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
12 changed files with 52 additions and 260 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 \
+1 -22
View File
@@ -365,12 +365,6 @@ async function main() {
if (def.required && params[key] === undefined) {
if (queryHasAlt && key === 'query') continue;
const cliName = op.cliHints?.name || op.name;
// #2822: when the missing param is the op's stdin-fed one, the usage
// line alone is misleading (the positionals may all be present — the
// pipe was just empty). Name the real problem.
if (op.cliHints?.stdin === key) {
console.error(`Error: required "${key}" is missing — stdin was empty or not piped. Pipe content on stdin or pass --${key.replace(/_/g, '-')}.`);
}
const positional = op.cliHints?.positional || [];
const usage = positional.map(p => `<${p}>`).join(' ');
console.error(`Usage: gbrain ${cliName} ${usage}`);
@@ -767,10 +761,6 @@ export function parseOpArgs(op: Operation, args: string[]): Record<string, unkno
const params: Record<string, unknown> = {};
const positional = op.cliHints?.positional || [];
let posIdx = 0;
// #2822: track which params came from positionals so a later flag that
// silently discards one (`gbrain put CONTENT --slug foo` — CONTENT was
// parsed as the slug) gets a stderr warning instead of vanishing.
const positionallySet = new Set<string>();
for (let i = 0; i < args.length; i++) {
const arg = args[i];
@@ -788,20 +778,13 @@ export function parseOpArgs(op: Operation, args: string[]): Record<string, unkno
if (paramDef?.type === 'boolean') {
params[key] = true;
} else if (i + 1 < args.length) {
if (positionallySet.has(key) && params[key] !== args[i + 1]) {
console.error(`Warning: ${arg} overrides the positional <${key}> value ${JSON.stringify(params[key])}.`);
}
params[key] = args[++i];
if (paramDef?.type === 'number') params[key] = Number(params[key]);
}
} else if (posIdx < positional.length) {
const key = positional[posIdx++];
const paramDef = op.params[key];
if (params[key] !== undefined && params[key] !== (paramDef?.type === 'number' ? Number(arg) : arg)) {
console.error(`Warning: positional <${key}> overrides the earlier --${key.replace(/_/g, '-')} value ${JSON.stringify(params[key])}.`);
}
params[key] = paramDef?.type === 'number' ? Number(arg) : arg;
positionallySet.add(key);
}
}
@@ -813,11 +796,7 @@ export function parseOpArgs(op: Operation, args: string[]): Record<string, unkno
console.error(`Error: stdin content exceeds ${MAX_STDIN} bytes. Split into smaller inputs.`);
process.exit(1);
}
// #2822: empty/whitespace-only stdin (cron with no input, broken pipe)
// stays UNSET so the required-param check rejects the call instead of
// silently writing an empty page (0 chunks, invisible to search and
// embed --stale).
if (stdinContent.trim().length > 0) params[op.cliHints.stdin] = stdinContent;
params[op.cliHints.stdin] = stdinContent;
}
return params;
-1
View File
@@ -48,7 +48,6 @@ const FRONTMATTER_RULE_NAMES: Record<ParseValidationCode, string> = {
NESTED_QUOTES: 'frontmatter-nested-quotes',
NON_STRING_FIELD: 'frontmatter-non-string-field',
EMPTY_FRONTMATTER: 'frontmatter-empty',
MULTI_FRONTMATTER: 'frontmatter-multi',
};
/** Codes whose lint findings are fixable by `gbrain frontmatter validate --fix`. */
+1 -22
View File
@@ -301,17 +301,6 @@ export async function importFromContent(
// silently fabricated a duplicate at (default, slug) — causing later
// bare-slug subqueries (getTags, deleteChunks, etc.) to crash with 21000.
const sourceId = opts.sourceId;
// #2822: reject empty/whitespace-only content before any work happens. An
// empty page writes 0 chunks — invisible to search AND to `embed --stale`
// (nothing to embed), so the mistake never surfaces. Empty content is
// always a caller bug (empty piped stdin, bad shell substitution). Thrown
// (not returned) so every wrapper site surfaces the message, matching the
// ContentSanityBlockError flow.
if (content.trim().length === 0) {
throw new Error(
`Content for "${slug}" is empty; refusing to write an empty page (0 chunks would be invisible to search and embed --stale).`,
);
}
// Reject oversized payloads before any parsing, chunking, or embedding happens.
// Uses Buffer.byteLength to count UTF-8 bytes the same way disk size would,
// so the network path behaves identically to the file path.
