Merge commit '2f9d11a59d68822c36b3e03fb3e535f48a8ace96' into gsta258-l6-prepush-guards

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

# Conflicts:
#	docs/pregate.md
This commit is contained in:
garrytan committed 2026-11-03 14:34:08 +00:00
commit 5660d09adc
141 files changed
+7184 -393

No files matched your search

+35 -1
View File
@@ -223,8 +223,42 @@ export function cell(text: string, cap = 400): string {
export const APPROVALS_TABLE_HEADER = '| Date | Asked | Exact words | Outcome | Covers | Still pending |';
export const APPROVALS_FOOTER_END = '(display only; not an approval citation).';
export function renderRecord(read: ReadResult, loc: StoreLocation, o: { heading?: boolean } = {}): string {
export type RepoVisibility = 'public' | 'private' | 'unknown';
/**
* The repository's visibility as GitHub reports it through gh, for the repository gh acts on by default (GH_REPO, its
* default repo, an upstream remote) and for the origin remote a PR is opened against: private only when every one
* asked is private, public when any is public, and unknown for anything gh cannot answer.
*/
export function repoVisibility(cwd: string, env: Env = process.env): RepoVisibility {
const ask = (repo: string[]): RepoVisibility => {
const r = spawnSync('gh', ['repo', 'view', ...repo, '--json', 'visibility', '-q', '.visibility'], { cwd, env, encoding: 'utf8', timeout: 15_000 });
if (r.status !== 0) return 'unknown';
const v = (r.stdout ?? '').trim().toLowerCase();
return v === 'private' || v === 'internal' ? 'private' : v === 'public' ? 'public' : 'unknown';
};
const origin = spawnSync('git', ['remote', 'get-url', 'origin'], { cwd, env, encoding: 'utf8', timeout: 15_000 });
const slug = origin.status === 0 ? /([^/:]+\/[^/]+?)(?:\.git)?\/*$/.exec(origin.stdout.trim())?.[1] : undefined;
const seen = [ask([]), ...(slug ? [ask([slug])] : [])];
return seen.includes('public') ? 'public' : seen.every(v => v === 'private') ? 'private' : 'unknown';
}
/**
* Full rows only where the repository is known private, or --full asks for them and GitHub does not report it public.
* A PR description keeps every earlier version in its edit history, so a public or unknown repository gets the
* redacted block: whether anything was approved and an opaque id of the record (a hash of its row ids), nothing else.
*/
export function renderMode(visibility: RepoVisibility, full: boolean): 'full' | 'redacted' {
return visibility === 'private' || (full && visibility === 'unknown') ? 'full' : 'redacted';
}
export function renderRecord(read: ReadResult, loc: StoreLocation, o: { heading?: boolean; mode?: 'full' | 'redacted' } = {}): string {
const head = o.heading === false ? [] : ['## Approvals', ''];
if (o.mode === 'redacted' && read.rows.length) {
const approved = read.rows.some(r => r.outcome === 'approved' || r.outcome === 'standing');
const id = 'r-' + createHash('sha256').update(read.rows.map(r => r.id).sort().join('\n')).digest('hex').slice(0, 12);
return [...head, `Approved: ${approved ? 'yes' : 'no'} (record ${id}). The full approval record stays private: this repository is public or its visibility is unknown.`, ''].join('\n');
}
if (read.state === 'unavailable') return [...head, `Record unavailable on this machine: no approval record for branch ${cell(loc.branch)} here, and the state root is ephemeral with no Drive mirror, so this is not proof that nothing was approved.`, ''].join('\n');
if (!read.rows.length) return [...head, `No approvals recorded for branch ${cell(loc.branch)}.`, ''].join('\n');
const lines = [APPROVALS_TABLE_HEADER, '|---|---|---|---|---|---|'];
+25 -8
View File
@@ -16,6 +16,7 @@ import * as fs from 'node:fs';
import * as path from 'node:path';
import { OUTSIDE_ROLE, sectionLoadFor, type Phase, type Scope } from './autoplan-prompts';
import { RunError, type RunState } from './autoplan-state';
import { redactFindingSpans } from './redact-engine';
export const KINDS = ['fix', 'feature'] as const;
export type AutoplanKind = typeof KINDS[number];
@@ -33,7 +34,7 @@ export const FIX_LINES = [
const LENS: Record<Phase, string> = { ceo: 'CEO', design: 'Design', dx: 'DX', eng: 'Eng' };
const INLINE_DRAFT_BYTES = 100_000;
const UNTRACKED_MAX = 200;
const SECRET_NAME = /(^|\/)(\.env(\.[^/]*)?|[^/]*\.(pem|key|p12|pfx|keystore)|id_(rsa|dsa|ecdsa|ed25519)|\.netrc|\.npmrc|\.pypirc|credentials(\.[^/]*)?|[^/]*secret[^/]*)$/i;
const SECRET_NAME = /(^|\/)(\.env[^/]*|[^/]*\.(pem|key|p12|pfx|keystore|jks)|id_(rsa|dsa|ecdsa|ed25519)(_sk)?|[._]netrc|\.npmrc|\.pypirc|\.pgpass|[^/]*credentials[^/]*|[^/]*secret[^/]*)$/i;
/** The first next's mode: lean for --kind feature, or for an unattended Capy run that names no kind; --full and --ceo-only keep today's run. */
export function leanMode(o: { kind?: AutoplanKind; full?: boolean; ceoOnly?: boolean; base?: string; sessionKind: string; env: NodeJS.ProcessEnv }): LeanMode | undefined {
@@ -56,7 +57,9 @@ export const draftLine = (d: DraftInput) =>
/**
* The draft code against the base: tracked changes since the merge base (committed and not) plus untracked files,
* the run directory excluded, written once to <dir>/draft.diff (read-only) and hashed. No git, no merge base, no
* change or a git command that failed is an honest none with its reason, never a partial draft.
* change or a git command that failed is an honest none with its reason, never a partial draft. The draft goes to
* a model provider, so a file whose name looks like a secret is never diffed (tracked or untracked), and every
* file's patch passes through the redact engine; one it cannot redact is left out with a note (GST-1164, #3914).
*/
export function captureDraft(cwd: string, dir: string, out: string, base?: string): DraftInput {
const git = (args: string[], at = cwd) => spawnSync('git', ['-c', 'core.quotepath=off', ...args], { cwd: at, encoding: 'utf8', timeout: 60_000, maxBuffer: 256 * 1024 * 1024 });
@@ -73,19 +76,33 @@ export function captureDraft(cwd: string, dir: string, out: string, base?: strin
const failed = (r: ReturnType<typeof git>, what: string, ok = [0]) => r.error || r.status === null || !ok.includes(r.status)
? `draft capture failed: ${what} ${r.error ? `errored (${r.error.message})` : `exited ${r.status ?? r.signal}`}${r.stderr?.trim() ? `: ${r.stderr.trim().split('\n')[0]}` : ''}` : undefined;
const raw = ['diff', '--no-color', '--no-ext-diff', '--no-textconv'];
const tracked = git([...raw, mergeBase, '--'], root);
const names = git(['diff', '--name-only', mergeBase, '--'], root);
const names = git(['diff', '--name-status', '-z', '-M', '-C', mergeBase, '--'], root);
const others = git(['ls-files', '--others', '--exclude-standard', '-z'], root);
const broken = failed(tracked, 'git diff') ?? failed(names, 'git diff --name-only') ?? failed(others, 'git ls-files');
const listed = failed(names, 'git diff') ?? failed(others, 'git ls-files');
if (listed) return { base: ref, none: listed };
const fields = names.stdout.split('\0');
const changed: string[] = [];
const secretTracked: string[][] = [];
for (let i = 0; fields[i];) {
const entry = /^[RC]/.test(fields[i++]!) ? [fields[i++]!, fields[i++]!] : [fields[i++]!];
changed.push(...entry);
if (entry.some(f => SECRET_NAME.test(f))) secretTracked.push(entry);
}
const tracked = git([...raw, '-M', mergeBase, '--', '.', ...secretTracked.flat().map(f => `:(exclude,literal)${f}`)], root);
const broken = failed(tracked, 'git diff');
if (broken) return { base: ref, none: broken };
const redacted = (patch: string, what: string) => redactFindingSpans(patch) ?? `# draft: ${what} left out (it could not be redacted safely)\n`;
const untracked = others.stdout.split('\0').filter(f => f && !inRun(f)).sort();
const parts = [tracked.stdout];
const parts = [
...tracked.stdout.split(/^(?=diff --git )/m).map(p => redacted(p, p.slice(0, p.indexOf('\n')))),
...secretTracked.map(e => `# draft: ${e.join(' -> ')} left out (${e.length > 1 ? 'a rename or copy of a file whose name looks like a secret' : 'its name looks like a secret'})\n`),
];
for (const f of untracked.slice(0, UNTRACKED_MAX)) {
if (SECRET_NAME.test(f)) { parts.push(`# draft: untracked ${f} left out (its name looks like a secret)\n`); continue; }
const d = git([...raw, '--no-index', '--', '/dev/null', f], root);
const bad = failed(d, `git diff --no-index ${f}`, [0, 1]);
if (bad) return { base: ref, none: bad };
parts.push(d.stdout);
parts.push(redacted(d.stdout, `untracked ${f}`));
}
if (untracked.length > UNTRACKED_MAX) parts.push(`# draft: ${untracked.length - UNTRACKED_MAX} more untracked files not shown\n`);
const text = parts.join('');
@@ -94,7 +111,7 @@ export function captureDraft(cwd: string, dir: string, out: string, base?: strin
fs.writeFileSync(file, text, process.platform === 'win32' ? { flag: 'wx' } : { flag: 'wx', mode: 0o444 });
const bytes = fs.readFileSync(file);
return { base: ref, root, merge_base: mergeBase, sha256: crypto.createHash('sha256').update(bytes).digest('hex'), bytes: bytes.length,
...(bytes.length > INLINE_DRAFT_BYTES ? { shown_bytes: INLINE_DRAFT_BYTES } : {}), files: [...new Set([...names.stdout.split('\n').filter(Boolean), ...untracked])].sort() };
...(bytes.length > INLINE_DRAFT_BYTES ? { shown_bytes: INLINE_DRAFT_BYTES } : {}), files: [...new Set([...changed, ...untracked])].sort() };
}
/** The merged brief (before the shared blocks) and the receipt line (after RESULT FORMAT) a lean phase's native prompt adds. */
+93 -18
View File
@@ -14,7 +14,9 @@
* max age, the run's own tree.json naming HEAD's tree);
* every run executed tests;
* with an inventory command, the runs' tree.json files cover every file it
* prints (a lane carries only what CI provably ran);
* prints (a lane carries only what CI provably ran); with named jobs instead
* (a lane with no test files, GST-1189), each of those jobs succeeded in the
* run's newest attempt, read by verifyRun like the profile's own jobs;
* every run still stands after the inventory ran (runChanged).
*
* Anything else returns a refusal with a reason code, and the caller runs the
@@ -25,12 +27,18 @@
* bounded by a timeout, and returns green, red with the failing jobs, still
* running, or none.
*/
import { runChanged, runReceiptFields, uncoveredFiles, verifyRun, type ApiRun, type RunDeps, type RunEvidence, type RunProfile, type RunSource } from './run-evidence';
import { normalPath, runChanged, runInventory, runReceiptFields, uncoveredFiles, verifyRun, type ApiRun, type RunDeps, type RunEvidence, type RunProfile, type RunSource } from './run-evidence';
export type CarryRefusal = 'dirty_tree' | 'no_profile' | 'no_run' | 'no_verified_run' | 'tree_mismatch' | 'no_tests'
| 'inventory_failed' | 'inventory_empty' | 'inventory_uncovered' | 'run_changed' | 'tree_changed' | 'api_error';
| 'inventory_failed' | 'inventory_empty' | 'inventory_uncovered' | 'inventory_skipped' | 'run_changed' | 'tree_changed' | 'api_error';
export type CarryRefused = { ok: false; reason: CarryRefusal; detail: string };
export type CarryVerdict = { ok: true; runs: RunEvidence[]; tests: number; receipt: string } | CarryRefused;
/**
* `tests` never claims a whole run's count for a narrow lane: with an inventory it is the number of the lane's inventory files
* the runs' tree.json covers (`unit: 'files'`; tree.json has no per-file test counts yet), with named jobs `n/a`, and only
* without either (the whole run carried) the runs' test count. `remainder`: inventory files every run skipped entirely
* (tree.json `skipped_files`), which the caller must run here with the lane's per_file runner before the lane counts.
*/
export type CarryVerdict = { ok: true; runs: RunEvidence[]; tests: number | 'n/a'; unit?: 'files'; notOwed?: string; remainder?: string[]; receipt: string } | CarryRefused;
/** How many of HEAD's runs per workflow are tried, newest first. */
const CANDIDATES = 10;
@@ -38,12 +46,13 @@ const no = (reason: CarryRefusal, detail: string): CarryRefused => ({ ok: false,
const newestFirst = (a: ApiRun, b: ApiRun) => (b.created_at ?? '').localeCompare(a.created_at ?? '') || b.id - a.id;
/** The newest run of `profile` on `head` that verifies; a refusal lists each candidate's reason, newest first. */
function newestVerified(src: RunSource, name: string, profile: RunProfile, head: string): { ok: true; evidence: RunEvidence } | CarryRefused {
function newestVerified(src: RunSource, name: string, profile: RunProfile, head: string, jobs: string[]): { ok: true; evidence: RunEvidence } | CarryRefused {
if (jobs.length) profile = { ...profile, jobs: [...new Set([...profile.jobs, ...jobs])] };
const listed = src.deps.runsForHead(profile.workflow, head).filter(r => profile.events.includes(r.event)).sort(newestFirst).slice(0, CANDIDATES);
if (!listed.length) return no('no_run', `no ${profile.workflow} run (${profile.events.join(', ')}) on ${head.slice(0, 12)} yet`);
const refusals: string[] = [];
for (const run of listed) {
const v = verifyRun(src.deps, { repo: src.repo, runRef: String(run.id), profileName: name, profile, vetoHeads: [head], now: src.now });
const v = verifyRun(src.deps, { repo: src.repo, runRef: String(run.id), profileName: name, profile, vetoHeads: [head], now: src.now, unchanged: src.unchanged });
if (v.ok) return v;
refusals.push(`run ${run.id}: ${v.reason}`);
}
@@ -52,9 +61,13 @@ function newestVerified(src: RunSource, name: string, profile: RunProfile, head:
/**
* The CI evidence for HEAD's tree under `profiles` (one verified run each), or why there is none. `wtree` reads the
* working-tree fingerprint (before the lookup and after it); `inventory` lists the files the runs must cover.
* working-tree fingerprint (before the lookup and after it); `inventory` lists the files the runs must cover, or `jobs`
* names the jobs that must have succeeded in each run's newest attempt (a lane with no test files; no test count needed).
