Compare commits

..
Author SHA1 Message Date
06eeb2890b fix(skillopt): emit proposed.md in no-mutate mode (#2635)
writeProposed now writes both best.md (optimizer's current-best pointer)
and proposed.md (the stable review artifact documented by --no-mutate);
the orchestrator returns the real proposed.md path instead of aliasing
best.md. Fix lands in the shared helper so both accept-branch call sites
are covered.

Takeover of #2719.

Co-authored-by: RerankerGuo <RerankerGuo@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-21 14:30:19 -07:00
7 changed files with 69 additions and 79 deletions
@@ -233,13 +233,14 @@ keep it or `git checkout` to throw it away. Nothing is committed for you.
**For a skill that ships with gbrain** (anything under the gbrain repo's own
`skills/`): SkillOpt refuses to overwrite it by default and writes the winner to
`skills/<name>/skillopt/best.md` instead, so an optimization pass can never
silently mutate a skill other people depend on. Two ways to handle that:
`skills/<name>/skillopt/proposed.md` instead (while keeping `best.md` as the
optimizer's current-best pointer), so an optimization pass can never silently
mutate a skill other people depend on. Two ways to handle that:
```bash
# See the proposed improvement without touching SKILL.md (works for ANY skill):
gbrain skillopt meeting-prep --split 1:1:1 --no-mutate
# → writes skills/meeting-prep/skillopt/best.md (the proposed rewrite), prints its path. Copy what you want.
# → writes skills/meeting-prep/skillopt/proposed.md, updates best.md, and prints the proposal path.
# Actually rewrite a bundled skill (explicit opt-in + an independent held-out set):
gbrain skillopt brain-ops --split 1:1:1 --allow-mutate-bundled \
+3 -11
View File
@@ -35,11 +35,10 @@
*
* The doctor renders both side by side.
*
* Drift contract: every check name that ships through doctor MUST appear in
* Drift contract: every check name that ships in doctor.ts MUST appear in
* exactly one set below. The drift-guard test in
* `test/doctor-categories.test.ts` enforces this by reading doctor check
* emitter sources via a tagged-string scan and asserting set membership
* exactly.
* `test/doctor-categories.test.ts` enforces this by reading doctor.ts source
* via a tagged-string scan and asserting set membership exactly.
*
* If you add a new doctor check, you MUST add its name to the appropriate
* set here. The categorize step in `src/commands/doctor.ts` falls through
@@ -68,15 +67,12 @@ export const BRAIN_CHECK_NAMES: ReadonlySet<string> = new Set([
'conversation_parser_probe_health',
'cross_modal_modality_backfill',
'cycle_freshness',
'dangling_aliases',
'effective_date_health',
'embed_staleness',
'embedding_column_registry',
'embedding_env_override',
'embedding_provider',
'embedding_width_consistency',
'embeddings',
'entity_link_coverage',
'eval_drift',
'extract_atoms_backlog',
'extract_health',
@@ -106,9 +102,7 @@ export const BRAIN_CHECK_NAMES: ReadonlySet<string> = new Set([
'stub_guard_24h',
'sync_failures',
'sync_freshness',
'takes_count',
'takes_weight_grid',
'timeline_coverage',
'unified_multimodal_coverage',
'voice_gate_health',
]);
@@ -176,14 +170,12 @@ export const META_CHECK_NAMES: ReadonlySet<string> = new Set([
'eval_capture',
'minions_migration',
'multi_source_drift',
'pack_upgrade_available',
'schema_pack_active',
'schema_pack_consistency',
'schema_pack_source_drift',
'schema_version',
'slug_fallback_audit',
'timeline_dedup_index',
'type_proliferation',
'upgrade_errors',
]);
+10 -4
View File
@@ -93,7 +93,13 @@ import { resolveLrSchedule } from './lr-schedule.ts';
import { preflight, formatPreflightReport } from './preflight.ts';
import { isRejected, loadRejectedBuffer, makeRejectedEntry, saveRejectedBuffer } from './rejected-buffer.ts';
import { runReflect, runOneShotRewrite, describeJudges } from './reflect.ts';
import { acceptCandidate, bestPath, revertAllPending, skillPath, writeProposed } from './version-store.ts';
import {
acceptCandidate,
proposedPath as proposedFilePath,
revertAllPending,
skillPath,
writeProposed,
} from './version-store.ts';
import { runValidationGate, scoreSkillOnTasks } from './validate-gate.ts';
import { ROLLOUT_SUCCESS_THRESHOLD } from './types.ts';
import type { SkillOptOpts, EditOp, RunReceipt, BenchmarkTask } from './types.ts';
@@ -702,9 +708,9 @@ async function runOptimizationLoop(
// to the catch's assignment values only (it can't prove the async callback ran).
