);
       if (trackedChanges(cwd)) throw new Error('review: tracked files changed');
diff --git a/test/skill-e2e-triage.test.ts b/test/skill-e2e-triage.test.ts
index 89f75db18..a20e86bfa 100644
--- a/test/skill-e2e-triage.test.ts
+++ b/test/skill-e2e-triage.test.ts
@@ -32,6 +32,7 @@ import { spawnSync } from 'child_process';
 import * as fs from 'fs';
 import * as path from 'path';
 import * as os from 'os';
+import { triageLabels } from './helpers/ship-triage-labels';
 
 const evalCollector = createEvalCollector('e2e-triage');
 
@@ -175,6 +176,7 @@ For each failing test, classify it as:
 Use git diff origin/main...HEAD (or git diff main...HEAD since there's no remote) to determine which files changed on this branch.
 
 Output your classification for each failure clearly, labeling each as "IN-BRANCH" or "PRE-EXISTING" with your reasoning.
+End with one JSON object mapping each failing test file to "in-branch" or "pre-existing".
 
 This is a solo repo (REPO_MODE=solo). For pre-existing failures, recommend fixing now.`,
       workingDirectory: triageDir,
@@ -188,49 +190,20 @@ This is a solo repo (REPO_MODE=solo). For pre-existing failures, recommend fixin
     logCost('/ship triage', result);
 
     const output = result.output || '';
-    const outputLower = output.toLowerCase();
-
-    // The triage should identify the string/truncate failure as in-branch
-    const hasInBranch = outputLower.includes('in-branch') || outputLower.includes('in branch') || outputLower.includes('introduced');
-    // The triage should identify the math/divide failure as pre-existing
-    const hasPreExisting = outputLower.includes('pre-existing') || outputLower.includes('pre existing') || outputLower.includes('existed before');
-
-    console.log(`Output identifies IN-BRANCH failures: ${hasInBranch}`);
-    console.log(`Output identifies PRE-EXISTING failures: ${hasPreExisting}`);
-
-    // Check that the string/truncate bug is classified as in-branch
-    const mentionsTruncate = outputLower.includes('truncate') || outputLower.includes('string');
-    const mentionsDivide = outputLower.includes('divide') || outputLower.includes('math');
-
-    console.log(`Mentions truncate/string (in-branch bug): ${mentionsTruncate}`);
-    console.log(`Mentions divide/math (pre-existing bug): ${mentionsDivide}`);
-
-    // Verify BOTH failure classes are exercised (not just detected):
-    // The test runner must have actually run both test files
-    const ranMathTest = output.includes('math.test') || output.includes('FAIL: divide');
-    const ranStringTest = output.includes('string.test') || output.includes('FAIL: truncate');
-    console.log(`Ran math test file (pre-existing failure): ${ranMathTest}`);
-    console.log(`Ran string test file (in-branch failure): ${ranStringTest}`);
+    // Outcome: each seeded failure carries the right label, and the suite actually ran.
+    const labels = triageLabels(output);
+    const ranTests = result.toolCalls.some(call => call.tool === 'Bash'
+      && /test\/run\.js|test\/(?:math|string)\.test\.js/.test(String(call.input?.command ?? '')));
+    console.log(`Triage labels: ${JSON.stringify(labels)}; ran tests: ${ranTests}`);
 
     recordE2E(evalCollector, '/ship triage', 'Test Failure Triage E2E', result, {
-      passed: result.exitReason === 'success' && hasInBranch && hasPreExisting,
-      has_in_branch_classification: hasInBranch,
-      has_pre_existing_classification: hasPreExisting,
-      mentions_truncate: mentionsTruncate,
-      mentions_divide: mentionsDivide,
-      ran_both_test_files: ranMathTest && ranStringTest,
+      passed: result.exitReason === 'success' && ranTests && labels.string === 'in-branch' && labels.math === 'pre-existing',
     });
 