*/
export function findCarry(src: RunSource, o: { profiles: string[]; head: string; tree: string; wtree: () => string | null; inventory?: () => string[] | { error: string } }): CarryVerdict {
export function findCarry(src: RunSource, o: { profiles: string[]; head: string; tree: string; wtree: () => string | null; inventory?: () => string[] | { error: string }; jobs?: string[]; perFile?: boolean }): CarryVerdict {
const jobs = o.jobs ?? [];
let covered: number | undefined;
let remainder: string[] = [];
const before = o.wtree();
if (before !== o.tree) return no('dirty_tree', `the working tree (${before ? before.slice(0, 12) : 'fingerprint unavailable'}) is not HEAD's committed tree ${o.tree.slice(0, 12)}; only a clean, committed tree carries`);
if (!o.profiles.length) return no('no_profile', 'no run profile is named');
@@ -63,20 +76,24 @@ export function findCarry(src: RunSource, o: { profiles: string[]; head: string;
for (const name of o.profiles) {
const profile = src.profiles[name];
if (!profile) return no('no_profile', `${name} is not a run_profiles name on the base branch`);
const v = newestVerified(src, name, profile, o.head);
const v = newestVerified(src, name, profile, o.head, jobs);
if (!v.ok) return v;
runs.push(v.evidence);
}
const other = runs.find(e => e.tree !== o.tree);
if (other) return no('tree_mismatch', `run ${other.run_id} tested tree ${other.tree.slice(0, 12)}, not ${o.tree.slice(0, 12)}`);
const empty = runs.find(e => !e.files.length || e.tests === 0);
const empty = jobs.length ? undefined : runs.find(e => !e.files.length || e.tests === 0);
if (empty) return no('no_tests', `run ${empty.run_id}'s tree.json ${empty.files.length ? 'records 0 tests' : 'lists no test files'}`);
if (o.inventory) {
const listed = o.inventory();
if ('error' in listed) return no('inventory_failed', listed.error);
if (!listed.length) return no('inventory_empty', 'the inventory command printed no files');
const missing = uncoveredFiles(listed, runs);
if (missing.length) return no('inventory_uncovered', `${missing.length} of ${listed.length} inventory file(s) are in no run's tree.json: ${missing.slice(0, 5).join(', ')}${missing.length > 5 ? ', …' : ''}`);
const absent = uncoveredFiles(listed, runs, 'covered');
const some = (fs: string[]) => `${fs.slice(0, 5).join(', ')}${fs.length > 5 ? ', …' : ''}`;
if (absent.length) return no('inventory_uncovered', `${absent.length} of ${listed.length} inventory file(s) are in no run's tree.json: ${some(absent)}`);
remainder = uncoveredFiles(listed, runs, 'uncovered');
if (remainder.length && !o.perFile) return no('inventory_skipped', `${remainder.length} of ${listed.length} inventory file(s) skipped every test in CI and the lane declares no per_file runner: ${some(remainder)}`);
covered = new Set(listed.map(normalPath)).size - remainder.length;
}
for (const e of runs) {
const changed = runChanged(src.deps, e.run_id, e.run_attempt);
@@ -87,6 +104,13 @@ export function findCarry(src: RunSource, o: { profiles: string[]; head: string;
}
const after = o.wtree();
if (after !== o.tree) return no('tree_changed', `the working tree changed while the runs were verified (now ${after ? after.slice(0, 12) : 'unreadable'})`);
if (jobs.length) {
const gates = [...new Set(o.profiles.map(n => src.profiles[n]?.skippable?.gate).filter(Boolean))];
const allSkipped = runs.every(e => jobs.every(j => e.skipped_jobs?.includes(j)));
const notOwed = allSkipped ? `CI skipped at this tree (gate job ${gates.join(', ')}, workflow unchanged)` : undefined;
return { ok: true, runs, tests: 'n/a', ...(notOwed ? { notOwed } : {}), receipt: `ci-carry tree=${o.tree.slice(0, 12)} jobs=${jobs.join(',')}${notOwed ? ` not_owed=${gates.join(',')}` : ''} ${runReceiptFields(runs, src.now)}` };
}
if (covered !== undefined) return { ok: true, runs, tests: covered, unit: 'files', ...(remainder.length ? { remainder } : {}), receipt: `ci-carry tree=${o.tree.slice(0, 12)} files=${covered}${remainder.length ? ` skipped_here=${remainder.length}` : ''} ${runReceiptFields(runs, src.now)}` };
const tests = runs.reduce((n, e) => n + (e.tests ?? e.files.length), 0);
return { ok: true, runs, tests, receipt: `ci-carry tree=${o.tree.slice(0, 12)} tests=${tests} ${runReceiptFields(runs, src.now)}` };
}
@@ -109,24 +133,75 @@ export async function waitForRun(deps: RunDeps, o: { workflow: string; events: s
}
}
/** `ship_lanes_from` (base branch only): exact lane command → the profiles whose runs carry it and the command listing its test files. */
export interface ShipLaneMapping { profiles: string[]; inventory: string }
/**
* `ship_lanes_from` (base branch only): exact lane command → the profiles whose runs carry it and either the command listing
* its test files (`inventory`) or the CI jobs that run it (`jobs`, for a lane with no test files such as a typecheck; GST-1189).
*/
export interface ShipLaneMapping { profiles: string[]; inventory?: string; jobs?: string[]; aliases?: string[]; per_file?: string }
/**
* The `ship_lanes_from` entry for an exact lane command: the entry it keys, else the entry that declares it in `aliases`
* (GST-1190: the same lane run under another exact spelling, such as an env prefix). Never normalized: an alias is a
* second exact command the base's owner declared, not a pattern.
*/
export function laneMapping(from: Record<string, ShipLaneMapping>, command: string): ShipLaneMapping | undefined {
if (Object.prototype.hasOwnProperty.call(from, command)) return from[command];
return Object.values(from).find(m => m.aliases?.includes(command));
}
/** findCarry's coverage rule for one mapping: its inventory (a pin-file command run in `cwd`, with a per_file runner or not) or its named jobs. */
export const mappingCoverage = (m: ShipLaneMapping, cwd: string): { inventory?: () => string[] | { error: string }; jobs?: string[]; perFile?: boolean } =>
m.jobs ? { jobs: m.jobs } : { inventory: () => runInventory(m.inventory!, cwd), perFile: !!m.per_file };
/**
* What `gstack-evidence record-ci` reported, read from its own last record line onward (the per_file remainder's output comes
* first, so a test printing look-alike lines cannot change the reading): the carried runs, count, unit and remainder, any
* not-owed reason, or the last CARRY_NONE reason.
*/
export function recordCiResult(stderr: string): { runs?: string; tests?: string; files?: boolean; notOwed?: string; here?: string; none?: string } {
const lines = stderr.split('\n');
const at = lines.findLastIndex(x => x.startsWith('gstack-evidence: recorded label=') && x.includes(' log=- source=ci runs='));
const none = lines.findLast(x => x.startsWith('CARRY_NONE: '))?.slice('CARRY_NONE: '.length);
if (at < 0) return none ? { none } : {};
const got = / tests_ran=(\d+|n\/a)( unit=files)? log=- source=ci runs=(\S+)/.exec(lines[at]!);
const here = / skipped_here=(\d+) tests_ran_here=(\S+)$/.exec(lines[at]!);
const notOwed = lines.slice(at + 1).find(x => x.startsWith('gstack-evidence: not owed: '))?.slice('gstack-evidence: not owed: '.length);
return { runs: got?.[3], tests: got?.[1], files: !!got?.[2], ...(notOwed ? { notOwed } : {}), ...(here ? { here: `skipped_here=${here[1]} file(s) tests_ran_here=${here[2]}` } : {}) };
}
/** The per_file runner command for `files`: `{files}` replaced by the paths, each single-quoted for bash. */
export const perFileCommand = (template: string, files: string[]): string =>
template.replaceAll('{files}', files.map(f => `'${f.replace(/'/g, `'\\''`)}'`).join(' '));
/** `<path>: <message>` errors for `ship_lanes_from`; empty when valid. Shared with lib/pregate/config.ts. */
export function shipLanesErrors(raw: unknown, profileNames: readonly string[]): Array<{ path: string; message: string }> {
if (raw === undefined) return [];
if (!raw || typeof raw !== 'object' || Array.isArray(raw)) return [{ path: 'ship_lanes_from', message: 'must be an object of exact lane command → { profiles: [run_profiles names], inventory: command }' }];
if (!raw || typeof raw !== 'object' || Array.isArray(raw)) return [{ path: 'ship_lanes_from', message: 'must be an object of exact lane command → { profiles: [run_profiles names], inventory: command } or { profiles, jobs: [job names] }' }];
const errors: Array<{ path: string; message: string }> = [];
const oneLine = (s: unknown) => typeof s === 'string' && !!s.trim() && s.length <= 512 && !/[\0\r\n]/.test(s);
const declared = new Map<string, string>(Object.keys(raw as object).map(k => [k, 'a key']));
for (const [command, m] of Object.entries(raw as Record<string, unknown>)) {
const at = `ship_lanes_from[${JSON.stringify(command)}]`;
if (!oneLine(command)) errors.push({ path: at, message: 'the key must be the exact lane command: one line under 512 characters' });
if (!m || typeof m !== 'object' || Array.isArray(m)) { errors.push({ path: at, message: 'must be { profiles, inventory }' }); continue; }
if (!m || typeof m !== 'object' || Array.isArray(m)) { errors.push({ path: at, message: 'must be { profiles, inventory } or { profiles, jobs }' }); continue; }
const o = m as Record<string, unknown>;
for (const k of Object.keys(o)) if (k !== 'profiles' && k !== 'inventory') errors.push({ path: `${at}.${k}`, message: 'unknown key; valid keys: profiles, inventory' });
for (const k of Object.keys(o)) if (!['profiles', 'inventory', 'jobs', 'aliases', 'per_file'].includes(k)) errors.push({ path: `${at}.${k}`, message: 'unknown key; valid keys: profiles, inventory, jobs, aliases, per_file' });
if ('per_file' in o && (!oneLine(o.per_file) || !(o.per_file as string).includes('{files}') || !('inventory' in o))) errors.push({ path: `${at}.per_file`, message: 'must be one command line under 512 characters containing {files}, on an entry with an inventory (it runs the inventory files CI skipped entirely)' });
if ('aliases' in o) {
const aliases = o.aliases;
if (!Array.isArray(aliases) || !aliases.length || aliases.some(a => !oneLine(a)) || new Set(aliases).size !== aliases.length) errors.push({ path: `${at}.aliases`, message: 'must be a non-empty array of distinct exact lane commands, each one line under 512 characters' });
else for (const a of aliases as string[]) {
if (declared.has(a)) errors.push({ path: `${at}.aliases`, message: `${JSON.stringify(a)} is already ${declared.get(a)}; each exact command maps to one entry` });
else declared.set(a, `an alias of ${JSON.stringify(command)}`);
}
}
if (!Array.isArray(o.profiles) || !o.profiles.length || o.profiles.some(p => typeof p !== 'string')) errors.push({ path: `${at}.profiles`, message: 'must be a non-empty array of run_profiles names' });
else for (const p of o.profiles as string[]) if (!profileNames.includes(p)) errors.push({ path: `${at}.profiles`, message: `${JSON.stringify(p)} is not a run_profiles name` });
if (!oneLine(o.inventory)) errors.push({ path: `${at}.inventory`, message: 'must be one command line under 512 characters that prints the lane\'s test files, one per line' });
if (('inventory' in o) === ('jobs' in o)) errors.push({ path: at, message: 'declare exactly one of inventory (a command listing the lane\'s test files) or jobs (the CI jobs that run it)' });
else if ('jobs' in o) {
const jobs = o.jobs;
if (!Array.isArray(jobs) || !jobs.length || jobs.some(j => typeof j !== 'string' || !oneLine(j) || j.length > 128) || new Set(jobs).size !== jobs.length) errors.push({ path: `${at}.jobs`, message: 'must be a non-empty array of distinct CI job names, each one line under 128 characters' });
} else if (!oneLine(o.inventory)) errors.push({ path: `${at}.inventory`, message: 'must be one command line under 512 characters that prints the lane\'s test files, one per line' });
}
return errors;
}
+21
View File
@@ -137,6 +137,27 @@ export function resolveBase(ref: string, cwd: string, head?: string): { ref: str
return sha ? { ref, sha } : null;
}
export type ResolvedBase = NonNullable<ReturnType<typeof resolveBase>>;
/** Why resolveBase read another ref than the name given (`local main is behind origin/main; base is origin/main`); null when the name stood as given. */
export function baseNote(name: string, base: ResolvedBase): string | null {
if (base.local === 'fallback') return `${base.unrelatedRemote ? `${base.unrelatedRemote} shares no history with HEAD` : `no remote-tracking branch for ${name}`}; base is the local ${name}, which may lag its remote`;
if (!base.local) return null;
return `${{ unrelated: `local ${name} shares no history with HEAD`, missing: `no local branch ${name}`, behind: `local ${name} is behind ${base.ref}`, ahead: `local ${name} is ahead of ${base.ref}`, diverged: `local ${name} has diverged from ${base.ref}` }[base.local]}; base is ${base.ref}`;
}
/**
* A --base argument as the commit a review or ship bin reads (GST-1036): resolveBase against HEAD, so a bare `main`
* reads origin/main over a stale local main, with the note on stderr as `<bin>: <note>` when it read another ref than
* the name. An unresolvable name comes back as given, so the caller's own git error still names it.
*/
export function baseArg(bin: string, name: string, cwd: string): string {
const base = resolveBase(name, cwd, 'HEAD');
const note = base && baseNote(name, base);
if (note) process.stderr.write(`${bin}: ${note}\n`);
return base?.sha ?? name;
}
/** A word for a printed shell line: as is when it is plain, single-quoted otherwise (ref names may hold `;`, `$`, `(`, `{1,2}`). */
function shellWord(s: string): string {
return /^[A-Za-z0-9._\/@^~+:-]+$/.test(s) ? s : `'${s.replace(/'/g, "'\\''")}'`;
+8
View File
@@ -83,6 +83,8 @@ export interface PregateCheck {
/** A remote_gate receipt accepted from a stamp-only predecessor tree (lib/pregate/remote-receipt.ts): the gated tree, the tree it was accepted for, the policy, and every differing path with its stamp rule. */
export interface PregateReceiptSubstitution { gated_tree: string; accepted_for: string; policy: string; files: Array<{ path: string; rule: string }> }
export interface PregateRemote { lane: string; platform: string; workflow: string; job: string; tests: string[]; receipt: string | null; cleared_by?: string; command?: string; substitution?: PregateReceiptSubstitution }
/** An owner-granted flaky exception (lib/pregate/flaky-exception.ts): the runs, cases and grant that cleared one obligation, bound to one head and tree. */
export interface PregateException { lane: string; repo: string; pr: number; head: string; tree: string; runs: string[]; cases: string[]; grant: string; by: string; words: string; binding: string }
export interface PregateFile {
schema_version: 1; tree: string; checks: PregateCheck[]; run?: string; stage1_ms?: number; stage2_ms?: number;
// C-pregate (C8): head/base identity, the policy sources, persisted requires-remote obligations and the one-line verdict.