const finalOutcome = outcome as 'accepted' | 'no_improvement' | 'aborted' | 'errored';
if (!mutateDecision.mutate && finalOutcome === 'accepted') {
// best.md was written by writeProposed() in the accept branch (no-mutate
// path); it doubles as proposed.md for human review. SKILL.md untouched.
proposedPath = bestPath(skillsDir, skillName);
// writeProposed() emitted both the best pointer and the stable review
// artifact in the accept branch. SKILL.md remains untouched.
proposedPath = proposedFilePath(skillsDir, skillName);
} else if (mutateDecision.mutate) {
mutatedSkillFile = finalOutcome === 'accepted';
}
+15 -9
View File
@@ -23,6 +23,7 @@
*
* history.json
* best.md
* proposed.md
* versions/
* v0001_e1_s1.md
* v0002_e1_s2.md
@@ -52,6 +53,10 @@ export function bestPath(skillsDir: string, skillName: string): string {
return path.join(skilloptDir(skillsDir, skillName), 'best.md');
}
export function proposedPath(skillsDir: string, skillName: string): string {
return path.join(skilloptDir(skillsDir, skillName), 'proposed.md');
}
export function skillPath(skillsDir: string, skillName: string): string {
return path.join(skillsDir, skillName, 'SKILL.md');
}
@@ -171,17 +176,18 @@ export function acceptCandidate(input: AcceptInput): AcceptResult {
}
/**
* Write the candidate to `best.md` (which doubles as `proposed.md`) WITHOUT
* touching SKILL.md or the history ledger. Used by the `--no-mutate` /
* bundled-without-allow paths: the optimizer found a better candidate but the
* caller opted out of in-place mutation, so we surface it for human review.
* Returns the path written. Atomic (.tmp + rename).
* Write the candidate to both `best.md` and `proposed.md` WITHOUT touching
* SKILL.md or the history ledger. `best.md` remains the optimizer's current
* best pointer; `proposed.md` is the stable human-review artifact promised by
* `--no-mutate`. Returns the proposal path. Each write is atomic (.tmp + rename).
*/
export function writeProposed(skillsDir: string, skillName: string, candidateText: string): string {
const p = bestPath(skillsDir, skillName);
fs.mkdirSync(path.dirname(p), { recursive: true });
atomicWrite(p, candidateText);
return p;
const best = bestPath(skillsDir, skillName);
const proposed = proposedPath(skillsDir, skillName);
fs.mkdirSync(path.dirname(best), { recursive: true });
atomicWrite(best, candidateText);
atomicWrite(proposed, candidateText);
return proposed;
}
/**
+18 -48
View File
@@ -1,10 +1,10 @@
/**
* Drift guard for src/core/doctor-categories.ts.
*
* Reads doctor check emitter source via a literal-string scan, enumerates every
* `name: '<...>'` Check name, and asserts each appears in exactly ONE category
* set. The union of the four sets must equal the discovered names exactly —
* no orphans, no extras.
* Reads src/commands/doctor.ts source via a literal-string scan, enumerates
* every `name: '<...>'` Check name, and asserts each appears in exactly ONE
* category set. The union of the four sets must equal the discovered names
* exactly — no orphans, no extras.