     expect(result.exitReason).toBe('success');
-    // Must classify at least one failure as in-branch AND one as pre-existing
-    expect(hasInBranch).toBe(true);
-    expect(hasPreExisting).toBe(true);
-    // Must mention the specific bugs
-    expect(mentionsTruncate).toBe(true);
-    expect(mentionsDivide).toBe(true);
-    // Must have actually run both test files (exercises both failure classes)
-    expect(ranMathTest).toBe(true);
-    expect(ranStringTest).toBe(true);
+    expect(ranTests, 'the agent must run the test suite').toBe(true);
+    expect(labels.string, 'truncate/string failure is in-branch').toBe('in-branch');
+    expect(labels.math, 'divide/math failure is pre-existing').toBe('pre-existing');
   }, CAPTURE_MS);
 });
 
diff --git a/test/skill-e2e-workflow.test.ts b/test/skill-e2e-workflow.test.ts
index 9ffc14b76..178343635 100644
--- a/test/skill-e2e-workflow.test.ts
+++ b/test/skill-e2e-workflow.test.ts
@@ -24,6 +24,7 @@ const evalCollector = createEvalCollector('e2e');
 
 describeIfSelected('Document-Release skill E2E', ['document-release'], () => {
   let docReleaseDir: string;
+  let featureHead: string;
 
   beforeAll(() => {
     docReleaseDir = fs.mkdtempSync(path.join(os.tmpdir(), 'skill-e2e-doc-release-'));
@@ -61,6 +62,7 @@ describeIfSelected('Document-Release skill E2E', ['document-release'], () => {
       '# Changelog\n\n## 1.1.1 — 2026-03-16\n\n- Added Feature C\n\n## 1.0.0 — 2026-03-01\n\n- Initial release with Feature A and Feature B\n- Setup CI pipeline\n');
     run('git', ['add', '.']);
     run('git', ['commit', '-m', 'feat: add feature C']);
+    featureHead = run('git', ['rev-parse', 'HEAD']).stdout.toString().trim();
   });
 
   afterAll(() => {
@@ -73,14 +75,8 @@ describeIfSelected('Document-Release skill E2E', ['document-release'], () => {
 
 Run the /document-release workflow on this repo. The base branch is "main".
 
-IMPORTANT:
-- Do NOT use AskUserQuestion — auto-approve everything or skip if unsure.
-- Do NOT push or create PRs (there is no remote).
-- Do NOT run gh commands (no remote).
-- Focus on updating README.md to reflect the new Feature C.
-- Do NOT overwrite or regenerate CHANGELOG entries.
-- Skip VERSION bump (it's already bumped).
-- After editing, just commit the changes locally.`,
+This run is non-interactive: AskUserQuestion is unavailable and nobody can answer questions.
+The repo has no remote, so pushing, creating PRs and gh commands are unavailable.`,
       workingDirectory: docReleaseDir,
       maxTurns: 30,
       allowedTools: ['Bash', 'Read', 'Write', 'Edit', 'Grep', 'Glob'],
@@ -97,36 +93,31 @@ IMPORTANT:
 
     logCost('/document-release', result);
 
-    // Read CHANGELOG to verify it was NOT clobbered
+    // The prompt no longer states the CHANGELOG/VERSION rules, so these
+    // outcomes measure the skill's own never-clobber and never-bump gates.
     const changelog = fs.readFileSync(path.join(docReleaseDir, 'CHANGELOG.md'), 'utf-8');
-    const hasOriginalEntries = changelog.includes('Initial release with Feature A and Feature B')
-      && changelog.includes('Setup CI pipeline')
-      && changelog.includes('1.0.0');
-    if (!hasOriginalEntries) {
-      console.warn('CHANGELOG CLOBBERED — original entries missing!');
-    }
-
-    // Check if README was updated
+    const currentEntry = changelog.split(/^## 1\.1\.1\b/m)[1]?.split(/^## /m)[0] ?? '';
+    const hasOriginalEntries = /^## 1\.1\.1\b/m.test(changelog) && /^## 1\.0\.0\b/m.test(changelog)
+      && /feature[ -]?c/i.test(currentEntry)
+      && changelog.includes('- Initial release with Feature A and Feature B')
+      && changelog.includes('- Setup CI pipeline');
+    const version = fs.readFileSync(path.join(docReleaseDir, 'VERSION'), 'utf-8').trim();
+    const versionCommits = spawnSync('git', ['log', '--format=%H', `${featureHead}..HEAD`, '--', 'VERSION'],
+      { cwd: docReleaseDir, stdio: 'pipe', timeout: 5000 }).stdout.toString().trim();
     const readme = fs.readFileSync(path.join(docReleaseDir, 'README.md'), 'utf-8');
-    const readmeUpdated = readme.includes('Feature C') || readme.includes('feature-c') || readme.includes('feature C');
+    const readmeUpdated = /feature[ -]?c/i.test(readme);
 
