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 50 additions and 137 deletions
+1 -5
View File
@@ -206,11 +206,7 @@ jobs:
needs: cache-check
if: needs.cache-check.outputs.hit != 'true'
runs-on: ubuntu-latest
# 20 (was 15): shard 4 runs ~14.5 min on master (dream.test.ts ~29s/test
# dominates it) and hits the 15-min ceiling on slower runners, cancelling
# mid-run with 0 test failures. Rebalancing via
# scripts/mine-shard-weights.ts is the real fix; this stops the bleeding.
timeout-minutes: 20
timeout-minutes: 15
strategy:
fail-fast: false
matrix:
@@ -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 \
+1 -34
View File
@@ -23,15 +23,6 @@
* hold conventions and shared rule files, not skills. Files like
* `_brain-filing-rules.md` live at the root and are not considered
* skills by either loader.
*
* ClawHub-installed workspace skills (#1767): a skill dir carrying
* `.clawhub/origin.json` is an externally-managed runtime integration
* (e.g. an email or catalog skill), not a gbrain-routable skill. The
* derive path SKIPS those so `gbrain doctor` resolver_health doesn't
* hard-fail on them — UNLESS the skill's SKILL.md frontmatter declares
* `triggers:`, which is the explicit opt-in to gbrain routing (and the
* same surface that makes it reachable). An explicit manifest.json that
* lists a ClawHub skill also keeps strict checking (verbatim path).
*/
import { existsSync, readFileSync, readdirSync, statSync } from 'fs';
@@ -69,27 +60,9 @@ function parseSkillName(skillMdPath: string): string | null {
}
}
/**
* Does the SKILL.md frontmatter declare a `triggers:` key? A ClawHub-
* installed skill that ships gbrain `triggers:` has explicitly opted in
* to gbrain routing and gets full resolver checks (#1767).
*/
function declaresTriggers(skillMdPath: string): boolean {
try {
const content = readFileSync(skillMdPath, 'utf-8');
const fmMatch = content.match(/^---\n([\s\S]*?)\n---/);
if (!fmMatch) return false;
return /^triggers:/m.test(fmMatch[1]);
} catch {
return false;
}
}
/**
* Walk skillsDir, return every `<skillsDir>/<dir>/SKILL.md` as a
* ManifestEntry. Dotfile and underscore-prefixed dirs are skipped, as
* are ClawHub-installed external skills that haven't opted in to gbrain
* routing via `triggers:` frontmatter (#1767).
* ManifestEntry. Dotfile and underscore-prefixed dirs are skipped.
*/
function deriveManifest(skillsDir: string): ManifestEntry[] {
const out: ManifestEntry[] = [];
@@ -120,12 +93,6 @@ function deriveManifest(skillsDir: string): ManifestEntry[] {
const skillMd = join(subdirAbs, 'SKILL.md');
if (!existsSync(skillMd)) continue;
// ClawHub-installed external skill (#1767): skip unless it opts in
// to gbrain routing by declaring `triggers:` in its frontmatter.
if (existsSync(join(subdirAbs, '.clawhub', 'origin.json')) && !declaresTriggers(skillMd)) {
continue;
}
const frontmatterName = parseSkillName(skillMd);
const name = frontmatterName && frontmatterName !== '' ? frontmatterName : entry;
out.push({ name, path: `${entry}/SKILL.md` });
+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;
}
/**
-29
View File
@@ -382,35 +382,6 @@ describe("DRY detection — checkResolvable", () => {
});
});
describe("#1767 — ClawHub workspace skills are not resolver-required", () => {
let dir: string;
afterEachCleanup(() => dir && rmSync(dir, { recursive: true, force: true }));
test("ClawHub skill without gbrain metadata produces no unreachable/mece_gap", () => {
dir = mkdtempSync(join(tmpdir(), "gbrain-clawhub-"));
// Native gbrain skill: routable via frontmatter triggers. No manifest.json
// (the OpenClaw derive path from the issue repro).