@@ -93,6 +95,8 @@ export interface PregateFile {
policy_user_path?: string;
/** This run's scratch directory under the state root (logs, scan files). */
scratch?: string;
/** Obligations cleared from runs with failures under an owner's grant (`clear --flaky-exception`). */
exceptions?: PregateException[];
}
export interface ShipReceipt {
schema_version: 1; head: string; base: string; pr?: number; version?: string; gated_tree?: string;
@@ -123,6 +127,8 @@ export interface ShipReceipt {
review_targeted?: { classes: string[]; rules: string; brief_sha: string; product_patch_id: string; status: 'done' | 'failed' | 'not-run'; findings: number; raised_by: string };
/** Step 5 failures gstack-base-red proved are the base's own (lib/base-red.ts): a conditional receipt the queue refuses until /ship --resume rewrites it without this block. */
base_red?: { cases: string[]; files: string[]; base: string; lane: string; expires_at: string };
/** The pregate's owner-granted flaky exceptions (gstack-pregate clear --flaky-exception): pregate only, never a required CI check. */
flaky_exception?: Array<{ lane: string; runs: string[]; cases: string[]; grant: string; by: string }>;
/** The base policy's review_exclude_paths at the reviewed head (lib/review-exclude.ts); absent when the base declares none. Carry and the queue compare `digest`. */
review_exclude?: { source: string; patterns: string[]; applied: string[]; excluded: number; excluded_bytes: number; kept: number; gate_tagged: string[]; patterns_sha256: string; list_sha256: string; digest: string };
}
@@ -234,6 +240,7 @@ export const ARTIFACT_SCHEMAS = {
head: str, base: str, generated_at: str, policy: str, policy_digest: str, policy_user_path: str, registry: str, verdict: str, scratch: str,
requires_remote: { type: 'array', items: { type: 'object', required: ['lane', 'platform', 'tests', 'receipt'], properties: { lane: str, platform: str, workflow: str, job: str, tests: strList, receipt: nullableStr, cleared_by: str, command: str,
substitution: { type: 'object', required: ['gated_tree', 'accepted_for', 'policy', 'files'], properties: { gated_tree: str, accepted_for: str, policy: str, files: { type: 'array', items: { type: 'object', required: ['path', 'rule'], properties: { path: str, rule: str } } } } } } } },
exceptions: { type: 'array', items: { type: 'object', required: ['lane', 'repo', 'pr', 'head', 'tree', 'runs', 'cases', 'grant', 'by', 'words', 'binding'], properties: { lane: str, repo: str, pr: { type: 'integer' }, head: str, tree: str, runs: strList, cases: strList, grant: str, by: str, words: str, binding: str } } },
},
},
'ship-receipt': {
@@ -250,6 +257,7 @@ export const ARTIFACT_SCHEMAS = {
carry: { type: 'object', required: ['from', 'reason', 'product_patch_id'], properties: { from: str, reason: str, product_patch_id: str, changed_paths: strList, merge_commits: { type: 'integer', minimum: 0 }, range_diff: str, covers: strList, reruns: strList, trees: strList, ancestry: { enum: ['linear', 'rewritten'] } } },
head_tree: str, product_patch_id: str, gstack_version: str,
base_red: { type: 'object', required: ['cases', 'files', 'base', 'lane', 'expires_at'], properties: { cases: strList, files: strList, base: str, lane: str, expires_at: str } },
flaky_exception: { type: 'array', items: { type: 'object', required: ['lane', 'runs', 'cases', 'grant', 'by'], properties: { lane: str, runs: strList, cases: strList, grant: str, by: str } } },
review_depth: { type: 'object', required: ['depth', 'class', 'runtime_lines', 'rules', 'base', 'gate_scan'], properties: { depth: { enum: ['light', 'standard', 'full'] }, class: { enum: ['docs', 'tests', 'code', 'gate'] }, runtime_lines: { type: 'integer', minimum: 0 }, rules: str, base: str, registry: nullableStr, gate_scan: { type: 'object', required: ['tagged', 'head'], properties: { tagged: { type: ['integer', 'null'], minimum: 0 }, head: str } }, reasons: strList, risk: { type: 'array', items: { enum: ['money', 'auth', 'secret', 'security', 'data'] } } } },
review_targeted: { type: 'object', required: ['classes', 'rules', 'brief_sha', 'product_patch_id', 'status', 'findings', 'raised_by'], properties: { classes: { type: 'array', items: { enum: ['money', 'gate', 'auth', 'secret'] } }, rules: str, brief_sha: str, product_patch_id: str, status: { enum: ['done', 'failed', 'not-run'] }, findings: { type: 'integer', minimum: 0 }, raised_by: str } },
review_exclude: { type: 'object', required: ['source', 'patterns', 'applied', 'excluded', 'excluded_bytes', 'kept', 'gate_tagged', 'patterns_sha256', 'list_sha256', 'digest'], properties: { source: str, patterns: strList, applied: strList, excluded: { type: 'integer', minimum: 0 }, excluded_bytes: { type: 'integer', minimum: 0 }, kept: { type: 'integer', minimum: 0 }, gate_tagged: strList, patterns_sha256: str, list_sha256: str, digest: str } },
+7 -3
View File
@@ -6,7 +6,10 @@
*
* Verdict file: <state root>/install-verdicts/<sha256(checkout realpath)>/<sha256(project realpath | "none")>.json,
* then the path the GSTACK_LAUNCH line names as GSTACK_INSTALL_VERDICT, then
* the schema 1 file keyed by the checkout alone (<sha256(checkout)>.json).
* the schema 1 file keyed by the checkout alone (<sha256(checkout)>.json). An
* install-profile verdict (for=review-voice) at the first path yields to the
* launched GSTACK_INSTALL_VERDICT, so a reviewer's install on a shared machine
* never shadows the session's own verdict (GST-1161).
* The project is the working directory's git toplevel unless that is the
* checkout itself. It is the one record the bun shim, the Bash twin,
* lib/pregate/runtime.ts and admission read.
@@ -102,11 +105,12 @@ export function installProject(checkout: string, cwd: string = process.cwd()): s
try { const real = fs.realpathSync(top); return real === checkout ? null : real; } catch { return null; }
}
/** The verdict file admission reads: (checkout, project), then GSTACK_INSTALL_VERDICT, then the schema 1 file. */
/** The verdict file admission reads: (checkout, project), then GSTACK_INSTALL_VERDICT, then the schema 1 file. An install-profile verdict at (checkout, project) admits no workflow, so the launched GSTACK_INSTALL_VERDICT outranks it (GST-1161). */
export function findInstallVerdict(checkout: string, project: string | null, env: Env): string {
const stateRoot = resolveStateRoot(env);
const keyed = installVerdictPath(checkout, stateRoot, project);
if (fs.existsSync(keyed)) return keyed;
if (fs.existsSync(keyed) && !INSTALL_PROFILES.includes(readInstallVerdict(keyed)?.for ?? '')) return keyed;
if (fs.existsSync(keyed) && (!env.GSTACK_INSTALL_VERDICT || !fs.existsSync(env.GSTACK_INSTALL_VERDICT))) return keyed;
if (env.GSTACK_INSTALL_VERDICT && fs.existsSync(env.GSTACK_INSTALL_VERDICT)) return env.GSTACK_INSTALL_VERDICT;
const legacy = legacyInstallVerdictPath(checkout, stateRoot);
return fs.existsSync(legacy) ? legacy : keyed;
+4 -1
View File
@@ -17,6 +17,7 @@
import { spawnSync } from 'node:child_process';
import * as fs from 'node:fs';
import * as path from 'node:path';
import { resolveBase } from './gate-diff/git';
import { findSection } from './markdown-sections';
import { scan } from './redact-engine';
import type { ResultCodeName } from './result-codes';
@@ -55,9 +56,11 @@ export function readCapture(file: string, roots: string[], what: string): string
return cleanText(fs.readFileSync(real), `${what} file`);
}
/** Resolve `ref` to a commit sha without letting it be read as an option. */
/** Resolve `ref` to a commit sha without letting it be read as an option; a bare branch name reads origin's tracking branch first (resolveBase, GST-1036). */
export function resolveCommit(cwd: string, ref: string): string | ShotError {
if (!ref || ref.startsWith('-') || /\s/.test(ref)) return { code: 'PR_SHOTS_INPUT_REFUSED', message: `ref ${JSON.stringify(ref)} is not a plain ref` };
const base = resolveBase(ref, cwd, 'HEAD');
if (base) return base.sha;
const r = git(cwd, ['rev-parse', '--verify', '--quiet', '--end-of-options', `${ref}^{commit}`]);
return r.status === 0 ? r.stdout.trim() : { code: 'PR_SHOTS_INPUT_NOT_FOUND', message: `base ref ${ref}` };
}
+38 -3
View File
@@ -11,9 +11,17 @@
* incident: a gbrain serial test landed without its manifest row and
* passed both the isolation check and pregate; `test/cli-goldens.test.ts`
* spawns the CLI as a subprocess, so no import graph selects it.
*
* GST-1186: a guard whose exact command already passed at this tree is not run
* again. #3901's carry (gstack-evidence record-ci) carries it from a verified
* CI run when the base's ship_lanes_from maps the command, else a counted
* ledger pass of the same command at the same working tree stands
* (gstack-evidence passed). ship_lanes_from is read from origin/<base> only, so
* the PR that adds a mapping cannot carry from it yet.
*/
import * as fs from 'node:fs';
import * as path from 'node:path';
import { laneMapping, recordCiResult } from '../../ci-carry';
import { matchesAny } from '../../ship-policy';
import type { Guard } from '../config';
import { baseInputs, sha256 } from '../context';
@@ -23,8 +31,29 @@ import { checkScratch } from '../run-scratch';
import { readRange, STALL_IDLE, supervise } from '../supervise';
import type { EnvironmentCause } from '../types';
import type { CheckDef, CheckResult, PregateContext } from '../types';
import { runScriptSync } from '../../script-invocation';
interface GuardRun { guard: Guard; exit: number | null; timedOut: boolean; stalled: boolean; tail: string[]; log: string; environment: EnvironmentCause | null }
interface GuardRun { guard: Guard; exit: number | null; timedOut: boolean; stalled: boolean; tail: string[]; log: string; environment: EnvironmentCause | null; prior?: string; carryNone?: string }
const EVIDENCE = path.resolve(import.meta.dir, '..', '..', '..', 'bin', 'gstack-evidence');
/**
* The pass that already covers this guard's exact command at this tree (GST-1186): a CI carry (or a CI skip the unchanged gate
* decided, printed as not owed), else a same-tree local ledger pass under the runtime the guard would launch with.
*/
function priorPass(ctx: PregateContext, guard: Guard, env: NodeJS.ProcessEnv): { prior?: string; carryNone?: string } {
const evidence = (args: string[]) => runScriptSync(EVIDENCE, args, { cwd: ctx.repoRoot, env, encoding: 'utf8', timeout: 600_000 });
let carryNone: string | undefined;
if (laneMapping(ctx.config.ship_lanes_from, guard.command)) {
const ci = evidence(['record-ci', '--label', `guard-${guard.id}`, '--base', ctx.base, '--', guard.command]);
const got = recordCiResult(String(ci.stderr ?? ''));
if (ci.status === 0 && got.runs) return { prior: got.notOwed ? `not owed: ${got.notOwed} runs=${got.runs}` : `carried runs=${got.runs}` };
carryNone = got.none ?? `record-ci exit ${ci.status}`;
}
const local = evidence(['passed', '--', guard.command]);
const hit = /^PASSED: (.+)$/m.exec(String(local.stdout ?? ''))?.[1];
return local.status === 0 && hit ? { prior: `reused ${hit}`, carryNone } : { carryNone };
}
async function runGuard(ctx: PregateContext, guard: Guard, env: NodeJS.ProcessEnv, dir: string): Promise<GuardRun> {
const timeoutMs = guard.timeout_s ? guard.timeout_s * 1000 : ctx.timeoutMs;
@@ -53,9 +82,14 @@ async function run(ctx: PregateContext): Promise<CheckResult> {
if (runtime.refusal) return { ...base, ...runtime.refusal, cause: 'environment', environment: launchCause('runtime', runtime.refusal.detail), lines: [`runtime: ${runtime.label}`, environmentLine(launchCause('runtime', runtime.refusal.detail))] };
const dir = checkScratch(ctx.scratch, 'guards');
const runs: GuardRun[] = [];
for (const g of selected) runs.push(await runGuard(ctx, g, runtime.env, dir));
for (const g of selected) {
const p = priorPass(ctx, g, runtime.env);
runs.push(p.prior ? { guard: g, exit: 0, timedOut: false, stalled: false, tail: [], log: '-', environment: null, ...p } : { ...(await runGuard(ctx, g, runtime.env, dir)), carryNone: p.carryNone });
}
const lines = [`runtime: ${runtime.label}`];
for (const r of runs) {
if (r.carryNone) lines.push(`CI_CARRY: none guard=${r.guard.id} (${r.carryNone})`);
if (r.prior) { lines.push(`GUARD: ${r.guard.id} ${r.prior}`); continue; }
lines.push(r.stalled ? `GUARD: ${r.guard.id} stalled: no output for ${ctx.stallTimeoutMs / 1000} s${STALL_IDLE}; process group killed (log ${r.log})` : r.timedOut ? `GUARD: ${r.guard.id} timed out after ${r.guard.timeout_s ?? ctx.timeoutMs / 1000} s (log ${r.log})` : `GUARD: ${r.guard.id} exit=${r.exit} (log ${r.log})`);
if (r.exit !== 0) lines.push(...r.tail.map(l => ` ${l}`));
if (r.environment) lines.push(` ${r.guard.id} ${environmentLine(r.environment)}`);
@@ -68,7 +102,8 @@ async function run(ctx: PregateContext): Promise<CheckResult> {
if (failed.length) return { ...withRuns, status: 'fail', cause: 'diff', code: 'PREGATE_GUARD_FAILED', detail: `${failed.length} guard(s) failed: ${failed.map(r => r.guard.id).join(', ')}${timedOut.length ? `; timed out: ${timedOut.map(r => r.guard.id).join(', ')}` : ''}${blocked.length ? `; blocked by the environment: ${blocked.map(r => r.guard.id).join(', ')}` : ''}`, fix: fixOf(failed) };
if (blocked.length) return environmentResult(withRuns, blocked[0]!.environment!, `${blocked.length} guard(s) could not run on this machine: ${blocked.map(r => `${r.guard.id} (${r.environment!.kind})`).join(', ')}${timedOut.length ? `; timed out: ${timedOut.map(r => r.guard.id).join(', ')}` : ''}`);
if (timedOut.length) return { ...withRuns, status: 'incomplete', code: timedOut.some(r => r.stalled) ? 'PREGATE_STALLED' : 'PREGATE_INCOMPLETE', detail: `${timedOut.length} guard(s) did not finish: ${timedOut.map(r => `${r.guard.id} (${r.stalled ? 'stalled' : 'timed out'})`).join(', ')}`, fix: 'fix the hang, or raise the guard\'s timeout_s, --timeout or --stall-timeout; an unfinished guard never passes' };
return { ...withRuns, detail: `${runs.length} guard(s) passed: ${runs.map(r => r.guard.id).join(', ')}` };
const prior = runs.filter(r => r.prior).length;
return { ...withRuns, detail: `${runs.length} guard(s) passed: ${runs.map(r => r.guard.id).join(', ')}${prior ? ` (${prior} from an earlier pass at this tree)` : ''}` };
}
export const guardsCheck: CheckDef = { id: 'guards', stage: 'preflight', tier: 2, downgradable: false, needsRepoCommands: true, run };
+4 -3
View File
@@ -40,7 +40,7 @@ import { selectLanes, type Selection } from '../lane-selection';
import { launchRuntime } from '../runtime';
import { cargoModel, type CargoModel } from '../rust';
import { adapterFor, readRange, STALL_IDLE, supervise, type Counter, type Supervised } from '../supervise';
import { verifyFromRun, type RunProfile } from '../../run-evidence';
import { ranFiles, verifyFromRun, type RunProfile } from '../../run-evidence';
import { matchesAny } from '../../ship-policy';
import type { CheckDef, CheckResult, PregateContext, RemoteObligation } from '../types';
@@ -643,9 +643,10 @@ function fromRun(ctx: PregateContext, sel: Selection): { line: string; passed: s
const names = 'error' in check ? [] : fromRunProfiles(ctx, check.profiles);
const v = 'error' in check ? { ok: false as const, reason: 'profiles_unavailable', detail: check.error } : verifyFromRun(check, ref, names, ctx.wtree, ctx.head);
if (!v.ok) return { line: `REUSE: refused reason=${v.reason} — ${v.detail}; every selected test runs here`, passed: [] };
const ran = new Set(v.evidence.files);
const ran = new Set(ranFiles(v.evidence));
const passed = sel.localTests.filter(t => ran.has(t));
return { line: `REUSE: accepted ${passed.length} of ${sel.localTests.length} selected local test file(s) from ${v.receipt}; ${sel.localTests.length - passed.length} still run`, passed, receipt: v.receipt };
const skipped = sel.localTests.filter(t => v.evidence.skipped?.includes(t)).length;
return { line: `REUSE: accepted ${passed.length} of ${sel.localTests.length} selected local test file(s) from ${v.receipt}; ${sel.localTests.length - passed.length} still run${skipped ? ` (${skipped} skipped every test in the run)` : ''}`, passed, receipt: v.receipt };
}
/**
+30 -7
View File
@@ -46,9 +46,13 @@
* tree, the command listing the gate's test files they must
* cover, and the command naming the selected files the gate
* itself never runs (gstack: paid files)
* ship_lanes_from exact /ship lane command → `{ profiles, inventory }`: the
* run profiles whose verified runs at HEAD's tree carry that
* lane, and the command listing its test files (lib/ci-carry.ts)
* ship_lanes_from exact /ship lane command → `{ profiles, inventory }` or
* `{ profiles, jobs }`: the run profiles whose verified runs
* at HEAD's tree carry that lane, and either the command
* listing its test files or the CI jobs that run it (a lane
* with no test files, GST-1189), plus optional `aliases`:
* other exact commands for the same lane (GST-1190)
* (lib/ci-carry.ts)
* allowlist `{ finding, paths?, pattern?, reason }` entries the `secrets`
* check accepts as benign: a gstack-redact finding id or a
* non-secret category, scoped by path globs and/or a regex
@@ -74,7 +78,8 @@ import { shipLanesErrors, type ShipLaneMapping } from '../ci-carry';
import type { ResultCodeName } from '../result-codes';
import { runProfileErrors, type RemoteGateRuns, type RunProfile } from '../run-evidence';
import { sealedErrors } from '../sealed-guard';
import { containedRelative, globToRegExp } from '../ship-policy';
import { containedRelative, globToRegExp, runningGstackVersion, SHIP_POLICY_PATH } from '../ship-policy';
import { cmpVersion, parseVersion } from '../version-source';
export const PREGATE_CONFIG_PATH = '.gstack/pregate.json';
export const PREGATE_CONFIG_SCHEMA = 'https://github.com/garrytan/gstack/docs/pregate.md#pregate-v1';
@@ -291,6 +296,22 @@ function git(cwd: string, args: string[]): { ok: boolean; out: string } {
return { ok: r.status === 0 && !r.error, out: (r.stdout ?? '').trim() };
}
/**
* GST-1190: an unknown key in a policy whose project floor (min_gstack_version in the same tree's ship-policy.json) is
* above this gstack belongs to a newer gstack. It still refuses, but as an upgrade, not as a typo.