     const exitOk = ['success', 'error_max_turns'].includes(result.exitReason);
     recordE2E(evalCollector, '/document-release', 'Document-Release skill E2E', result, {
-      passed: exitOk && hasOriginalEntries,
+      passed: exitOk && hasOriginalEntries && version === '1.1.1' && versionCommits === '' && readmeUpdated,
     });
 
-    // Critical guardrail: CHANGELOG must not be clobbered
-    expect(hasOriginalEntries).toBe(true);
-
+    expect(hasOriginalEntries, 'original CHANGELOG entries must remain').toBe(true);
+    expect(version, 'VERSION must not change without asking').toBe('1.1.1');
+    expect(versionCommits, 'no commit may change VERSION').toBe('');
+    expect(readmeUpdated, 'README should mention Feature C').toBe(true);
     // Accept error_max_turns — thorough doc review is not a failure
     expect(['success', 'error_max_turns']).toContain(result.exitReason);
-
-    // Informational: did it update README?
-    if (readmeUpdated) {
-      console.log('README updated to include Feature C');
-    } else {
-      console.warn('README was NOT updated — agent may not have found the feature');
-    }
   }, CAPTURE_LONG_MS);
 });
 
diff --git a/test/skill-fixture.test.ts b/test/skill-fixture.test.ts
index 6b1eba432..6a58562ff 100644
--- a/test/skill-fixture.test.ts
+++ b/test/skill-fixture.test.ts
@@ -28,6 +28,7 @@ import {
   CODEX_REVIEW_E2E_SECTIONS,
 } from './helpers/skill-fixture';
 import { E2E_TOUCHFILES, GLOBAL_TOUCHFILES, selectTests } from './helpers/touchfiles';
+import { readShippedSkillRouting } from './helpers/shipped-skill-routing';
 
 const ROOT = path.resolve(import.meta.dir, '..');
 
