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
9 changed files with 60 additions and 103 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 \
-7
View File
@@ -438,13 +438,6 @@ export async function runApplyMigrations(args: string[]): Promise<void> {
const result = await m.orchestrator(orchestratorOptsFrom(cli));
if (result.status === 'failed') {
console.error(`Migration v${m.version} reported status=failed.`);
// Surface each failed phase's detail — the ledger records it, but
// the operator needs it on stderr to act (#921).
for (const p of result.phases) {
if (p.status === 'failed') {
console.error(` phase ${p.name}: ${p.detail ?? '(no detail)'}`);
}
}
// Record the attempt as 'partial' (not 'complete') so the cap counts
// it. Don't let a failed orchestrator look like it never ran.
try {
+11 -15
View File
@@ -186,6 +186,17 @@ async function phaseBFenceFacts(
const localPathById = new Map<string, string | null>();
for (const s of sources) localPathById.set(s.id, s.local_path);
// Dirty-tree refusal: check every source's local_path before writing.
for (const [id, localPath] of localPathById) {
if (localPath && isLocalPathDirty(localPath)) {
return {
name: 'fence_facts',
status: 'failed',
detail: `source "${id}" has uncommitted changes in ${localPath}. Commit or stash, then re-run.`,
};
}
}
// Walk legacy rows in (source_id, entity_slug) groups for per-page
// atomic writes.
const legacy = await engine.executeRaw<LegacyFactRow>(
@@ -224,21 +235,6 @@ async function phaseBFenceFacts(
groups.set(key, list);
}
// Dirty-tree refusal: check ONLY the sources we are about to write
// into. A dirty tree in an unrelated source (or zero fenceable rows
// at all) must not block a no-op or a targeted backfill (#927).
const targetSourceIds = new Set([...groups.keys()].map(k => k.split('\0')[0]));
for (const id of targetSourceIds) {
const localPath = localPathById.get(id);
if (localPath && isLocalPathDirty(localPath)) {
return {
name: 'fence_facts',
status: 'failed',
detail: `source "${id}" has uncommitted changes in ${localPath}. Commit or stash, then re-run.`,
};
}
}
for (const [key, group] of groups) {
const [sourceId, entitySlug] = key.split('\0');
const localPath = localPathById.get(sourceId)!;
+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;
}
/**
-13
View File
@@ -180,16 +180,3 @@ describe('runApplyMigrations exit codes (v0.36.1.x #1062)', () => {
expect(src).toMatch(/All migrations up to date[\s\S]{0,80}process\.exit\(0\)/);
});
});
// #921: a failed orchestrator must print each failed phase's detail to
// stderr — not just "reported status=failed" — so the operator can act
// without digging through the ledger.
describe('failed migration prints phase detail (#921)', () => {
test('runner loops result.phases and console.errors failed phase details', async () => {
const { readFileSync } = await import('fs');
const src = readFileSync('src/commands/apply-migrations.ts', 'utf8');
expect(src).toMatch(
/reported status=failed[\s\S]{0,400}for \(const p of result\.phases\)[\s\S]{0,200}p\.status === 'failed'[\s\S]{0,200}console\.error\([\s\S]{0,80}p\.name[\s\S]{0,80}p\.detail/,
);
});
});
+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');
+1 -48
View File
@@ -10,11 +10,10 @@
* __setTestEngineOverride so we don't need a configured brain.
*/
import { describe, test, expect, beforeAll, afterAll, beforeEach, afterEach } from 'bun:test';
import { describe, test, expect, beforeAll, afterAll, beforeEach } from 'bun:test';
import { mkdtempSync, rmSync, existsSync, readFileSync, writeFileSync, mkdirSync } from 'node:fs';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { execFileSync } from 'node:child_process';
import { PGLiteEngine } from '../src/core/pglite-engine.ts';
import { v0_32_2, __setTestEngineOverride, __testing } from '../src/commands/migrations/v0_32_2.ts';
@@ -239,52 +238,6 @@ describe('phaseBFenceFacts — happy path backfill', () => {
});
});
describe('phaseBFenceFacts — dirty-tree refusal scoping (#927)', () => {
let dirtyDir: string;
beforeEach(async () => {
// A second source whose local_path is a git repo with uncommitted changes.
dirtyDir = mkdtempSync(join(tmpdir(), 'mig-v0_32_2-dirty-'));
execFileSync('git', ['-C', dirtyDir, 'init', '-q']);
writeFileSync(join(dirtyDir, 'uncommitted.md'), 'dirty', 'utf-8');
// eslint-disable-next-line @typescript-eslint/no-explicit-any
await (engine as any).db.query(
`INSERT INTO sources (id, name, local_path) VALUES ('other', 'other', $1)`,
[dirtyDir],
);
});
afterEach(async () => {
// eslint-disable-next-line @typescript-eslint/no-explicit-any
await (engine as any).db.query(`DELETE FROM sources WHERE id = 'other'`);
rmSync(dirtyDir, { recursive: true, force: true });
});
test('no legacy facts at all → complete, dirty unrelated source ignored', async () => {
const r = await __testing.phaseBFenceFacts(engine, OPTS);
expect(r.status).toBe('complete');
expect(r.detail).toContain('scanned=0');
});
test('facts scoped to a clean source fence despite dirty unrelated source', async () => {
await seedLegacyFact({ entity_slug: 'people/alice', fact: 'Founded Acme' });
const r = await __testing.phaseBFenceFacts(engine, OPTS);
expect(r.status).toBe('complete');
expect(r.detail).toContain('fenced=1');
expect(existsSync(join(brainDir, 'people/alice.md'))).toBe(true);
});
test('still refuses when the TARGETED source is dirty', async () => {
await seedLegacyFact({ entity_slug: 'people/alice', fact: 'F1', source_id: 'other' });
const r = await __testing.phaseBFenceFacts(engine, OPTS);
expect(r.status).toBe('failed');
expect(r.detail).toContain('"other"');
expect(r.detail).toContain('uncommitted changes');
});
});
describe('phaseCVerify', () => {
test('returns complete when fence + DB row counts match', async () => {
await seedLegacyFact({ entity_slug: 'people/alice', fact: 'F1' });
+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);