@@ -325,17 +314,7 @@ export async function importFromContent(
};
}
const parsed = parseMarkdown(content, slug + '.md', { activePack: opts.activePack, validate: true });
// #2743: reject stacked frontmatter (the double-put corruption class —
// already-serialized markdown re-wrapped in fresh frontmatter). gray-matter
// parses only the first block; the second would land verbatim in the body
// and poison every subsequent round-trip. Only MULTI_FRONTMATTER rejects
// here — the other validation codes keep their lint-only semantics.
const multiFm = parsed.errors?.find(e => e.code === 'MULTI_FRONTMATTER');
if (multiFm) {
throw new Error(`MULTI_FRONTMATTER: ${multiFm.message} (slug "${slug}", line ${multiFm.line})`);
}
const parsed = parseMarkdown(content, slug + '.md', { activePack: opts.activePack });
// v0.42 (#1699 trust boundary): strip gate-owned markers from UNTRUSTED
// input. parseMarkdown preserves every frontmatter key except type/title/
+1 -46
View File
@@ -11,8 +11,7 @@ export type ParseValidationCode =
| 'NULL_BYTES'
| 'NESTED_QUOTES'
| 'NON_STRING_FIELD'
| 'EMPTY_FRONTMATTER'
| 'MULTI_FRONTMATTER';
| 'EMPTY_FRONTMATTER';
export interface ParseValidationError {
code: ParseValidationCode;
@@ -332,50 +331,6 @@ function collectValidationErrors(
});
}
}
// 9. MULTI_FRONTMATTER (#2743) — a second ---…--- block right after the
// closing fence is stacked frontmatter (the double-put corruption class:
// already-serialized markdown re-wrapped in fresh frontmatter).
// gray-matter parses only the first block and silently leaves the second
// in the body. Heuristic: first non-empty line after the close is `---`,
// a later `---` closes it, EVERY line between is frontmatter-shaped
// (YAML `key:`, `- ` list item, `#` comment, indented continuation, or
// blank — the issue's "stop at the first non-frontmatter character"
// spec), and at least one is a `key:` line. A lone `---` stays a
// markdown horizontal rule, and an hrule followed by prose — even
// colon-prefixed prose like `Note: …` mixed with plain lines — is body
// content, not a stacked block.
let afterClose = closeLine + 1;
while (afterClose < lines.length && lines[afterClose].trim().length === 0) afterClose++;
if (afterClose < lines.length && lines[afterClose].trim() === '---') {
let secondClose = -1;
for (let i = afterClose + 1; i < lines.length; i++) {
const trimmed = lines[i].trim();
if (trimmed === '---') {
secondClose = i;
break;
}
const yamlShaped =
trimmed.length === 0 ||
/^[A-Za-z_][\w-]*\s*:/.test(trimmed) ||
trimmed.startsWith('- ') ||
trimmed === '-' ||
trimmed.startsWith('#') ||
/^\s/.test(lines[i]);
if (!yamlShaped) break; // first non-frontmatter line → body prose, not a stacked block
}
if (
secondClose > afterClose + 1 &&
lines.slice(afterClose + 1, secondClose).some(l => /^\s*[A-Za-z_][\w-]*\s*:/.test(l))
) {
errors.push({
code: 'MULTI_FRONTMATTER',
message:
'Stacked frontmatter: a second ---…--- block follows the frontmatter (double-put corruption); merge into a single frontmatter block',
line: afterClose + 1,
});
}
}
}
/**
+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;
}
/**
+1 -70
View File
@@ -1,8 +1,4 @@
import { afterEach, describe, expect, test } from 'bun:test';
import { spawnSync } from 'child_process';
import { join, resolve } from 'path';
import { mkdtempSync, rmSync } from 'fs';
import { tmpdir } from 'os';
import { describe, expect, test } from 'bun:test';
import { parseOpArgs } from '../src/cli.ts';
import { operationsByName } from '../src/core/operations.ts';
@@ -24,70 +20,5 @@ describe('parseOpArgs', () => {
source_id: 'gstack-code-repo-0e4763c9',
});
});
describe('positional/flag overwrite warning (#2822)', () => {
const errors: string[] = [];
const origError = console.error;
const captureErrors = () => {
console.error = (...args: unknown[]) => errors.push(args.join(' '));
};
afterEach(() => {
console.error = origError;
errors.length = 0;
});
test('a flag that overwrites a positional value warns to stderr', () => {
captureErrors();
const params = parseOpArgs(operationsByName.query, ['positional text', '--query', 'flag text']);
expect(params.query).toBe('flag text');
expect(errors.some(e => e.includes('Warning') && e.includes('--query'))).toBe(true);
});
test('a positional that overwrites an earlier flag value warns to stderr', () => {
captureErrors();
const params = parseOpArgs(operationsByName.query, ['--query', 'flag text', 'positional text']);
expect(params.query).toBe('positional text');
expect(errors.some(e => e.includes('Warning') && e.includes('<query>'))).toBe(true);
});
test('no warning when flag and positional agree', () => {
captureErrors();
parseOpArgs(operationsByName.query, ['same', '--query', 'same']);
expect(errors).toEqual([]);
});
});
});
describe('gbrain put — empty non-TTY stdin rejects (#2822)', () => {
const REPO = resolve(import.meta.dir, '..');
const CLI = join(REPO, 'src', 'cli.ts');
const runPut = (input: string) => {
// Isolated HOME so a regression can never write into a real brain.