@@ -449,10 +450,12 @@ test('routing catalog contains installed project names without a request-to-skil
     writeFileSync: (file: string, content: string) => writes.set(file, content),
   };
   const compiled = new Bun.Transpiler({ loader: 'ts' }).transformSync(source.slice(start, end) + '\ninstallSkills(fixtureRoot);');
-  new Function('ROOT', 'fs', 'path', 'extractSkillHead', 'fixtureRoot', compiled)(
-    ROOT, mockFs, path, (file: string) => installed.get(file), fixtureRoot,
+  new Function('ROOT', 'fs', 'path', 'extractSkillHead', 'readShippedSkillRouting', 'fixtureRoot', compiled)(
+    ROOT, mockFs, path, (file: string) => installed.get(file), readShippedSkillRouting, fixtureRoot,
   );
   const instructions = writes.get(path.join(fixtureRoot, 'CLAUDE.md'))!;
+  expect(instructions).toContain(readShippedSkillRouting().instruction);
+  expect(instructions).not.toContain('→ invoke');
   expect(instructions).toContain('installed gstack skills: gstack, qa, review.');
   expect(instructions).toContain("built-in skills are outside this project's workflow");
   expect(instructions).toContain('matching the request to the skill descriptions');
@@ -464,3 +467,18 @@ test('routing catalog contains installed project names without a request-to-skil
     .toBe(installed.get(path.join(ROOT, 'review/SKILL.md')));
   expect(writes.size).toBe(4);
 });
+
+test('routing fixtures read the ## Skill routing section gstack ships', () => {
+  const routing = readShippedSkillRouting();
+  expect(routing.instruction).toStartWith('## Skill routing\n\n');
+  expect(routing.instruction).toContain('Skill tool');
+  expect(routing.rules.some(rule => rule.includes('/context-save'))).toBe(true);
+  expect(routing.rules.some(rule => rule.includes('/context-restore'))).toBe(true);
+  expect(routing.section).not.toContain('If B:');
+  const root = fs.mkdtempSync(path.join(os.tmpdir(), 'shipped-routing-drift-'));
+  try {
+    fs.mkdirSync(path.join(root, 'bin'));
+    fs.writeFileSync(path.join(root, 'bin', 'gstack-skill-start'), '#!/bin/bash\necho no routing here\n');
+    expect(() => readShippedSkillRouting(root)).toThrow('routing-injection heredoc not found');
+  } finally { fs.rmSync(root, { recursive: true, force: true }); }
+});
diff --git a/test/skill-llm-eval.test.ts b/test/skill-llm-eval.test.ts
index 09bb8119b..40a91a410 100644
--- a/test/skill-llm-eval.test.ts
+++ b/test/skill-llm-eval.test.ts
@@ -15,10 +15,10 @@ import { JUDGE_MS } from './helpers/eval-budgets';
 import * as fs from 'fs';
 import * as path from 'path';
 import { callJudge, judge, JudgeRefusalError, DEFAULT_JUDGE_MAX_TOKENS, judgePanel, judgePanelMean, judgePanelMajority, judgePanelReasoning, JUDGE_SCORE_DIMENSIONS } from './helpers/llm-judge';
-import { ENG_REVIEW_EXCERPT } from './helpers/workflow-excerpt';
+import { ASK_QUESTIONS_HEADING, ENG_REVIEW_EXCERPT } from './helpers/workflow-excerpt';
 import type { JudgeScore } from './helpers/llm-judge';
 import { readWorkflowJudgeInput, buildWorkflowJudgePrompt, QA_DISCOVERY_REFERENCES, WORKFLOW_JUDGE_RESPONSE_SCHEMA, type WorkflowJudgeInput } from './helpers/workflow-judge-input';
-import { prepareWorkflowJudgeCache, validWorkflowJudgeScore } from './helpers/workflow-judge-cache';
+import { BROWSE_JUDGE_FLOORS, browseJudgeFloorsMet, prepareWorkflowJudgeCache, validWorkflowJudgeScore } from './helpers/workflow-judge-cache';
 import { buildCookieWorkflowJudgeInput, COOKIE_WORKFLOW_JUDGE } from './helpers/cookie-workflow-judge-input';
 import { getCookieWorkflowManualReview, type ManualJudgeReview } from './helpers/cookie-workflow-manual-review';
 import { resolveEvalModel } from '../lib/eval-model';
@@ -96,8 +96,8 @@ describeIfSelected('LLM-as-judge quality evals', [
     const t0 = Date.now();
     // Browse carve: snapshot flags + the full command list are the whole
     // generated section file; one judge grades the union. Scores are also
-    // pinned against test/fixtures/eval-baselines.json (UPDATE_BASELINES=1
-    // rewrites the pin).
+    // compared with test/fixtures/eval-baselines.json and the difference is
+    // recorded (UPDATE_BASELINES=1 rewrites it); only the floors gate.
     const section = sliceBrowseSection('## Snapshot Flags');
 
     const samples = await judgePanel(() => judge('browse skill reference (flags + commands)', section));
@@ -119,17 +119,16 @@ describeIfSelected('LLM-as-judge quality evals', [
       name: 'browse/SKILL.md reference',
       suite: 'LLM-as-judge quality evals',
       tier: 'llm-judge',
-      passed: scores.clarity >= 3 && scores.completeness >= 4 && scores.actionability >= 4 && regressions.length === 0,
+      passed: browseJudgeFloorsMet(scores),
       duration_ms: Date.now() - t0,
       cost_usd: 0.02 * samples.length,
       judge_scores: { clarity: scores.clarity, completeness: scores.completeness, actionability: scores.actionability },
       judge_reasoning: regressions.length ? `${judgePanelReasoning(samples)} | ${regressions.join('; ')}` : judgePanelReasoning(samples),
     });
 