*
* This is the structural failure the v0.41.19.0 plan-eng-review caught:
* doctor.ts grows new checks regularly; without this guard, the
@@ -25,30 +25,26 @@ import {
} from '../src/core/doctor-categories.ts';
const DOCTOR_TS_PATH = join(import.meta.dir, '..', 'src', 'commands', 'doctor.ts');
const ONBOARD_CHECKS_TS_PATH = join(import.meta.dir, '..', 'src', 'core', 'onboard', 'checks.ts');
const CHECK_SOURCE_PATHS = [DOCTOR_TS_PATH, ONBOARD_CHECKS_TS_PATH];
function enumerateCheckNames(): Set<string> {
const source = readFileSync(DOCTOR_TS_PATH, 'utf-8');
const names = new Set<string>();
for (const path of CHECK_SOURCE_PATHS) {
const source = readFileSync(path, 'utf-8');
// 1) Inline object-literal form: `{ name: 'foo', ... }`.
for (const m of source.matchAll(/name:\s*['"]([a-z][a-z0-9_]+)['"]/g)) {
names.add(m[1]);
}
// 2) Helper-function form: `const name = 'foo';` inside a check helper.
// Catches checks like `nightly_quality_probe_health` and
// `conversation_facts_backlog` that build the Check from a captured
// name constant.
for (const m of source.matchAll(/const\s+name\s*=\s*['"]([a-z][a-z0-9_]+)['"]/g)) {
names.add(m[1]);
}
// 1) Inline object-literal form: `{ name: 'foo', ... }`.
for (const m of source.matchAll(/name:\s*['"]([a-z][a-z0-9_]+)['"]/g)) {
names.add(m[1]);
}
// 2) Helper-function form: `const name = 'foo';` inside a check helper.
// Catches checks like `nightly_quality_probe_health` and
// `conversation_facts_backlog` that build the Check from a captured
// name constant.
for (const m of source.matchAll(/const\s+name\s*=\s*['"]([a-z][a-z0-9_]+)['"]/g)) {
names.add(m[1]);
}
return names;
}
describe('doctor-categories drift guard', () => {
test('every doctor-emitted check name belongs to exactly one category set', () => {
test('every check name in doctor.ts source belongs to exactly one category set', () => {
const discovered = enumerateCheckNames();
const allCategorized = new Set<string>([
...BRAIN_CHECK_NAMES,
@@ -63,7 +59,7 @@ describe('doctor-categories drift guard', () => {
}
if (missing.length > 0) {
throw new Error(
`These check names appear in doctor check emitters but are not categorized in ` +
`These check names appear in doctor.ts but are not categorized in ` +
`src/core/doctor-categories.ts: ${missing.sort().join(', ')}. ` +
`Add each to BRAIN/SKILL/OPS/META_CHECK_NAMES.`,
);
@@ -90,7 +86,7 @@ describe('doctor-categories drift guard', () => {
expect(dupes).toEqual([]);
});
test('every categorized name is currently used in doctor check emitters (no stale entries)', () => {
test('every categorized name is currently used in doctor.ts source (no stale entries)', () => {
const discovered = enumerateCheckNames();
const allCategorized = new Set<string>([
...BRAIN_CHECK_NAMES,
@@ -128,14 +124,6 @@ describe('categorizeCheck', () => {
expect(categorizeCheck('sync_freshness')).toBe('brain');
});
test('returns the right category for onboard data-quality check names', () => {
expect(categorizeCheck('embed_staleness')).toBe('brain');
expect(categorizeCheck('entity_link_coverage')).toBe('brain');
expect(categorizeCheck('timeline_coverage')).toBe('brain');
expect(categorizeCheck('takes_count')).toBe('brain');
expect(categorizeCheck('dangling_aliases')).toBe('brain');
});
test('returns the right category for a known skill name', () => {
expect(categorizeCheck('resolver_health')).toBe('skill');
expect(categorizeCheck('skill_conformance')).toBe('skill');
@@ -152,24 +140,6 @@ describe('categorizeCheck', () => {