*/
function newerGstackErrors(parsed: { config: PregateConfig; errors: ConfigError[]; declaresLanes: boolean }, policyText: string | null): typeof parsed {
const unknown = (e: ConfigError) => e.message.startsWith('unknown key');
if (!policyText || !parsed.errors.some(unknown)) return parsed;
let min: unknown;
try { min = (JSON.parse(policyText) as Record<string, unknown> | null)?.min_gstack_version; } catch { return parsed; }
const running = runningGstackVersion();
const want = typeof min === 'string' ? parseVersion(min) : null;
const have = parseVersion(running);
if (!want || !have || cmpVersion(want, have) <= 0) return parsed;
return { ...parsed, errors: parsed.errors.map(e => unknown(e) ? { ...e, code: 'PREGATE_GSTACK_TOO_OLD' as const, message: `needs gstack >= ${min as string}; upgrade (this gstack is ${running} and does not know this key)` } : e) };
}
export function loadConfigFromBase(repoRoot: string, base: string, knownChecks: readonly string[] = []): ConfigLoad {
const ref = `origin/${base.replace(/^origin\//, '')}`;
const sha = git(repoRoot, ['rev-parse', '--verify', '--quiet', `${ref}^{commit}`]);
@@ -302,13 +323,15 @@ export function loadConfigFromBase(repoRoot: string, base: string, knownChecks:
if (!entry.ok || entry.out) return { unavailable: true, base: base.replace(/^origin\//, ''), detail: `${ref}:${PREGATE_CONFIG_PATH} could not be read and git cannot confirm it is absent` };
return { config: DEFAULT_CONFIG, errors: [], present: false, label: `${ref}@${sha.out.slice(0, 12)} (no ${PREGATE_CONFIG_PATH})`, kind: 'none' };
}
return { ...parseConfig(show.out, knownChecks), present: true, label: `${ref}@${sha.out.slice(0, 12)}:${PREGATE_CONFIG_PATH}`, kind: 'base' };
const policy = git(repoRoot, ['show', `${sha.out}:${SHIP_POLICY_PATH}`]);
return { ...newerGstackErrors(parseConfig(show.out, knownChecks), policy.ok ? policy.out : null), present: true, label: `${ref}@${sha.out.slice(0, 12)}:${PREGATE_CONFIG_PATH}`, kind: 'base' };
}
export function loadConfigFromHead(repoRoot: string, knownChecks: readonly string[] = []): ConfigLoad {
const file = path.join(repoRoot, PREGATE_CONFIG_PATH);
if (!fs.existsSync(file)) return { config: DEFAULT_CONFIG, errors: [], present: false, label: `head (no ${PREGATE_CONFIG_PATH})`, kind: 'none' };
return { ...parseConfig(fs.readFileSync(file, 'utf8'), knownChecks), present: true, label: `head (unreviewed) ${PREGATE_CONFIG_PATH}`, kind: 'head' };
const policy = path.join(repoRoot, SHIP_POLICY_PATH);
return { ...newerGstackErrors(parseConfig(fs.readFileSync(file, 'utf8'), knownChecks), fs.existsSync(policy) ? fs.readFileSync(policy, 'utf8') : null), present: true, label: `head (unreviewed) ${PREGATE_CONFIG_PATH}`, kind: 'head' };
}
/** A symlink to nothing on the way to `file` (a dotfiles link not yet cloned): a policy the user meant to have, not an absent one. */
@@ -365,7 +388,7 @@ export function explainConfig(): string {
'remote_gate: the repo\'s remote test gate command; under 8 cores the test tier becomes a requires-remote obligation naming it, and gstack-pregate --remote runs it (it must print PREGATE_REMOTE: tree=<tree> tests=<n> failed=<n>) (docs/pregate.md#remote-gate)',
'run_profiles: name → { workflow, events, jobs, artifact, max_age_hours?, os? }; the GitHub Actions runs gstack-pregate clear --run verifies as evidence on the identical tree (docs/pregate.md#run-evidence)',
'remote_gate_runs: { profiles, inventory, skipped? }; the run profiles that together clear the remote gate obligation, the command listing the gate\'s test files they must cover, and an optional command printing which selected files (PREGATE_FILES_FILE) the gate itself does not run, which then need no run (docs/pregate.md#run-evidence)',
'ship_lanes_from: exact /ship lane command → { profiles, inventory }; a verified run of each profile at HEAD\'s clean tree whose tree.json covers the inventory carries the lane instead of running it again (docs/pregate.md#ci-carry)',
'ship_lanes_from: exact /ship lane command → { profiles, inventory } or { profiles, jobs }, with optional aliases (other exact commands for the same lane); a verified run of each profile at HEAD\'s clean tree whose tree.json covers the inventory, or whose newest attempt ran every named job to success, carries the lane instead of running it again (docs/pregate.md#ci-carry)',
`allowlist: [{ finding, paths?, pattern?, reason }]: benign findings the secrets check accepts and names as allowlisted. finding is a gstack-redact id (env.kv, google.api_key, pii.phone.e164, …) or a category (${ALLOWLISTABLE_CATEGORIES.join(', ')}); paths are globs; pattern is a regex matched at the finding's position in the added line; one of the two is required, reason always. Never ${NON_ALLOWLISTABLE.join(' or ')} or engine.* (docs/pregate.md#allowlist)`,
'no .gstack/pregate.json on the base: <state root>/projects/<slug>/pregate.json is read instead (state root: GSTACK_STATE_ROOT, else GSTACK_HOME, else GSTACK_STATE_DIR, else a gstack plugin data dir, else ~/.gstack, per docs/state-root.md; run --explain prints the resolved path; outside the repo, so a PR cannot reach it; a path inside the checkout is ignored)',
].join('\n') + '\n';
+403
View File
@@ -0,0 +1,403 @@
/**
* pregate flaky exception (GST-1096, docs/pregate.md#flaky-exception): an owner-granted exception that lets GitHub
* Actions runs with failures clear one requires-remote obligation. `gstack-pregate clear --remote <lane> --run <run>
* --flaky-exception "<test file>::<case>" … --pr <n> --grant <approval id>` (lib/pregate/index.ts clearByRuns) takes
* the runs only when:
*
* - every failure is named from the run's own records: each job of the attempt reconciles (only test-shard jobs, and
* the aggregate that needs them, may fail), the plan and every shard's strict result are present and complete (a
* timed-out or unattributed shard is incomplete evidence), and each failed shard's log names no failure its result
* omits;
* - the named cases are exactly the failures, at most three, none in a paid test file or a test of a security scan,
* and each file is outside the PR's diff and the diff's import closure (lib/pregate/import-graph.ts);
* - the runs tested pregate.json's own head and tree (no stamp-only substitution), the PR's head is that head, and an
* approved gstack-approvals row in the owner's words carries the scope `exceptionBinding` computes over repository,
* PR, head, tree, lane, runs with their attempts and cases, so it cannot be rebound to another head or case list.
*
* It is pregate only: nothing here touches a required CI check, which still needs its own green run at the head.
* base_red is not used: it proves a failure by reproducing it on the base, and a timing flake passes there.