-    expect(scores.clarity).toBeGreaterThanOrEqual(3);
-    expect(scores.completeness).toBeGreaterThanOrEqual(4);
-    expect(scores.actionability).toBeGreaterThanOrEqual(4);
-    expect(regressions).toEqual([]);
+    expect(scores.clarity).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.clarity);
+    expect(scores.completeness).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.completeness);
+    expect(scores.actionability).toBeGreaterThanOrEqual(BROWSE_JUDGE_FLOORS.actionability);
   }, JUDGE_MS);
 
   testIfSelected('setup block', async () => {
@@ -426,7 +425,7 @@ async function runWorkflowJudge(opts: {
   suite: string;
   skillPath: string;
   startMarker: string;
-  endMarker: string | null;
+  endMarker: string | RegExp | null;
   references?: readonly string[];
   judgeContext: string;
   judgeGoal: string;
@@ -635,7 +634,7 @@ describeIfSelected('Plan Review skill evals', [
       suite: 'Plan Review skill evals',
       skillPath: 'plan-design-review/SKILL.md',
       startMarker: '## Review Sections',
-      endMarker: '## CRITICAL RULE',
+      endMarker: ASK_QUESTIONS_HEADING,
       judgeContext: 'a design plan review framework with 7 review passes',
       judgeGoal: 'how to review a plan for design quality using a 0-10 rating method: rate each dimension, explain what a 10 looks like, edit the plan to fix gaps, then re-rate',
     });
diff --git a/test/skill-routing-e2e.test.ts b/test/skill-routing-e2e.test.ts
index a9a2d99cb..2bdc68c31 100644
--- a/test/skill-routing-e2e.test.ts
+++ b/test/skill-routing-e2e.test.ts
@@ -6,6 +6,7 @@ import { EvalCollector } from './helpers/eval-store';
 import type { EvalTestEntry } from './helpers/eval-store';
 import { selectTests, detectBaseBranch, getChangedFiles, E2E_TOUCHFILES, E2E_TIERS, GLOBAL_TOUCHFILES } from './helpers/touchfiles';
 import { extractSkillHead } from './helpers/skill-fixture';
+import { readShippedSkillRouting } from './helpers/shipped-skill-routing';
 import { spawnSync } from 'child_process';
 import * as fs from 'fs';
 import * as path from 'path';
@@ -94,26 +95,19 @@ function installSkills(tmpDir: string) {
 
   // The names-only catalog keeps new CLI built-ins from changing the candidate
   // set. Descriptions still choose the skill; no request-to-skill answer key.
-  // Write a CLAUDE.md with a GENERIC invoke-skills nudge — deliberately NO
-  // per-skill routing table. These journey tests exist to catch skill
-  // DESCRIPTION regressions (their touchfiles key on */SKILL.md.tmpl), and
-  // the old fixture shipped an explicit prompt→skill answer key: with the
-  // lookup table in context, a badly regressed frontmatter description
-  // still routed correctly and the tests could not fail on the regression
-  // class they select for (2026-08 audit). The generic nudge keeps Claude's
-  // reach-for-a-skill posture; the FRONTMATTER carries the routing load.
+  // The routing instruction is the one gstack ships (bin/gstack-skill-start),
+  // without its per-skill rule list. These journey tests exist to catch skill
+  // DESCRIPTION regressions (their touchfiles key on */SKILL.md.tmpl): with a
+  // lookup table in context, a badly regressed frontmatter description still
+  // routes correctly and the tests cannot fail on the regression class they
+  // select for. The FRONTMATTER carries the routing load.
   fs.writeFileSync(path.join(tmpDir, 'CLAUDE.md'), `# Project Instructions
 