expect(categorizeCheck('upgrade_errors')).toBe('meta');
});
test('returns the right category for onboard schema-pack check names without warning', () => {
const originalWrite = process.stderr.write.bind(process.stderr);
const captured: string[] = [];
(process.stderr as { write: typeof process.stderr.write }).write = ((
chunk: string | Uint8Array,
) => {
captured.push(typeof chunk === 'string' ? chunk : Buffer.from(chunk).toString());
return true;
}) as typeof process.stderr.write;
try {
expect(categorizeCheck('pack_upgrade_available')).toBe('meta');
expect(categorizeCheck('type_proliferation')).toBe('meta');
expect(captured.filter((c) => c.includes('[doctor-categories]'))).toEqual([]);
} finally {
(process.stderr as { write: typeof process.stderr.write }).write = originalWrite;
}
});
test('unknown check name falls through to meta with a stderr warn (once per process)', () => {
const originalWrite = process.stderr.write.bind(process.stderr);
const captured: string[] = [];
+4 -4
View File
@@ -39,6 +39,7 @@ import { runSkillOpt } from '../../src/core/skillopt/orchestrator.ts';
import {
bestPath,
loadHistory,
proposedPath,
skillPath,
} from '../../src/core/skillopt/version-store.ts';
import { loadRejectedBuffer } from '../../src/core/skillopt/rejected-buffer.ts';
@@ -741,7 +742,7 @@ describe('skillopt T3 — F11 held-out gate, ablation opts, no-DB-pollution', ()
} finally { fixture.cleanup(); }
});
test('--no-mutate writes proposed.md (best.md), leaves SKILL.md untouched', async () => {
test('--no-mutate writes proposed.md and best.md, leaves SKILL.md untouched', async () => {
const fixture = setupFixture(SKILL_PEOPLE_ONLY, CITATIONS_BENCHMARK);
try {
installStub({
@@ -753,10 +754,9 @@ describe('skillopt T3 — F11 held-out gate, ablation opts, no-DB-pollution', ()
const result = await runOnce(fixture, { noMutate: true });
expect(result.outcome).toBe('accepted');
expect(result.mutatedSkillFile).toBe(false);
expect(result.proposedPath).toBeDefined();
// proposed.md (best.md) exists and carries the improvement.
expect(fs.existsSync(result.proposedPath!)).toBe(true);
expect(result.proposedPath).toBe(proposedPath(fixture.skillsDir, SKILL));
expect(fs.readFileSync(result.proposedPath!, 'utf8')).toContain('## Citations');
expect(fs.readFileSync(bestPath(fixture.skillsDir, SKILL), 'utf8')).toContain('## Citations');
// SKILL.md on disk is UNCHANGED (still People-only).
const skill = fs.readFileSync(skillPath(fixture.skillsDir, SKILL), 'utf8');
expect(skill).not.toContain('## Citations');
+15
View File
@@ -12,9 +12,11 @@ import {
bestPath,
historyPath,
loadHistory,
proposedPath,
revertAllPending,
skillPath,
versionsDir,
writeProposed,
} from '../../src/core/skillopt/version-store.ts';
let tmpDir: string;
@@ -79,6 +81,19 @@ describe('acceptCandidate (D8 two-phase commit)', () => {
});
});
describe('writeProposed', () => {
test('writes distinct best and proposed artifacts without mutating SKILL.md (#2635)', () => {
const candidate = '---\nname: test\n---\nproposed body\n';
const written = writeProposed(tmpDir, SKILL, candidate);
expect(written).toBe(proposedPath(tmpDir, SKILL));
expect(fs.readFileSync(bestPath(tmpDir, SKILL), 'utf8')).toBe(candidate);
expect(fs.readFileSync(proposedPath(tmpDir, SKILL), 'utf8')).toBe(candidate);
expect(fs.readFileSync(skillPath(tmpDir, SKILL), 'utf8')).toContain('baseline body');
});
});
describe('revertAllPending (D8 crash recovery)', () => {
test('no-op when no pending rows', () => {
const reverted = revertAllPending(tmpDir, SKILL);