mkdirSync(join(dir, "query"), { recursive: true });
writeFileSync(
join(dir, "query", "SKILL.md"),
`---\nname: query\ndescription: test\ntriggers:\n - "what do we know"\n---\n\n# query\n`
);
// ClawHub-installed integration: no triggers, no resolver row.
mkdirSync(join(dir, "agentmail", ".clawhub"), { recursive: true });
writeFileSync(
join(dir, "agentmail", ".clawhub", "origin.json"),
JSON.stringify({ registry: "https://clawhub.ai", slug: "agentmail" })
);
writeFileSync(join(dir, "agentmail", "SKILL.md"), `---\nname: agentmail\ndescription: email integration\n---\n\n# agentmail\n`);
const report = checkResolvable(dir);
const agentmailIssues = report.issues.filter(i => i.skill === "agentmail");
expect(agentmailIssues).toEqual([]);
expect(report.ok).toBe(true);
expect(report.summary.total_skills).toBe(1);
});
});
describe("v0.22.4 regression — actual repo skills/ has 0 errors", () => {
test("repo skills/ pass check-resolvable cleanly (zero errors AND zero warnings)", () => {
// The v0.22.4 (Part A) contract was zero warnings AND zero errors.
+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');
-49
View File
@@ -166,55 +166,6 @@ describe('loadOrDeriveManifest', () => {
expect(r.skills.map(s => s.name)).toEqual(['apple', 'mango', 'zebra']);
});
// #1767 — ClawHub-installed workspace skills are external integrations,
// not gbrain-routable skills. The derive path skips them unless they
// opt in via `triggers:` frontmatter.
it('skips ClawHub-origin skills without triggers frontmatter (#1767)', () => {
const dir = scratch();
writeSkill(dir, 'query', 'query');
writeSkill(dir, 'agentmail', 'agentmail');
mkdirSync(join(dir, 'agentmail', '.clawhub'), { recursive: true });
writeFileSync(
join(dir, 'agentmail', '.clawhub', 'origin.json'),
JSON.stringify({ registry: 'https://clawhub.ai', slug: 'agentmail' })
);
const r = loadOrDeriveManifest(dir);
expect(r.derived).toBe(true);
expect(r.skills.map(s => s.name)).toEqual(['query']);
});
it('includes ClawHub-origin skills that opt in via triggers frontmatter (#1767)', () => {
const dir = scratch();
writeSkill(dir, 'agentmail', 'agentmail');
mkdirSync(join(dir, 'agentmail', '.clawhub'), { recursive: true });
writeFileSync(
join(dir, 'agentmail', '.clawhub', 'origin.json'),
JSON.stringify({ registry: 'https://clawhub.ai', slug: 'agentmail' })
);
writeFileSync(
join(dir, 'agentmail', 'SKILL.md'),
`---\nname: agentmail\ndescription: test\ntriggers:\n - "send email"\n---\n\n# agentmail\n`
);
const r = loadOrDeriveManifest(dir);
expect(r.derived).toBe(true);
expect(r.skills.map(s => s.name)).toEqual(['agentmail']);
});
it('keeps ClawHub-origin skills listed in an explicit manifest.json (#1767)', () => {
// Explicit manifest.json is a deliberate declaration — strict checking stays.
const dir = scratch();
writeSkill(dir, 'agentmail', 'agentmail');
mkdirSync(join(dir, 'agentmail', '.clawhub'), { recursive: true });
writeFileSync(
join(dir, 'agentmail', '.clawhub', 'origin.json'),
JSON.stringify({ registry: 'https://clawhub.ai', slug: 'agentmail' })
);
writeManifest(dir, { skills: [{ name: 'agentmail', path: 'agentmail/SKILL.md' }] });
const r = loadOrDeriveManifest(dir);
expect(r.derived).toBe(false);
expect(r.skills.map(s => s.name)).toEqual(['agentmail']);
});
it('treats dirs without SKILL.md as not-a-skill', () => {
const dir = scratch();
writeSkill(dir, 'query', 'query');
+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);