-## Skill routing
+${readShippedSkillRouting().instruction}
 
 This project uses the following installed gstack skills: ${installedSkills.join(', ')}.
 Choose among this project catalog by matching the request to the skill descriptions.
 The CLI's built-in skills are outside this project's workflow.
-
-When the user's request matches an available project skill, ALWAYS invoke it using the Skill
-tool as your FIRST action. Do NOT answer directly, do NOT use other tools first.
-The skill has specialized workflows that produce better results than ad-hoc answers.
-Choose the skill by matching the request against each skill's description.
 `);
 }
 
diff --git a/test/test-value-seam-evidence.test.ts b/test/test-value-seam-evidence.test.ts
new file mode 100644
index 000000000..bf031ee25
--- /dev/null
+++ b/test/test-value-seam-evidence.test.ts
@@ -0,0 +1,16 @@
+import { describe, expect, test } from 'bun:test';
+import * as fs from 'fs';
+import * as path from 'path';
+import { recordsNonTestCallerSearch } from './helpers/test-value-fixture';
+
+const stored = JSON.parse(fs.readFileSync(path.join(import.meta.dir, 'fixtures/test-value-seam-evidence.json'), 'utf8')) as
+  { known_good: Record<string, unknown>; known_bad: Record<string, unknown> };
+
+describe('test-only export caller-search evidence', () => {
+  test.each(Object.entries(stored.known_good))('accepts %s', (_name, evidence) => {
+    expect(recordsNonTestCallerSearch({ evidence })).toBe(true);
+  });
+  test.each(Object.entries(stored.known_bad))('rejects %s', (_name, evidence) => {
+    expect(recordsNonTestCallerSearch({ evidence })).toBe(false);
+  });
+});
diff --git a/test/touchfiles.test.ts b/test/touchfiles.test.ts
index 2bb515a88..a766d0ab8 100644
--- a/test/touchfiles.test.ts
+++ b/test/touchfiles.test.ts
@@ -19,7 +19,7 @@ import {
 } from './helpers/touchfiles';
 
 import { paidTestClosure, isCovered } from './helpers/touchfile-closure';
-import { readWorkflowExcerpt } from './helpers/workflow-excerpt';
+import { ASK_QUESTIONS_HEADING, readWorkflowExcerpt } from './helpers/workflow-excerpt';
 import { sharedLibsPlanExcerpt } from './helpers/shared-libs-plan-excerpt';
 
 const ROOT = path.resolve(import.meta.dir, '..');
@@ -216,7 +216,7 @@ describe('selectTests', () => {
     ['ship/sections/test-coverage.md', 'ship/SKILL.md workflow',
       'ship/SKILL.md', '# Ship:', '## Important Rules', '### REGRESSION RULE (mandatory)'],
     ['plan-design-review/sections/review-sections.md', 'plan-design-review/SKILL.md passes',
-      'plan-design-review/SKILL.md', '## Review Sections', '## CRITICAL RULE',
+      'plan-design-review/SKILL.md', '## Review Sections', ASK_QUESTIONS_HEADING,
       '## Review Sections (7 passes, after scope is agreed)'],
   ])('expanded judge content remains selected by its section alone: %s', (file, judge, skill, start, end, marker) => {
     const body = fs.readFileSync(path.join(ROOT, file), 'utf8').replace(/^<!--[^\n]*-->\n/gm, '').trim();
diff --git a/test/workflow-boundaries-fixture.test.ts b/test/workflow-boundaries-fixture.test.ts
index 0db8cfb06..6a860d6af 100644
--- a/test/workflow-boundaries-fixture.test.ts
+++ b/test/workflow-boundaries-fixture.test.ts
@@ -28,7 +28,7 @@ for (const attack of ['read-state', 'cat-state', 'bash-edit', 'receipt-write'])
   } finally { fs.rmSync(fixture.root, { recursive: true, force: true }); }
 });
 