*/
import { createHash } from 'node:crypto';
import { spawnSync } from 'node:child_process';
import * as path from 'node:path';
import { locate, readRecord } from '../approval-record';
import type { PregateException, PregateFile } from '../headless-artifacts';
import { parseRunRef, verifyRun, type ApiJob, type ApiRun, type RunArtifactDeps, type RunDeps, type RunEvidence, type RunProfile } from '../run-evidence';
import { wtreeFingerprint } from '../script-invocation';
import { resolveBaseBranch } from '../ship-policy';
import type { FreeCiPlan, FreeCiResult, FreeShardOutcome } from '../../scripts/test-free-shards';
import { isConfigUnavailable, loadConfigFromBase } from './config';
import { buildGraph, testsFor } from './import-graph';
export type FailureDeps = RunArtifactDeps;
export interface GrantRow { id: string; outcome: string; scope: string; words: string; by: string }
export interface FlakyInput { cases: string[]; pr?: number; grant?: string; grantRow?: (id: string) => GrantRow | null }
export type FlakyRefusal = { refused: 'usage' | 'PREGATE_RECEIPT_INVALID'; message: string; fix?: string };
/** What a flaky exception is granted for; `exceptionBinding` hashes exactly these fields. */
export type ExceptionSubject = Pick<PregateException, 'lane' | 'repo' | 'pr' | 'head' | 'tree' | 'runs' | 'cases'>;
export interface FlakyRuns { evidence: RunEvidence[]; red: Array<{ evidence: RunEvidence; failures: string[] }> }
export const MAX_FLAKY_CASES = 3;
export const FLAKY_SCOPE_PREFIX = 'flaky-exception:';
/** The receipt field an exception clear adds; verify and verdictOf read it back. */
export const EXCEPTION_RECEIPT = / exception=(\S+)/;
/** The strict shard runner's artifacts and job names (scripts/test-free-shards.ts --ci-run) in the workflows that upload them. */
interface ShardLayout { plan: string; result: string; logs: string; shardJob: string; aggregate: string }
const SHARD_LAYOUTS: Record<string, ShardLayout> = {
'free-tests.yml': { plan: 'free-plan', result: 'free-result-', logs: 'free-test-shard-logs-', shardJob: 'free-suite', aggregate: 'free-tests' },
'windows-free-tests.yml': { plan: 'windows-plan', result: 'windows-result-', logs: 'windows-free-test-shard-logs-', shardJob: 'windows-free-shard', aggregate: 'windows-free-tests' },
};
/** A test of a security scanner (/cso, secrets, gitleaks, the sealed guard, redaction, OSV): never excepted. */
const SECURITY_TEST = /(?:^|[-_.])(?:cso|secrets?|gitleaks|sealed|redact\w*|osv|security|sanitize\w*|injection|inject)(?=[-_.])/;
const SHA = /^[0-9a-f]{40}$/;
const LOG_FILE_HEADER = /^(\S.*\.test\.(?:[cm]?[jt]s|tsx|jsx)):$/;
const sha256 = (text: string) => createHash('sha256').update(text).digest('hex');
const short = (s: string | undefined) => (s ?? '').slice(0, 12);
const shellQuote = (s: string) => `'${s.replace(/'/g, `'\\''`)}'`;
/** The gstack-approvals scope that grants one exception: a digest of every field it binds, cases in sorted order. */
export function exceptionBinding(e: ExceptionSubject): string {
const lines = [`lane=${e.lane}`, `repo=${e.repo}`, `pr=${e.pr}`, `head=${e.head}`, `tree=${e.tree}`, ...e.runs.map(r => `run=${r}`), ...[...e.cases].sort().map(c => `case=${c}`)];
return FLAKY_SCOPE_PREFIX + sha256(lines.join('\n'));
}
/** The approvals rows of this checkout's branch record (lib/approval-record.ts), looked up by id. */
export function approvalGrants(repoRoot: string, env: NodeJS.ProcessEnv = process.env): (id: string) => GrantRow | null {
return id => {
const loc = locate({ cwd: repoRoot, env });
return 'code' in loc ? null : readRecord(loc).rows.find(r => r.id === id) ?? null;
};
}
/** The `--flaky-exception` values as `<file>::<case>` strings, or the usage refusal. */
export function parseCases(cases: string[]): string[] | FlakyRefusal {
if (cases.length > MAX_FLAKY_CASES) return { refused: 'usage', message: `--flaky-exception names at most ${MAX_FLAKY_CASES} cases; got ${cases.length}` };
const bad = cases.find(c => { const i = c.indexOf('::'); return i <= 0 || i + 2 >= c.length; });
if (bad !== undefined) return { refused: 'usage', message: `--flaky-exception takes "<test file>::<case>", the file and the test name as the run printed them; got ${JSON.stringify(bad)}` };
if (new Set(cases).size !== cases.length) return { refused: 'usage', message: '--flaky-exception names each case once' };
return cases;
}
/** One-line summary for verify's PREGATE line: `pass-with-exception (<cases>; grant <id>)`. */
export function describeExceptions(file: PregateFile): string | null {
const list = file.exceptions ?? [];
if (!list.length) return null;
return `pass-with-exception (${list.flatMap(e => e.cases).join(', ')}; grant ${list.map(e => e.grant).join(', ')})`;
}
/** READY's line for each exception: it waives the pregate obligation, never a required CI check. */
export function flakyExceptionLines(file: PregateFile): string[] {
return (file.exceptions ?? []).map(e => `FLAKY_EXCEPTION: ${e.lane} cases=${e.cases.join(' | ')} run=${e.runs.join(',')} grant=${e.grant} by=${e.by} (pregate only; every required CI check still needs its own green run at this head)`);
}
/** verify's offline check: every exception receipt has its entry, bound to this head and tree, with an intact binding; null when all hold. */
export function exceptionProblem(file: PregateFile): string | null {
const list = file.exceptions ?? [];
const remote = file.requires_remote ?? [];
for (const r of remote) {
const grant = EXCEPTION_RECEIPT.exec(r.receipt ?? '')?.[1];
if (!grant) continue;
const e = list.find(x => x.lane === r.lane && x.grant === grant);
if (!e) return `${r.lane}'s receipt names flaky exception grant ${grant}, but pregate.json has no exceptions entry for it; clear again`;
if (e.head !== file.head || e.tree !== file.tree) return `${r.lane}'s flaky exception (grant ${grant}) is bound to head ${short(e.head)} tree ${short(e.tree)}, not pregate.json's head ${short(file.head)} tree ${short(file.tree)}; a new head needs a new run and a new grant`;
if (exceptionBinding(e) !== e.binding) return `${r.lane}'s flaky exception (grant ${grant}): the binding does not match its recorded fields (granted scope ${e.binding}); clear again`;
const runs = / runs=(\S+)/.exec(r.receipt ?? '')?.[1];
if (runs !== e.runs.join(',')) return `${r.lane}'s receipt names runs ${runs ?? 'none'}, but its flaky exception (grant ${grant}) was granted for ${e.runs.join(',')}; clear again`;
}
const orphan = list.find(e => !remote.some(r => r.lane === e.lane && EXCEPTION_RECEIPT.exec(r.receipt ?? '')?.[1] === e.grant));
return orphan ? `pregate.json records a flaky exception for ${orphan.lane} (grant ${orphan.grant}) that no receipt uses; clear again` : null;
}
/** Why `file`'s exceptions cannot stand for this PR head: each binds the PR and head it was granted for, and a same-tree commit or another PR reusing the record never inherits it. */
export function exceptionConsumerProblem(file: PregateFile, head: string, pr?: string): string | null {
const e = (file.exceptions ?? []).find(x => x.head !== head || (pr !== undefined && String(x.pr) !== pr));
return e ? `flaky exception grant ${e.grant} is bound to PR #${e.pr} at ${short(e.head)}, not ${pr ? `PR #${pr} at ` : ''}${short(head)}; a new head needs a new run and a new grant` : null;
}
/** Why red run `id` no longer stands as completed, failed attempt `attempt`; read again once the whole clear is assembled. */
export function redRunChanged(deps: RunDeps, id: number, attempt: number): string | null {
const again = deps.run(id);
return again.run_attempt === attempt && again.status === 'completed' && again.conclusion === 'failure' ? null : `run ${id} moved to attempt ${again.run_attempt} (${again.status === 'completed' ? again.conclusion : again.status}) while it was being verified`;
}
/** verify's ok-line suffix naming each exception. */
export function exceptionSummary(file: PregateFile): string {
return (file.exceptions ?? []).map(e => `flaky exception ${e.cases.join(', ')} (grant ${e.grant}, by ${e.by}, run ${e.runs.join(',')})`).join('; ');
}
const invalid = (lane: string, reason: string, detail: string, fix?: string): FlakyRefusal => ({ refused: 'PREGATE_RECEIPT_INVALID', message: `FLAKY_EXCEPTION: refused reason=${reason} — ${detail}; ${lane} stays uncleared`, ...(fix ? { fix } : {}) });
const errText = (e: unknown) => String((e as Error)?.message ?? e).slice(0, 300);
/**
* The runs for `names` (one per profile, matched by workflow): a green run verifies as `clear --run` verifies it, a red
* run through `readRunFailures`. At least one must be red, or no exception is needed.
*/
export function flakyRuns(lane: string, check: { repo: string; deps: RunDeps; profiles: Record<string, RunProfile>; now: Date; failures?: FailureDeps }, names: string[], refs: string[], vetoHead: string | undefined): FlakyRuns | FlakyRefusal {
if (!check.failures) return { refused: 'usage', message: 'the run artifacts cannot be read here (no GitHub origin), so failures cannot be named' };
const runs: Array<{ ref: string; id: number; run: ApiRun }> = [];
for (const ref of refs) {
const id = parseRunRef(ref, check.repo);
if (id === null) return invalid(lane, 'run_ref_invalid', `${JSON.stringify(ref)} is not a run id or a run URL on ${check.repo}`);
try { runs.push({ ref, id, run: check.deps.run(id) }); }
catch (e) { return invalid(lane, 'api_error', `reading run ${id} failed: ${errText(e)}`, `check gh auth status and that gh can read ${check.repo}'s Actions runs, then rerun the same clear`); }
}
if (runs.every(r => r.run.status === 'completed' && r.run.conclusion === 'success')) return invalid(lane, 'run_green', `${runs.map(r => `run ${r.id} attempt ${r.run.run_attempt}`).join(', ')} passed; clear with --run alone, no exception is needed`);
const out: FlakyRuns = { evidence: [], red: [] };
const used = new Set<number>();
for (const name of names) {
const profile = check.profiles[name]!;
const pick = runs.find(r => !used.has(r.id) && r.run.path === `.github/workflows/${profile.workflow}`) ?? runs.find(r => !used.has(r.id));
if (!pick) return { refused: 'usage', message: `${lane} needs one run per profile (${names.join(', ')})` };
used.add(pick.id);
if (pick.run.conclusion === 'success') {
const v = verifyRun(check.deps, { repo: check.repo, runRef: pick.ref, profileName: name, profile, vetoHeads: vetoHead ? [vetoHead] : [], now: check.now });
if (!v.ok) return invalid(lane, v.reason, `${name}: ${v.detail}`);
out.evidence.push(v.evidence);
continue;
}
const r = readRunFailures(check.deps, check.failures, { repo: check.repo, id: pick.id, run: pick.run, profileName: name, profile, now: check.now });
if (!r.ok) return invalid(lane, r.reason, r.detail);
out.evidence.push(r.evidence);
out.red.push({ evidence: r.evidence, failures: r.failures });
}
return out;
}
type Read<T> = { ok: true } & T | { ok: false; reason: string; detail: string };
const fail = (reason: string, detail: string) => ({ ok: false as const, reason, detail });
/** The one JSON file of artifact `name`, parsed; a message when it is missing, ambiguous or unreadable. */
function jsonArtifact<T>(art: FailureDeps, id: number, name: string): T | { error: string } {
let files: Record<string, string>;
try { files = art.artifactFiles(id, name); } catch (e) { return { error: errText(e) }; }
const json = Object.entries(files).filter(([f]) => f.endsWith('.json'));
if (json.length !== 1) return { error: `${name} holds ${json.length} JSON files, not one` };
try { return JSON.parse(json[0]![1]) as T; } catch (e) { return { error: `${json[0]![0]} is not JSON: ${errText(e)}` }; }
}
const isError = (v: unknown): v is { error: string } => !!v && typeof v === 'object' && 'error' in v && typeof (v as { error: unknown }).error === 'string';
const escapeRe = (s: string) => s.replace(/[.*+?^${{}()|[\]\\]/g, '\\$&');
/**
* A red run under `profile` read from its own records (B8: a red run uploads no tree.json): every job of the attempt, the
* plan, every shard's strict result and each failed shard's logs. Returns the evidence (the plan's files, the tests the
* shards ran) and the failed cases as `<file>::<test name>`, or the refusal naming what could not be reconciled.
*/
function readRunFailures(deps: RunDeps, art: FailureDeps, o: { repo: string; id: number; run: ApiRun; profileName: string; profile: RunProfile; now: Date }): Read<{ evidence: RunEvidence; failures: string[] }> {
const { run, profile: p, id } = o;
const layout = SHARD_LAYOUTS[p.workflow];
if (!layout) return fail('no_shard_results', `${p.workflow} uploads no strict shard results gstack can read (only ${Object.keys(SHARD_LAYOUTS).join(' and ')} do), so its failures cannot be named`);
const at = `run ${id} attempt ${run.run_attempt}`;
try {
if ((run.repository?.full_name ?? '').toLowerCase() !== o.repo.toLowerCase()) return fail('other_repository', `${at} belongs to ${run.repository?.full_name ?? 'an unknown repository'}, not ${o.repo}`);
if ((run.head_repository?.full_name ?? '').toLowerCase() !== o.repo.toLowerCase()) return fail('fork_run', `${at} ran for ${run.head_repository?.full_name ?? 'an unknown repository'}, not ${o.repo}`);
if (run.path !== `.github/workflows/${p.workflow}`) return fail('workflow_mismatch', `${at} ran ${run.path ?? 'an unknown workflow'}, not .github/workflows/${p.workflow}`);
if (!p.events.includes(run.event)) return fail('event_mismatch', `${at} is a ${run.event} run; ${o.profileName} accepts ${p.events.join(', ')}`);
if (run.status !== 'completed') return fail('run_in_progress', `${at} is ${run.status}`);
if (run.conclusion !== 'failure') return fail('run_not_failure', `${at} concluded ${run.conclusion}; only a run that finished with test failures can carry an exception (a cancelled or timed-out run is incomplete evidence)`);
if (!SHA.test(run.head_sha)) return fail('api_error', `${at} has no head commit`);
const jobs = deps.jobs(id, run.run_attempt);
const finishedAt = jobs.map(j => j.completed_at).filter((t): t is string => !!t).sort().pop() ?? run.updated_at;
const finished = Date.parse(finishedAt ?? '');
const maxAge = p.max_age_hours ?? 24;
if (!Number.isFinite(finished)) return fail('api_error', `${at} has no completion time`);
if (o.now.getTime() - finished > maxAge * 3600_000) return fail('run_stale', `${at} finished ${finishedAt}, more than ${maxAge} h before now`);
const newer = deps.runsForHead(p.workflow, run.head_sha).find(r => r.id !== id && (r.run_started_at ?? r.created_at ?? '') >= (run.created_at ?? ''));
if (newer) return fail('newer_run', `run ${newer.id} of ${p.workflow} on ${short(run.head_sha)} started after ${at}; clear with the newest run`);
const shardJobs = reconcileJobs(jobs, layout, p, at);
if ('reason' in shardJobs) return fail(shardJobs.reason, shardJobs.detail);
const plan = jsonArtifact<FreeCiPlan>(art, id, layout.plan);
if (isError(plan)) return fail('shard_incomplete', `${at} has no readable ${layout.plan} (${plan.error})`);
if (plan.version !== 1 || typeof plan.id !== 'string' || typeof plan.revision !== 'string' || !Array.isArray(plan.shards) || !plan.shards.length
|| plan.shards.some((s, i) => s.shard !== i + 1 || !Array.isArray(s.files) || s.files.some(f => typeof f !== 'string'))) return fail('result_mismatch', `${at}'s ${layout.plan} is not a strict shard plan`);
const headTree = deps.commitTree(run.head_sha);
if (!headTree) return fail('api_error', `git cannot read the tree of ${run.head_sha}`);
if (plan.revision !== run.head_sha && (!SHA.test(plan.revision) || deps.commitTree(plan.revision) !== headTree)) return fail('tree_mismatch', `${at} tested ${short(plan.revision)}, whose tree is not its head ${short(run.head_sha)}'s tree ${short(headTree)}`);
if (shardJobs.size !== plan.shards.length || plan.shards.some(s => !shardJobs.has(s.shard))) return fail('shard_mismatch', `${at} has ${layout.shardJob} job(s) for shard(s) ${[...shardJobs.keys()].sort((a, b) => a - b).join(', ') || 'none'}, but its plan has ${plan.shards.length} shard(s)`);
const failures: string[] = [];
let tests = 0;
for (const s of plan.shards) {
const res = jsonArtifact<FreeCiResult>(art, id, `${layout.result}${s.shard}`);
if (isError(res)) return fail('shard_incomplete', `shard ${s.shard} has no readable ${layout.result}${s.shard} (${res.error})`);
if (res.planId !== plan.id || res.revision !== plan.revision || res.outcome?.shard !== s.shard || JSON.stringify(res.outcome.files) !== JSON.stringify(s.files)) return fail('result_mismatch', `shard ${s.shard}'s result is not shard ${s.shard} of plan ${plan.id}`);
const named = shardFailures(res, s.shard);
if ('reason' in named) return fail(named.reason, named.detail);
const job = shardJobs.get(s.shard)!;
if ((job.conclusion === 'failure') !== (named.final.length > 0)) return fail('shard_mismatch', `shard ${s.shard}'s job concluded ${job.conclusion}, but its result names ${named.final.length} final failure(s)`);
if (job.conclusion === 'failure') {
const problem = logProblem(art, id, `${layout.logs}${s.shard}`, s.shard, s.files, named.all);
if (problem) return fail(problem.reason, problem.detail);
}
failures.push(...named.final);
tests += res.outcome.summary?.testsRan ?? 0;
}
const again = deps.run(id);
if (again.run_attempt !== run.run_attempt || again.status !== 'completed' || again.conclusion !== run.conclusion) return fail('run_changed', `run ${id} moved to attempt ${again.run_attempt} (${again.status === 'completed' ? again.conclusion : again.status}) while it was being verified`);
return { ok: true, evidence: { run_id: id, run_attempt: run.run_attempt, head_sha: run.head_sha, tree: headTree, profile: o.profileName, event: run.event, finished_at: finishedAt!, files: plan.shards.flatMap(s => s.files), tests }, failures: [...new Set(failures)] };
} catch (e) {
return fail('api_error', `reading run ${id} failed: ${errText(e)}`);
}
}
/** Shard number → job, once every job of the attempt is accounted for: only shard jobs and the aggregate may fail. */
function reconcileJobs(jobs: ApiJob[], layout: ShardLayout, p: RunProfile, at: string): Map<number, ApiJob> | { reason: string; detail: string } {
const shardRe = new RegExp(`^${escapeRe(layout.shardJob)} \\((\\d+)\\)$`);
const shards = new Map<number, ApiJob>();
for (const j of jobs) {
if (j.conclusion === null) return { reason: 'shard_incomplete', detail: `${at}'s ${j.name} job has not finished` };
const m = shardRe.exec(j.name);
if (m) {
if (shards.has(Number(m[1]))) return { reason: 'shard_mismatch', detail: `${at} has two jobs named ${j.name}` };
if (j.conclusion !== 'success' && j.conclusion !== 'failure') return { reason: 'shard_incomplete', detail: `${at}'s ${j.name} job concluded ${j.conclusion}` };
shards.set(Number(m[1]), j);
continue;
}
if (j.name === layout.aggregate) continue;
const allowed = p.jobs.includes(j.name) ? ['success'] : ['success', 'skipped', 'neutral'];
if (!allowed.includes(j.conclusion)) return { reason: 'unnamed_failure', detail: `${at}'s ${j.name} job concluded ${j.conclusion}; only test-shard failures can be named` };
}
const missing = p.jobs.find(n => !jobs.some(j => j.name === n));
if (missing) return { reason: 'job_missing', detail: `${at} has no job named ${missing}` };
const aggregate = jobs.filter(j => j.name === layout.aggregate);
if (aggregate.length !== 1) return { reason: 'job_missing', detail: `${at} has ${aggregate.length} jobs named ${layout.aggregate}` };
if (![...shards.values()].some(j => j.conclusion === 'failure')) return { reason: 'unnamed_failure', detail: `${at} failed, but no ${layout.shardJob} job did, so the failure is not a test case` };
if (aggregate[0]!.conclusion !== 'failure') return { reason: 'shard_mismatch', detail: `${at}'s ${layout.aggregate} job concluded ${aggregate[0]!.conclusion} although a shard failed` };
return shards;
}
function completeSummary(o: FreeShardOutcome): boolean {
const s = o.summary;
return !!s && Number.isInteger(s.testsRan) && s.testsRan! >= 0 && s.filesRan === o.files.length && (o.files.length === 0 || s.sawTerminalSummary === true);
}
/**
* A shard's final failures (`<file>::<test>`) and every failure it recorded. A retry that passed recovered its files; a
* retry that failed replaces the first run's failures for the files it ran. Anything unattributed or incomplete refuses.