const home = mkdtempSync(join(tmpdir(), 'gbrain-put-empty-'));
try {
return spawnSync('bun', [CLI, 'put', 'inbox/empty-stdin-test'], {
stdio: ['pipe', 'pipe', 'pipe'],
input,
encoding: 'utf-8',
timeout: 60_000,
env: { ...process.env, HOME: home, GBRAIN_SKIP_STARTUP_HOOKS: '1' },
});
} finally {
rmSync(home, { recursive: true, force: true });
}
};
test('empty stdin exits 1 and names the missing content param', () => {
const res = runPut('');
expect(res.status).toBe(1);
expect(res.stderr).toContain('content');
expect(res.stderr).toContain('stdin');
}, 90_000);
test('whitespace-only stdin also exits 1', () => {
const res = runPut(' \n\t\n');
expect(res.status).toBe(1);
expect(res.stderr).toContain('stdin');
}, 90_000);
});
+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');
-29
View File
@@ -708,32 +708,3 @@ body unchanged
expect(shortCircuited).toBe(true);
});
});
describe('importFromContent — empty content guard (#2822)', () => {
test('empty string throws instead of writing an invisible 0-chunk page', async () => {
const engine = mockEngine();
await expect(importFromContent(engine, 'inbox/empty', '', { noEmbed: true })).rejects.toThrow(/empty/i);
expect((engine as any)._calls.find((c: any) => c.method === 'putPage')).toBeUndefined();
});
test('whitespace-only content throws', async () => {
const engine = mockEngine();
await expect(importFromContent(engine, 'inbox/ws', ' \n\t \n', { noEmbed: true })).rejects.toThrow(/empty/i);
});
});
describe('importFromContent — stacked frontmatter rejection (#2743)', () => {
test('double-put shaped content (two ---…--- blocks) throws MULTI_FRONTMATTER', async () => {
const engine = mockEngine();
const md = '---\ntitle: outer\n---\n\n---\ntitle: inner\ntype: concept\n---\n\nreal body';
await expect(importFromContent(engine, 'inbox/double', md, { noEmbed: true })).rejects.toThrow(/MULTI_FRONTMATTER/);
expect((engine as any)._calls.find((c: any) => c.method === 'putPage')).toBeUndefined();
});
test('normal content with horizontal rules in the body still imports', async () => {
const engine = mockEngine();
const md = '---\ntitle: ok\ntype: concept\n---\n\nprose before\n\n---\n\nprose after the rule';
const result = await importFromContent(engine, 'inbox/hrule', md, { noEmbed: true });
expect(result.status).toBe('imported');
});
});
-50
View File
@@ -256,56 +256,6 @@ body`;
});
});
describe('MULTI_FRONTMATTER (#2743)', () => {
test('stacked frontmatter immediately after the close fence', () => {
const md = `${fence}\ntitle: outer\n${fence}\n${fence}\ntitle: inner\ntype: concept\n${fence}\n\nbody`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).toContain('MULTI_FRONTMATTER');
});
test('stacked frontmatter with a blank line between blocks (serializeMarkdown shape)', () => {
const md = `${fence}\ntitle: outer\n${fence}\n\n${fence}\ntitle: inner\n${fence}\n\nbody`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).toContain('MULTI_FRONTMATTER');
});
test('horizontal rules in the body are NOT flagged', () => {
const md = `${fence}\ntitle: ok\n${fence}\n\nsome prose\n\n${fence}\n\nmore prose\n\n${fence}\n\nend`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).not.toContain('MULTI_FRONTMATTER');
});
test('hrule pair at body start without YAML-shaped lines is NOT flagged', () => {
const md = `${fence}\ntitle: ok\n${fence}\n\n${fence}\n\nplain prose between rules\n\n${fence}\n\nend`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).not.toContain('MULTI_FRONTMATTER');
});
test('timeline sentinel form is NOT flagged', () => {
const md = `${fence}\ntitle: ok\n${fence}\n\nbody text\n\n${fence}\n\n## Timeline\n- 2024-01-01: thing`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).not.toContain('MULTI_FRONTMATTER');
});
test('body hrule + colon-prefixed prose (`Note: …`) mixed with plain lines is NOT flagged', () => {
const md = `${fence}\ntitle: ok\ntype: concept\n${fence}\n\n${fence}\n\nNote: remember to follow up\n\nlots of plain prose here\n\n${fence}\n\nmore prose`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).not.toContain('MULTI_FRONTMATTER');
});
test('fence pairing stops at the first non-frontmatter line (no far-fence pairing across prose)', () => {
const md = `${fence}\ntitle: ok\n${fence}\n\n${fence}\n\n${'plain prose line\n'.repeat(40)}TODO: fix the widget\n${'more prose\n'.repeat(40)}${fence}\nend`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).not.toContain('MULTI_FRONTMATTER');
});
test('stacked block with list-valued keys is still flagged', () => {
const md = `${fence}\ntitle: outer\n${fence}\n\n${fence}\ntitle: inner\ntags:\n - a\n - b\n${fence}\n\nbody`;
const parsed = parseMarkdown(md, undefined, { validate: true });
expect(parsed.errors!.map(e => e.code)).toContain('MULTI_FRONTMATTER');
});
});
test('error.line is set for line-bearing errors', () => {
const md = `${fence}\ntype: concept\n${fence}\n# Heading inline\n\nbody\x00drop`;
const parsed = parseMarkdown(md, undefined, { validate: true });
+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);