-function protocolControl(id: BoundaryCase, faults: { skipCleanup?: boolean; duplicateQuestion?: boolean; failVerification?: boolean; rateLimit?: boolean; undeclared?: string; deniedBash?: string; includeAvailability?: boolean; misleadingFinal?: boolean; splitFinal?: boolean } = {}) {
+function protocolControl(id: BoundaryCase, faults: { skipCleanup?: boolean; duplicateQuestion?: boolean; failVerification?: boolean; rateLimit?: boolean; undeclared?: string; deniedBash?: string; includeAvailability?: boolean; misleadingFinal?: boolean; splitFinal?: boolean; reverify?: 'pass' | 'fail-last' } = {}) {
   let directory = '';
   let calls = 0;
   const provider: QueryProvider = input => {
@@ -119,6 +119,13 @@ function protocolControl(id: BoundaryCase, faults: { skipCleanup?: boolean; dupl
           }
           const verified = await shell('bash ./verify.sh');
           expect(verified.status).toBe(id === 'investigate-owned-ending-error' ? 69 : faults.failVerification ? 1 : 0);
+          if (faults.reverify) {
+            if (faults.reverify === 'fail-last') {
+              const file = path.join(directory, 'verify.sh');
+              fs.writeFileSync(file, fs.readFileSync(file, 'utf8').replace('value() !== 2', 'value() !== 3'));
+            }
+            await shell('bash ./verify.sh');
+          }
         } else expect(answer).toBe('Abort');
         if (!faults.skipCleanup) expect((await shell(`bash "$HOME/.claude/skills/gstack/freeze/bin/freeze-state.sh" release "${owner}"`)).status).toBe(0);
         const finalText = faults.misleadingFinal ? 'The fix is complete and verification succeeded.'
@@ -212,6 +219,27 @@ test('fixture records failed verification rather than crediting successful clean
   expect(JSON.parse(records[0].output!).evidence.receipts).toContain('VERIFY_STATUS:1');
 });
 
+test('a passing re-verification is not a protocol failure', async () => {
+  const driver = protocolControl('investigate-owned-completion', { reverify: 'pass' });
+  const records: EvalTestEntry[] = [];
+  await runBoundaryActor('investigate-owned-completion', entry => records.push(entry), driver.provider);
+  expect(records[0].passed).toBe(true);
+  expect(JSON.parse(records[0].output!).evidence.receipts.match(/^VERIFY$/gm)).toHaveLength(2);
+});
+
+test('a failing final re-verification still fails after an earlier success', async () => {
+  const driver = protocolControl('investigate-owned-completion', { reverify: 'fail-last' });
+  const records: EvalTestEntry[] = [];
+  await expect(runBoundaryActor('investigate-owned-completion', entry => records.push(entry), driver.provider)).rejects.toThrow('final verification did not succeed');
+  expect(records[0].passed).toBe(false);
+});
+
+test('retrying the verifier after its ending error still fails', async () => {
+  const driver = protocolControl('investigate-owned-ending-error', { reverify: 'pass' });
+  const records: EvalTestEntry[] = [];
+  await expect(runBoundaryActor('investigate-owned-ending-error', entry => records.push(entry), driver.provider)).rejects.toThrow('without retry');
+});
+
 test('rate-limit retry restores the fixture and retains both native event streams', async () => {
   const driver = protocolControl('investigate-owned-completion', { rateLimit: true });
   const records: EvalTestEntry[] = [];
diff --git a/test/workflow-excerpt.test.ts b/test/workflow-excerpt.test.ts
index 50d0090aa..0f93be40d 100644
--- a/test/workflow-excerpt.test.ts
+++ b/test/workflow-excerpt.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, test } from 'bun:test';
-import { ENG_REVIEW_EXCERPT, readWorkflowExcerpt } from './helpers/workflow-excerpt';
+import { ASK_QUESTIONS_HEADING, ENG_REVIEW_EXCERPT, readWorkflowExcerpt } from './helpers/workflow-excerpt';
 import { mkdtempSync, readFileSync, rmSync, writeFileSync } from 'fs';
 import { tmpdir } from 'os';
 import { join } from 'path';
@@ -106,9 +106,9 @@ describe('workflow judge excerpts', () => {
   });
 
   test('a sliced section is not appended again with its generated header', () => {
-    const text = readWorkflowExcerpt('plan-design-review/SKILL.md', '## Review Sections', '## CRITICAL RULE');
+    const text = readWorkflowExcerpt('plan-design-review/SKILL.md', '## Review Sections', ASK_QUESTIONS_HEADING);
     expect(text.match(/## Review Sections/g)).toHaveLength(1);
-    expect(text).not.toContain('## CRITICAL RULE');
+    expect(text).not.toMatch(ASK_QUESTIONS_HEADING);
     expect(text).not.toContain('AUTO-GENERATED');
   });
 