*/
function shardFailures(r: FreeCiResult, n: number): { final: string[]; all: Set<string> } | { reason: string; detail: string } {
const o = r.outcome;
const key = (f: { file: string | null; testName: string }) => `${f.file}::${f.testName}`;
if (o.status === 'timed-out') return { reason: 'shard_incomplete', detail: `shard ${n} timed out, so the tests after the hang never ran` };
if (!completeSummary(o)) return { reason: 'shard_incomplete', detail: `shard ${n}'s result has no complete run summary` };
if (o.status === 'passed') {
if (o.exitCode !== 0 || o.failingFiles.length || o.unattributedFailures || r.retry) return { reason: 'result_mismatch', detail: `shard ${n} passed with failures recorded` };
return { final: [], all: new Set() };
}
if (o.status !== 'failed') return { reason: 'result_mismatch', detail: `shard ${n} has status ${String(o.status)}` };
if (o.unattributedFailures) return { reason: 'unnamed_failure', detail: `shard ${n}: ${o.unattributedFailures} failure(s) the runner could not attribute to a test file` };
const first = o.failures ?? [];
const unnamed = first.filter(f => !f.file || !o.files.includes(f.file) || !f.testName);
if (!first.length || unnamed.length) return { reason: 'unnamed_failure', detail: `shard ${n} failed ${first.length ? `with ${unnamed.length} failure(s) not attributed to one of its files` : 'without naming a failure'}` };
if (JSON.stringify([...new Set(first.map(f => f.file))].sort()) !== JSON.stringify([...o.failingFiles].sort())) return { reason: 'result_mismatch', detail: `shard ${n}'s named failures and failing files disagree` };
const all = new Set(first.map(key));
const t = r.retry;
if (!t) return { final: [...all], all };
if (t.status === 'timed-out' || !completeSummary(t)) return { reason: 'shard_incomplete', detail: `shard ${n}'s retry did not complete` };
if (t.files.some(f => !o.failingFiles.includes(f))) return { reason: 'result_mismatch', detail: `shard ${n}'s retry ran files that had not failed` };
const again = t.failures ?? [];
if (t.unattributedFailures || (t.status === 'failed' && (!again.length || again.some(f => !f.file || !t.files.includes(f.file))))) return { reason: 'unnamed_failure', detail: `shard ${n}'s retry failed without naming each failure` };
if (t.status === 'passed' && (again.length || t.exitCode !== 0)) return { reason: 'result_mismatch', detail: `shard ${n}'s retry passed with failures recorded` };
for (const f of again) all.add(key(f));
const final = [...first.filter(f => !t.files.includes(f.file!)), ...again].map(key);
return { final: [...new Set(final)], all };
}
/** Why a failed shard's logs do not reconcile with its result: missing, an unhandled error between tests, or a failure the result omits. */
function logProblem(art: FailureDeps, id: number, name: string, n: number, planned: string[], recorded: Set<string>): { reason: string; detail: string } | null {
let files: Record<string, string>;
try { files = art.artifactFiles(id, name); } catch (e) { return { reason: 'logs_missing', detail: `shard ${n} failed and has no readable ${name} (${errText(e)})` }; }
const logs = Object.entries(files).filter(([f]) => f.endsWith('.log'));
if (!logs.length) return { reason: 'logs_missing', detail: `shard ${n} failed and its ${name} holds no .log file` };
const canonical = (printed: string) => {
const stripped = printed.replace(/\\/g, '/').replace(/^(?:\.{1,2}\/)+/, '');
return planned.includes(stripped) ? stripped : planned.find(p => stripped.endsWith(`/${p}`) || p.endsWith(`/${stripped}`)) ?? stripped;
};
const { classifyBunTestOutputLine, parseBunFailureResult, stripAnsiLine } = require('../../scripts/lib/shard-engine') as typeof import('../../scripts/lib/shard-engine');
for (const [, text] of logs) {
let file: string | null = null;
let recap = false;
let unhandled = 0;
for (const raw of text.split('\n')) {
const line = stripAnsiLine(raw).replace(/^::group::/, '').replace(/\r$/, '');
if (/^\d+ tests? failed:$/.test(line)) { recap = true; file = null; continue; }
if (classifyBunTestOutputLine(line) === 'unhandled-between-tests') { unhandled++; continue; }
const header = LOG_FILE_HEADER.exec(line);
if (header) { if (!recap) file = canonical(header[1]!); continue; }
const failed = recap ? null : parseBunFailureResult(line);
if (failed !== null && !recorded.has(`${file}::${failed}`)) return { reason: 'unnamed_failure', detail: `shard ${n}'s log names ${file}::${failed}, which its result does not` };
}
if (unhandled) return { reason: 'unnamed_failure', detail: `shard ${n}'s log shows ${unhandled} unhandled error(s) between tests` };
}
return null;
}
/** The PR's changed paths against the merge base with origin/<base>, and each test the diff reaches through the import graph. */
function prScope(repoRoot: string, file: PregateFile): { diff: Set<string>; closure: Map<string, string[]> } | { error: string } {
const git = (args: string[]) => { const r = spawnSync('git', args, { cwd: repoRoot, encoding: 'utf8', timeout: 60_000, maxBuffer: 64 * 1024 * 1024 }); return r.status === 0 ? r.stdout : null; };
const base = resolveBaseBranch(repoRoot);
const mb = git(['merge-base', `refs/remotes/origin/${base}`, '--end-of-options', file.head!])?.trim();
if (!mb) return { error: `git cannot find the merge base of origin/${base} and ${short(file.head)}` };
const listed = git(['diff', '--name-status', '-z', '--no-renames', '--end-of-options', mb, file.head!]);
if (listed === null) return { error: `git diff ${short(mb)} ${short(file.head)} failed` };
const parts = listed.split('\0').filter(Boolean);
const diff = new Set<string>();
const deleted: string[] = [];
for (let i = 0; i + 1 < parts.length; i += 2) { diff.add(parts[i + 1]!); if (parts[i] === 'D') deleted.push(parts[i + 1]!); }
const loaded = loadConfigFromBase(repoRoot, base);
if (isConfigUnavailable(loaded) || loaded.errors.length) return { error: `the base branch's .gstack/pregate.json cannot be read, so its dependency declarations are unknown` };
const at = () => { const fp = wtreeFingerprint({ cwd: repoRoot }); return 'tree' in fp ? fp.tree : null; };
if (at() !== file.tree) return { error: `the working tree is not pregate.json's tree ${short(file.tree)}; the import closure must be read from that tree` };
const declarations = loaded.config.dependencies;
const graph = buildGraph(repoRoot, { declarations, extraFiles: deleted });
const closure = new Map<string, string[]>();
for (const changed of diff) for (const t of testsFor(graph, changed, declarations).tests) closure.set(t, [...(closure.get(t) ?? []), changed]);
if (at() !== file.tree) return { error: 'the working tree changed while the import closure was read' };
return { diff, closure };
}
/**
* After the runs verified and covered the obligation: the named cases must be exactly the failures, in no paid or
* security test, outside the PR's diff and import closure, the PR's head must be pregate.json's, and the grant row must
* carry this exception's binding in the owner's words. Returns the exception to record.
*/
export function flakyDecision(file: PregateFile, lane: string, check: { repo: string; failures?: FailureDeps }, runs: FlakyRuns, input: FlakyInput & { cases: string[] }, repoRoot: string): { exception: PregateException } | FlakyRefusal {
const label = runs.red.map(r => `run ${r.evidence.run_id} attempt ${r.evidence.run_attempt}`).join(', ');
const failures = [...new Set(runs.red.flatMap(r => r.failures))];
const extra = failures.filter(f => !input.cases.includes(f));
if (extra.length) return invalid(lane, 'unnamed_failure', `${label} failed ${extra.slice(0, 5).join(', ')}${extra.length > 5 ? ` and ${extra.length - 5} more` : ''}, which no --flaky-exception names`);
const notFailed = input.cases.find(c => !failures.includes(c));
if (notFailed) return invalid(lane, 'case_not_failed', `${notFailed} did not fail in ${label}; its failures are ${failures.join(', ')}`);
const files = [...new Set(input.cases.map(c => c.slice(0, c.indexOf('::'))))];
const { isPaidTestFile } = require('../../test/helpers/paid-test-set') as typeof import('../../test/helpers/paid-test-set');
const paid = files.find(f => isPaidTestFile(f));
if (paid) return invalid(lane, 'case_paid', `${paid} is a paid test file; the paid gate covers it, never an exception`);
const security = files.find(f => SECURITY_TEST.test(path.posix.basename(f)));
if (security) return invalid(lane, 'case_security', `${security} tests a security scan, which an exception never covers`);
let pr: { head_sha: string; state: string; head_repo: string | null };
try { pr = check.failures!.pull(input.pr!); }
catch (e) { return invalid(lane, 'api_error', `reading PR #${input.pr} failed: ${errText(e)}`); }
if (pr.state !== 'open') return invalid(lane, 'pr_mismatch', `PR #${input.pr} is ${pr.state}`);
if ((pr.head_repo ?? '').toLowerCase() !== check.repo.toLowerCase()) return invalid(lane, 'pr_mismatch', `PR #${input.pr}'s head is on ${pr.head_repo ?? 'an unknown repository'}, not ${check.repo}`);
if (pr.head_sha !== file.head) return invalid(lane, 'pr_mismatch', `PR #${input.pr}'s head is ${short(pr.head_sha)}, not pregate.json's head ${short(file.head)}`);
const scope = prScope(repoRoot, file);
if ('error' in scope) return invalid(lane, 'closure_failed', scope.error);
const touched = files.find(f => scope.diff.has(f));
if (touched) return invalid(lane, 'case_in_diff', `${touched} is in the PR's diff`);
const reached = files.find(f => scope.closure.has(f));
if (reached) return invalid(lane, 'case_in_closure', `${reached} depends on a file the PR changes (${scope.closure.get(reached)!.slice(0, 5).join(', ')})`);
const subject: ExceptionSubject = { lane, repo: check.repo, pr: input.pr!, head: file.head!, tree: file.tree, runs: runs.evidence.map(e => `${e.run_id}/${e.run_attempt}`), cases: [...input.cases].sort() };
const binding = exceptionBinding(subject);
if (!input.grant) {
const ask = `Accept flaky ${subject.cases.join(', ')} in run ${subject.runs.join(',')} for PR #${subject.pr} at ${short(subject.head)}? Pregate only; required CI still needs a green run.`.slice(0, 300);
return invalid(lane, 'grant_needed', `every failure in ${label} is named (${subject.cases.join(', ')}) and outside the PR's diff and its import closure; the owner must grant the exception`,
`ask the owner, record their words with gstack-approvals add --ask ${shellQuote(ask)} --words "<the owner's exact words>" --by "<owner>" --scope "${binding}" --source pregate-flaky-exception, then rerun this clear with --grant <the row id it prints>`);
}
const row = input.grantRow?.(input.grant) ?? null;
if (!row) return invalid(lane, 'grant_missing', `no gstack-approvals row ${input.grant} in this branch's record`);
if (row.scope !== binding) return invalid(lane, 'grant_mismatch', `row ${row.id}'s scope ${JSON.stringify(row.scope.slice(0, 80))} is not this exception's binding ${binding}; a grant covers one repository, PR, head, tree, lane, run attempt and case list, so a new head needs a new grant`);
if (row.outcome !== 'approved') return invalid(lane, 'grant_not_approved', `row ${row.id} records outcome ${row.outcome}, not approved`);
if (/^agent:/i.test(row.by)) return invalid(lane, 'grant_not_owner', `row ${row.id} records an agent's words (${row.by}); only the owner grants an exception`);
return { exception: { ...subject, grant: row.id, by: row.by, words: row.words, binding } };
}
+35 -11
View File
@@ -14,7 +14,7 @@ import { spawnSync } from 'node:child_process';
import { createHash } from 'node:crypto';
import * as fs from 'node:fs';
import * as path from 'node:path';
import { ARTIFACT_SCHEMAS, checkSchema, type PregateFile, type PregateReceiptSubstitution, type PregateRemote, type ValidationError } from '../headless-artifacts';
import { ARTIFACT_SCHEMAS, checkSchema, type PregateException, type PregateFile, type PregateReceiptSubstitution, type PregateRemote, type ValidationError } from '../headless-artifacts';
import { wtreeFingerprint, type WtreeResult } from '../script-invocation';
import { resolveStateRoot } from '../state-root';
import { guardsCheck } from './checks/guards';
@@ -28,6 +28,7 @@ import { timingCheck } from './checks/timing';
import { NON_DOWNGRADABLE, PREGATE_CONFIG_PATH, userPolicyAbsent } from './config';
import { sha256, withFingerprint } from './context';
import { baseBranchOf, baseMismatchFix, describeSubstitution, substituteStampOnlyTree } from './remote-receipt';
import { EXCEPTION_RECEIPT, exceptionProblem, exceptionSummary, flakyDecision, flakyRuns, parseCases, redRunChanged, type FailureDeps, type FlakyInput } from './flaky-exception';
import { buildRunCheck } from './run-check';
import { resolveBaseBranch } from '../ship-policy';
import { parseRunReceipt, parseRunRef, runChanged, runReceiptFields, uncoveredFiles, verifyRun, type ApiJob, type ApiRun, type RemoteGateRuns, type RunDeps, type RunEvidence, type RunProfile } from '../run-evidence';
@@ -139,6 +140,7 @@ export function verdictOf(file: PregateFile): string {
if (statuses.has('incomplete') || statuses.has('not_installed')) return 'incomplete';
const uncleared = (file.requires_remote ?? []).filter(r => !r.receipt);
if (uncleared.length) return `requires-remote (${uncleared.length} uncleared)`;
if ((file.requires_remote ?? []).some(r => EXCEPTION_RECEIPT.test(r.receipt ?? ''))) return 'pass-with-exception';
if (statuses.has('requires-remote')) return 'requires-remote cleared';
if (statuses.has('warn')) return 'warn';
return 'pass';
@@ -239,8 +241,12 @@ export function verifyPregate(repoRoot: string, file: PregateFile, treeNow: Wtre
// A run receipt was verified online at clear for its tree; offline it must still name this tree (or carry its substitution).