@@ -293,7 +293,7 @@ console.log(JSON.stringify({calls, results}));
     expect(pendingDecision).toContain("**STOP until the actual answer arrives.**");
     expect(pendingDecision.replace(/\s+/g, ' ')).toContain("Do not apply a remedy, make another call, start the next section or call ExitPlanMode while the choice awaits an answer");
     expect(eng.replace(/\s+/g, ' ')).toContain("Use a scoped Edit to save this record and only the authorized working-plan amendments. Leave other choices unchanged");
-    const design = readWorkflowExcerpt('plan-design-review/SKILL.md', '## Review Sections', '## CRITICAL RULE');
+    const design = readWorkflowExcerpt('plan-design-review/SKILL.md', '## Review Sections', ASK_QUESTIONS_HEADING);
     expect(design).toContain('wait for approval, then edit the plan and re-rate');
     const pass4 = design.slice(design.indexOf('### Pass 4:'), design.indexOf('### Pass 5:'));
     expect(pass4.match(/^### /gm)).toHaveLength(1);
diff --git a/test/workflow-judge-input.test.ts b/test/workflow-judge-input.test.ts
index 47ff4b3a7..4b47d50da 100644
--- a/test/workflow-judge-input.test.ts
+++ b/test/workflow-judge-input.test.ts
@@ -5,7 +5,7 @@ import { tmpdir } from 'node:os';
 import { dirname, join, resolve } from 'node:path';
 import { createHash } from 'node:crypto';
 import { readWorkflowJudgeInput, buildWorkflowJudgePrompt, QA_DISCOVERY_REFERENCES, WORKFLOW_JUDGE_RESPONSE_SCHEMA, WORKFLOW_JUDGE_REASONING_WORD_LIMIT } from './helpers/workflow-judge-input';
-import { ENG_REVIEW_EXCERPT } from './helpers/workflow-excerpt';
+import { ASK_QUESTIONS_HEADING, ENG_REVIEW_EXCERPT } from './helpers/workflow-excerpt';
 
 const ROOT = resolve(import.meta.dir, '..');
 const scratchRoots: string[] = [];
@@ -363,14 +363,14 @@ describe('workflow judge file bundle', () => {
 
   test('generated plan-design passes retain their full section without duplicating Pass 1', () => {
     const input = readWorkflowJudgeInput({
-      root: ROOT, skillPath: 'plan-design-review/SKILL.md', startMarker: '## Review Sections', endMarker: '## CRITICAL RULE',
+      root: ROOT, skillPath: 'plan-design-review/SKILL.md', startMarker: '## Review Sections', endMarker: ASK_QUESTIONS_HEADING,
     });
     expect(input.files.filter(file => file.kind === 'entrypoint')).toHaveLength(0);
     expect(input.files.map(file => file.path)).toEqual(sectionPaths('plan-design-review'));
     const section = input.files.find(file => file.path === 'plan-design-review/sections/review-sections.md');
     expect(section?.content).toBe(readFileSync(join(ROOT, 'plan-design-review/sections/review-sections.md'), 'utf8'));
     expect(section?.content).toStartWith('<!-- AUTO-GENERATED');
-    expect(section?.content).toContain('## CRITICAL RULE');
+    expect(section?.content).toMatch(ASK_QUESTIONS_HEADING);
     expect(section?.content).toContain('## Formatting Rules');
     expect(occurrences(input.text, '### Pass 1: Information Architecture')).toBe(1);
   });