const foreignRuns = (file.requires_remote ?? []).filter(r => { const run = parseRunReceipt(r.receipt ?? ''); return run && !r.substitution && run.tree !== file.tree.slice(0, 12); });
if (foreignRuns.length) return { ok: false, code: 'PREGATE_REMOTE_UNCLEARED', message: `${foreignRuns.length} run receipt(s) verified for another tree: ${foreignRuns.map(r => `${r.lane} (${r.receipt})`).join(', ')}; clear again with runs on tree ${file.tree.slice(0, 12)}`, current, uncleared: foreignRuns };
// A flaky exception was granted for one head, tree, run attempt and case list; recorded for another, or edited, it clears nothing.
const exception = exceptionProblem(file);
if (exception) return { ok: false, code: 'PREGATE_REMOTE_UNCLEARED', message: exception, current, uncleared: [] };
const substituted = (file.requires_remote ?? []).filter(r => r.substitution).map(r => `${r.lane}: ${describeSubstitution(r.substitution!)}`);
return { ok: true, message: `pregate ${file.verdict} is current for tree ${file.tree.slice(0, 12)}${substituted.length ? `; ${substituted.join('; ')}` : ''}`, current, uncleared };
const excepted = exceptionSummary(file);
return { ok: true, message: `pregate ${file.verdict} is current for tree ${file.tree.slice(0, 12)}${substituted.length ? `; ${substituted.join('; ')}` : ''}${excepted ? `; ${excepted}` : ''}`, current, uncleared };
}
/**
@@ -264,10 +270,10 @@ function userPolicyChange(file: PregateFile): string | null {
return `${absent} now holds a user-owned policy, and pregate.json was recorded under the defaults; rerun gstack-pregate`;
}
export interface ClearInput { runUrl?: string; bundleLane?: string; receiptPath?: string; runs?: string[]; evidence?: RunCheck }
export interface ClearInput { runUrl?: string; bundleLane?: string; receiptPath?: string; runs?: string[]; evidence?: RunCheck; flaky?: FlakyInput }
/** What a run clear needs: the base branch's run profiles, the verifier's dependencies, the clock, and the remote gate's inventory. */
export interface RunCheck { repo: string; deps: RunDeps; profiles: Record<string, RunProfile>; gateRuns: RemoteGateRuns | null; now: Date; inventory: (command: string, tree: string, selected?: string[]) => string[] | { error: string } }
export type ClearDecision = { receipt: string; attested: boolean; substitution?: PregateReceiptSubstitution } | { refused: 'usage' | 'PREGATE_RECEIPT_INVALID'; message: string; fix?: string };
export interface RunCheck { repo: string; deps: RunDeps; profiles: Record<string, RunProfile>; gateRuns: RemoteGateRuns | null; now: Date; unchanged?: (p: RunProfile) => string | null; inventory: (command: string, tree: string, selected?: string[]) => string[] | { error: string }; failures?: FailureDeps }
export type ClearDecision = { receipt: string; attested: boolean; substitution?: PregateReceiptSubstitution; exception?: PregateException } | { refused: 'usage' | 'PREGATE_RECEIPT_INVALID'; message: string; fix?: string };
/**
* What `clear` may record for one obligation. A `pregate.json/remote_gate`
@@ -289,12 +295,21 @@ export type ClearDecision = { receipt: string; attested: boolean; substitution?:
* `--run` (and `--run-url` on a lane the base branch profiles) goes to
* clearByRuns, which accepts only verified Actions runs; a profiled lane
* refuses `--bundle-lane`, and the remote gate clears by runs only when the
* base declares remote_gate_runs.
* base declares remote_gate_runs. `flaky` (`--flaky-exception`) takes runs with failures the same way under an owner's
* grant (lib/pregate/flaky-exception.ts).
*/
export function clearDecision(file: PregateFile, lane: string, input: ClearInput, repoRoot: string): ClearDecision {
const list = file.requires_remote ?? [];
const target = list.find(r => r.lane === lane);
if (!target) return { refused: 'usage', message: `no requires-remote obligation named ${lane}; obligations: ${list.map(r => r.lane).join(', ') || '(none)'}` };
if (input.flaky) {
const cases = parseCases(input.flaky.cases);
if (!Array.isArray(cases)) return cases;
const refs = [...(input.runs ?? []), ...(input.runUrl ? [input.runUrl] : [])];
if (!refs.length) return { refused: 'usage', message: '--flaky-exception needs --run <id|url>: an exception reads GitHub Actions runs' };
if (!input.flaky.pr) return { refused: 'usage', message: '--flaky-exception needs --pr <n>, the pull request the grant binds' };
return clearByRuns(file, lane, refs, input.evidence, repoRoot, { ...input.flaky, cases });
}
const profiled = input.evidence && (lane === REMOTE_GATE_LANE ? input.evidence.gateRuns : input.evidence.profiles[lane]);
if (input.runs?.length || (input.runUrl && profiled)) return clearByRuns(file, lane, [...(input.runs ?? []), ...(input.runUrl ? [input.runUrl] : [])], input.evidence, repoRoot);
if (input.bundleLane && profiled && lane !== REMOTE_GATE_LANE) return { refused: 'usage', message: `${lane} has a run profile on the base branch; clear it with --run <id|url>, which verifies the run` };
@@ -377,7 +392,7 @@ function treeSubstitution(file: PregateFile, gatedTree: string, lane: string, re
* does not run (never an inventory file); any other lane needs one verified run of the profile named after it. The run
* tree must be pregate.json's tree or a stamp-only predecessor of it, exactly as a gate report's tree must.
*/
function clearByRuns(file: PregateFile, lane: string, refs: string[], check: RunCheck | undefined, repoRoot: string): ClearDecision {
function clearByRuns(file: PregateFile, lane: string, refs: string[], check: RunCheck | undefined, repoRoot: string, flaky?: FlakyInput & { cases: string[] }): ClearDecision {
if (!check) return { refused: 'usage', message: `--run needs the base branch's run_profiles; none could be read` };
const isGate = lane === REMOTE_GATE_LANE;
const names = isGate ? check.gateRuns?.profiles ?? [] : check.profiles[lane] ? [lane] : [];
@@ -409,13 +424,17 @@ function clearByRuns(file: PregateFile, lane: string, refs: string[], check: Run
}
return null;
};
const evidence = assign(0, new Set());
const red = flaky ? flakyRuns(lane, check, names, refs, file.head) : null;
if (red && 'refused' in red) return red;
const evidence = red ? red.evidence : assign(0, new Set());
if (!evidence) {
const failed = names.map(name => `${name}: ${refs.map(ref => { const v = verdict(name, ref); return v.ok ? `${ref} ok` : `${ref} reason=${v.reason} ${v.detail}`; }).join('; ')}`);
return { refused: 'PREGATE_RECEIPT_INVALID', message: `REUSE: refused reason=no_verified_run — no run per profile verifies (${failed.join(' | ')}); ${lane} stays uncleared` };
}
const trees = [...new Set(evidence.map(e => e.tree))];
if (trees.length > 1) return { refused: 'PREGATE_RECEIPT_INVALID', message: `REUSE: refused reason=tree_mismatch — the runs tested different trees (${trees.map(t => t.slice(0, 12)).join(', ')}); ${lane} stays uncleared` };
const off = red && evidence.find(e => e.head_sha !== file.head || e.tree !== file.tree);
if (off) return { refused: 'PREGATE_RECEIPT_INVALID', message: `FLAKY_EXCEPTION: refused reason=head_mismatch — run ${off.run_id} attempt ${off.run_attempt} tested head ${off.head_sha.slice(0, 12)} tree ${off.tree.slice(0, 12)}, not pregate.json's head ${(file.head ?? '').slice(0, 12)} tree ${file.tree.slice(0, 12)}; an exception binds the exact head and tree, never a stamp-only predecessor; ${lane} stays uncleared` };
const sub = treeSubstitution(file, trees[0]!, lane, repoRoot);
if ('refused' in sub) return { ...sub, message: `REUSE: refused reason=tree_not_stamp_only — ${sub.message}` };
const idle = evidence.filter(e => (e.tests ?? e.files.length) === 0);
@@ -438,24 +457,29 @@ function clearByRuns(file: PregateFile, lane: string, refs: string[], check: Run
}
const missing = uncoveredFiles([...required], evidence);
if (missing.length) return { refused: 'PREGATE_RECEIPT_INVALID', message: `REUSE: refused reason=${isGate ? 'inventory_uncovered' : 'tests_uncovered'} — ${missing.length} of the ${required.size} test file(s) ${lane} must cover appear in no run's tree.json (${missing.slice(0, 5).join(', ')}${missing.length > 5 ? ', …' : ''}); ${lane} stays uncleared` };
const granted = red ? flakyDecision(file, lane, check, red, flaky!, repoRoot) : null;
if (granted && 'refused' in granted) return granted;
for (const e of evidence) {
let changed: string | null;
try { changed = runChanged(check.deps, e.run_id, e.run_attempt); }
try { changed = red?.red.some(r => r.evidence === e) ? redRunChanged(check.deps, e.run_id, e.run_attempt) : runChanged(check.deps, e.run_id, e.run_attempt); }
catch (err: any) { changed = `reading run ${e.run_id} again failed: ${String(err?.message ?? err).slice(0, 200)}`; }
if (changed) return { refused: 'PREGATE_RECEIPT_INVALID', message: `REUSE: refused reason=run_changed — ${changed}; ${lane} stays uncleared` };
}
const tests = evidence.reduce((n, e) => n + (e.tests ?? e.files.length), 0);
const accepted = sub.substitution ? ` accepted-for=${file.tree.slice(0, 12)} stamp-only=${sub.substitution.files.length}` : '';
return { receipt: `${isGate ? 'remote-gate' : 'run'} tree=${trees[0]!.slice(0, 12)} tests=${tests}${skippedCount ? ` skipped=${skippedCount}` : ''}${accepted} ${runReceiptFields(evidence, check.now)}`, attested: false, substitution: sub.substitution };
const exception = granted?.exception;
return { receipt: `${isGate ? 'remote-gate' : 'run'} tree=${trees[0]!.slice(0, 12)} tests=${tests}${skippedCount ? ` skipped=${skippedCount}` : ''}${exception ? ` exception=${exception.grant} cases=${exception.cases.length}` : ''}${accepted} ${runReceiptFields(evidence, check.now)}`, attested: false, substitution: sub.substitution, ...(exception ? { exception } : {}) };
}
/** Record the receipt that clears one obligation (a run URL, an evidence-bundle lane label, or a remote gate report, with the stamp-only substitution when one was accepted). */
export function clearObligation(file: PregateFile, lane: string, receipt: string, by: string, substitution?: PregateReceiptSubstitution): PregateFile {
export function clearObligation(file: PregateFile, lane: string, receipt: string, by: string, substitution?: PregateReceiptSubstitution, exception?: PregateException): PregateFile {
const list = file.requires_remote ?? [];
const target = list.find(r => r.lane === lane);
if (!target) throw new Error(`no requires-remote obligation named ${lane}; obligations: ${list.map(r => r.lane).join(', ') || '(none)'}`);
target.receipt = receipt; target.cleared_by = by;
if (substitution) target.substitution = substitution; else delete target.substitution;
const exceptions = [...(file.exceptions ?? []).filter(e => e.lane !== lane), ...(exception ? [exception] : [])];
if (exceptions.length) file.exceptions = exceptions; else delete file.exceptions;
file.verdict = verdictOf(file);
return file;
}
+46 -7
View File
@@ -11,27 +11,66 @@ import { spawnSync } from 'node:child_process';
import * as fs from 'node:fs';
import * as os from 'node:os';
import * as path from 'node:path';
import { ghRunDeps, runInventory } from '../run-evidence';
import { ghArtifactDeps, ghRunDeps, runInventory, type RunProfile } from '../run-evidence';
import { wtreeFingerprint } from '../script-invocation';
import { resolveBaseBranch } from '../ship-policy';
import { normalizeRemoteUrl } from '../ship-queue-train-record';
import { REMOTE_GATE_LANE } from './checks/lanes';
import { PREGATE_CONFIG_PATH, isConfigUnavailable, loadConfigFromBase } from './config';
import { PREGATE_CONFIG_PATH, isConfigUnavailable, loadConfigFromBase, loadConfigFromHead } from './config';
import type { RunCheck } from './index';
/**
* `profiled` says whether `lane` may have a run profile: an unreadable base counts as profiled, so a lane is
* recorded unverified only once the base shows it has none.
* Null when a skippable profile's workflow, its declared `scripts` and every repo path its gate job names are byte-identical
* at HEAD and origin/<base> (GST-1190); else the first that differs. Named paths are read from the gate job's block in the base
* workflow (after `${{ … }}` prefixes, and also under each working-directory), plus package.json when the job runs a package
* script. Files the gate reaches indirectly (imports, a package script's own files) count only when `scripts` declares them.
*/
export function buildRunCheck(root: string, lane: string | null, allowRepoCommands: boolean, base = resolveBaseBranch(root)): RunCheck | { error: string; profiled?: boolean } {
const loaded = loadConfigFromBase(root, base);
export function gateInputsChanged(root: string, base: string, p: RunProfile): string | null {
const ref = `origin/${base.replace(/^origin\//, '')}`;
const wf = `.github/workflows/${p.workflow}`;
const gate = p.skippable?.gate;
if (!gate) return 'the profile names no gate job';
const blob = (rev: string, f: string) => { const r = spawnSync('git', ['rev-parse', '--verify', '--quiet', `${rev}:${f}`], { cwd: root, encoding: 'utf8', timeout: 30_000 }); return r.status === 0 ? r.stdout.trim() : null; };
const show = spawnSync('git', ['show', `${ref}:${wf}`], { cwd: root, encoding: 'utf8', timeout: 30_000, maxBuffer: 8 * 1024 * 1024 });
if (show.status !== 0) return `${wf} is not on ${ref}`;
const lines = show.stdout.split('\n');
const jobsAt = lines.findIndex(l => /^jobs:\s*$/.test(l));
const blocks: string[][] = [];
for (let i = jobsAt + 1; jobsAt >= 0 && i < lines.length && !/^\S/.test(lines[i]!); i++) {
if (/^ [\w-]+:\s*$/.test(lines[i]!)) blocks.push([lines[i]!]);
else blocks.at(-1)?.push(lines[i]!);
}
const block = blocks.find(b => b[0]!.trim() === `${gate}:` || b.some(l => /^ name:/.test(l) && l.trim() === `name: ${gate}`));
if (!block) return `${wf} on ${ref} has no job ${gate}`;
const text = block.join('\n');
const dirs = ['', ...[...text.matchAll(/working-directory:\s*['"]?([\w.\/-]+)/g)].map(m => `${m[1]!.replace(/^\.?\/+|\/+$/g, '')}/`)];
const tokens = [...text.matchAll(/[\w.\/-]+/g)].map(m => m[0]).filter(t => t.includes('/') || /\.(sh|ts|js|mjs|cjs|py|rb)$/.test(t));
const named = [...new Set(tokens.flatMap(t => dirs.map(d => path.posix.normalize(d + t.replace(/^\.?\/+/, ''))).filter(n => !n.startsWith('..'))))];
const pkg = /\b(bun|npm|pnpm|yarn)\s+(run\s+)?[\w:-]+/.test(text) ? dirs.map(d => `${d}package.json`) : [];
for (const f of [wf, ...(p.skippable!.scripts ?? []), ...named, ...pkg]) {
const a = blob(ref, f);
const b = blob('HEAD', f);
if (a === null && b === null && f !== wf && !p.skippable!.scripts?.includes(f)) continue;
if (a === null || a !== b) return `${f} differs from ${ref}`;
}
return null;
}
/**
* `profiled` says whether `lane` may have a run profile: an unreadable base counts as profiled, so a lane is
* recorded unverified only once the base shows it has none. `policy: 'head'` reads the branch's profiles instead, only
* for gstack-ship-preflight --simulate, a dry run that records and clears nothing.
*/
export function buildRunCheck(root: string, lane: string | null, allowRepoCommands: boolean, base = resolveBaseBranch(root), policy: 'base' | 'head' = 'base'): RunCheck | { error: string; profiled?: boolean } {
const loaded = policy === 'head' ? loadConfigFromHead(root) : loadConfigFromBase(root, base);
if (isConfigUnavailable(loaded)) return { error: `the base branch's ${PREGATE_CONFIG_PATH} is unreadable (${loaded.detail}); run profiles come only from the base`, profiled: true };
if (loaded.errors.length) return { error: `the base branch's ${PREGATE_CONFIG_PATH} is invalid: ${loaded.errors.map(e => `${e.path}: ${e.message}`).join('; ')}`, profiled: true };
const origin = spawnSync('git', ['config', '--get', 'remote.origin.url'], { cwd: root, encoding: 'utf8', timeout: 30_000 });
const repo = /^https:\/\/github\.com\/(.+)$/.exec(normalizeRemoteUrl(origin.status === 0 ? origin.stdout.trim() : ''))?.[1];
if (!repo) return { error: 'origin is not a github.com repository; run evidence is read from GitHub Actions', profiled: lane === null ? false : lane === REMOTE_GATE_LANE ? !!loaded.config.remote_gate_runs : !!loaded.config.run_profiles[lane] };
return {
repo, deps: ghRunDeps(repo, root), profiles: loaded.config.run_profiles, gateRuns: loaded.config.remote_gate_runs, now: new Date(),
repo, deps: ghRunDeps(repo, root), failures: ghArtifactDeps(repo), profiles: loaded.config.run_profiles, gateRuns: loaded.config.remote_gate_runs, now: new Date(),
unchanged: p => gateInputsChanged(root, base, p),
inventory: (command, tree, selected) => {
if (!allowRepoCommands) return { error: 'the inventory command is a pin-file command; pass --allow-repo-commands' };
const at = () => { const fp = wtreeFingerprint({ cwd: root }); return 'tree' in fp ? fp.tree : null; };
+15 -1
View File
@@ -661,13 +661,27 @@ const SCHEMA_FIELD_VALUE = /^[a-z][a-z0-9]*(?:[_.-][a-z][a-z0-9]*)+$/;
function isSchemaFieldIdentifier(span: string, match: RegExpExecArray): boolean {
return SCHEMA_FIELD_NAME.test(match[0]) && SCHEMA_FIELD_VALUE.test(span);
}
/**
* GST-1064: the first argument of a lookup call (`get`, `getenv`, `read`,
* `load`, `loads`, `open`, `pop`, `fetch`) that is a lowercase file name
* (`files.get("key.json")`) or a snake_case key of two or three words
* (`resp.json().get("access_token")`) names what the call reads. Eight distinct
* letters already reach 3.0 bits, so entropy alone cannot tell it from a
* secret. Any other callee (`make_password("orange_monkey_pass")`), a grouping
* paren, a later argument or a fallback (`os.getenv("X", "orange_monkey_pass")`)
* is a value and is still checked, and so is a first argument with a digit, an
* uppercase letter, kebab words or four or more snake_case words.
*/
const LOOKUP_ARGUMENT =
/(?<![\w$])(get|getenv|read|load|loads|open|pop|fetch)\([ \t]*(["'])(?:[a-z]+(?:_[a-z]+){1,2}|[a-z]+(?:[_-][a-z]+)*\.(?:json|ya?ml|toml|ini|cfg|conf|env|pem|crt|cer|key|txt|csv|xml|cookie))\2/g;
function isCallExpression(span: string, match: RegExpExecArray): boolean {
if (!CALL_SHAPED_VALUE.test(span)) return false;
const { start } = spanBounds(match);
if (match.input[start - 1] === '"' || match.input[start - 1] === "'") return false;
const call = match.input.slice(start + span.indexOf("(")).split("\n", 1)[0];
if (!call.includes(")") && call.trim() !== "(") return false;
return !carriesSecretLiteral(call);
const line = match.input.slice(start).split("\n", 1)[0];
return !carriesSecretLiteral(line.replace(LOOKUP_ARGUMENT, "$1("));
}
/**
+1 -1
View File
@@ -313,7 +313,7 @@ export function branchAddedLines(cwd: string, since: string, files: string[], to
return out;
}
const escapeRegExp = (t: string) => t.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
export const escapeRegExp = (t: string) => t.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
const VERSION_RE = '(?<v>\\d+(?:\\.\\d+){1,3})(?![0-9.])';
/**
* The built-in stamp forms as scanners, the version in the `v` group: `(was X)`, `since: X` and `vX`.
+5
View File
@@ -16,6 +16,11 @@ export const RELEASE_RESULT_CODES = {
summary: 'the ship policy sets min_gstack_version above this gstack, so this gstack refuses to read it',
fix: 'upgrade gstack (bin/gstack-upgrade, or the session install with --for <skill>) to the version named, then rerun',
},
PREGATE_GSTACK_TOO_OLD: {
anchor: 'pregate-gstack-too-old',
summary: 'the pregate policy uses a key this gstack does not know, and the project floor (min_gstack_version in .gstack/ship-policy.json) is above this gstack, so the key belongs to a newer gstack; the policy stays refused',
fix: 'upgrade gstack (bin/gstack-upgrade, or the session install with --for <skill>) to at least the version named, then rerun',
},
QUEUE_NOT_READY: {
anchor: 'queue-not-ready',
summary: 'the lane is not ready to report: the working tree, HEAD or CI state does not match the pushed PR head',
+12 -5
View File
@@ -10,6 +10,7 @@
import { spawnSync } from 'node:child_process';
import * as fs from 'node:fs';
import * as path from 'node:path';
import { resolveBase, type ResolvedBase } from './gate-diff/git';
import { GENERATED_PATH, parseRegistry } from './regen';
import { RESULT_CODES } from './result-codes';
import { classifyPath, dataCandidate, svgCandidate, svgDrawing, type ClassInput, type DepthError, type PathLines, type RegistryInfo } from './review-depth';
@@ -31,6 +32,8 @@ export interface Collected extends ClassInput {
untracked: string[];
consumed: string[] | 'all';
drawings: string[];
/** collectLive only: the ref resolveBase read for the base name (GST-1036), null when nothing resolved. */
base?: ResolvedBase | null;
}
function git(root: string, args: string[]): { ok: boolean; status: number | null; out: string; err: string } {
@@ -255,16 +258,20 @@ export function collectCommit(root: string, baseSha: string, headSha: string): C
return result(diff, { root: skillDirs, registry, errors, mergeBase, untracked: [], consumed, drawings });
}
/** The working tree against the merge base with origin/<base> (or <base>; a remote-qualified base reads refs/remotes/<base> only), untracked files counted per path; the registry is the base ref's commit. */
/**
* The working tree against the merge base with the base resolveBase reads (lib/gate-diff/git.ts: a bare name is
* origin/<base> when it shares history with HEAD, else the local branch; a remote-qualified base reads
* refs/remotes/<base> only), untracked files counted per path; the registry is the base ref's commit.
*/
export function collectLive(root: string, base: string): Collected {
const errors: DepthError[] = [];
const qualified = base.startsWith('origin/') || git(root, ['remote']).out.split('\n').some(r => r && base.startsWith(`${r}/`));
const ref = (qualified ? [`refs/remotes/${base}`] : [`refs/remotes/origin/${base}`, `refs/heads/${base}`]).find(r => git(root, ['rev-parse', '--verify', '-q', `${r}^{commit}`]).ok) ?? base;
const resolved = resolveBase(base, root, 'HEAD');
const ref = resolved?.sha ?? base;
const registry = loadRegistryAt(root, ref);
const mb = git(root, ['merge-base', ref, 'HEAD']);
if (!mb.ok) {
errors.push({ code: 'numstat_failed', detail: oneLine(mb.err || `no merge base with ${ref}`), input: `git merge-base ${ref} HEAD`, repair: `fetch the base first (git fetch origin ${base})` });
return result({ perPath: [], deleted: [], stampOnly: [], created: [], added: {}, removed: {} }, { root, registry, errors, mergeBase: null, untracked: [], consumed: 'all' });
return { ...result({ perPath: [], deleted: [], stampOnly: [], created: [], added: {}, removed: {} }, { root, registry, errors, mergeBase: null, untracked: [], consumed: 'all' }), base: resolved };
}
const mergeBase = mb.out.trim();
const diff = readDiff(root, [mergeBase], errors);
@@ -284,5 +291,5 @@ export function collectLive(root: string, base: string): Collected {
const drawings = drawingsAmong(diff.perPath, root, p => {
try { return fs.statSync(path.join(root, p)).size <= DRAWING_READ_CAP ? fs.readFileSync(path.join(root, p)) : null; } catch { return null; }
});
return result(diff, { root, registry, errors, mergeBase, untracked, consumed, drawings });
return { ...result(diff, { root, registry, errors, mergeBase, untracked, consumed, drawings }), base: resolved };
}
+32 -21
View File
@@ -93,7 +93,16 @@ const SVG_MARKUP = /^<(?:!--|!\[CDATA\[|\?xml\s|!DOCTYPE\s+svg(?:\s+PUBLIC\s+(["
export const SVG_REFUSED = [/&(?!(?:amp|lt|gt|quot|apos);)/, /\\/, /@import/i, /url\((?!\s*#|['"]\s*#)/i, /expression\s*\(/i, /-moz-binding|\bbehavior\s*:/i];
/** A drawing's CSS, a style element's text or a style attribute's value, holds declarations only: no @ rule and no function but url(#id). */
const SVG_STYLE_OPEN = /<(?:[A-Za-z_][\w.-]*:)?style\b/gi;
const SVG_STYLE_CLOSE = /^<\/(?:[A-Za-z_][\w.-]*:)?style\s*>/i;
const SVG_STYLE_CLOSE = /^<\/(?:[A-Za-z_][\w.-]*:)?style[ \t\n\r]*>/i;
/**
* A style element's opening tag, read as quoted name="value" pairs separated by the ASCII whitespace XML and HTML share: any
* other shape (a value holding < or >, an unquoted or valueless attribute, />, another space) refuses, so it cannot hide a closing tag.
*/
const SVG_STYLE_TAG = /^<(?:[A-Za-z_][\w.-]*:)?style(?:[ \t\n\r]+[\w.:-]+[ \t\n\r]*=[ \t\n\r]*(?:"[^"<>]*"|'[^'<>]*'))*[ \t\n\r]*>/i;
/** An element name, ended by ASCII whitespace, / or >: a non-ASCII character in the name (a prefix such as gé:) refuses rather than reading as its ASCII start. */
const SVG_TAG_NAME = /^<\/?(?:[A-Za-z_][\w.-]*:)?([A-Za-z_][\w.-]*)(?=[ \t\n\r/>]|$)/;
/** Every name= in the text, non-ASCII letters included so a prefix such as xé:href is read whole and refused. */
const SVG_ATTR = /(?<![\w.:\-\u0080-\uffff])([A-Za-z_\u0080-\uffff][\w.:\-\u0080-\uffff]*)\s*=\s*(?=(?:"([^"]*)"|'([^']*)'|([^\s"'<>=]+)))/g;
const SVG_CSS_REF = /url\(\s*(['"]?)#[\w.:-]*\1\s*\)/gi;
const SVG_CSS_REFUSED = /[@(]/;
/** An href or src value a drawing may hold: a same-document reference, or a raster image inline. */
@@ -102,10 +111,12 @@ const SVG_HREF = /^\s*(?:#|data:image\/(?:png|jpe?g|gif|webp);base64,[A-Za-z0-9+
/**
* Whether an SVG's whole text is a drawing (GST-911): every element in SVG_ELEMENTS, no event handler, every href or src a
* same-document reference or an inline raster image, CSS of declarations only, and nothing SVG_REFUSED names, read after character references are
* decoded. Every name= in the text is read, so a stray quote cannot hide a later attribute inside its value; an unquoted value
* stops at the next =, which keeps the scan linear. A style element's CSS is walked with indexOf to its closing tag, dropping
* comments and reading CDATA as CSS, so neither can hide a closing tag; any other markup inside it, or a style element that never
* closes, refuses. Anything else, an SVG that can run code included, stays code.
* decoded. Every name= in the text is read (SVG_ATTR), so a stray quote cannot hide a later attribute inside its value, and a name
* holding a non-ASCII character refuses, like an element name (SVG_TAG_NAME); an unquoted value
* stops at the next =, which keeps the scan linear. A style element's opening tag is read quote by quote (SVG_STYLE_TAG), and its CSS is walked with indexOf to its closing tag over the raw
* text, so a character reference cannot fake a closing tag or a comment, and any & in it refuses (GST-940). Comments and CDATA are
* read as CSS, so neither can hide a closing tag and an HTML-only comment end (<!--> or --!>) exposes nothing unread; any other
* markup inside it, or a style element that never closes, refuses. Anything else, an SVG that can run code included, stays code.
*/
export function svgDrawing(text: string): boolean {
const t = decodeCharRefs(text);
@@ -113,36 +124,36 @@ export function svgDrawing(text: string): boolean {
const cssDrawing = (css: string) => !SVG_CSS_REFUSED.test(css.replace(SVG_CSS_REF, ''));
for (let at = 0; ;) {
SVG_STYLE_OPEN.lastIndex = at;
const open = SVG_STYLE_OPEN.exec(t);
const open = SVG_STYLE_OPEN.exec(text);
if (!open) break;
let i = t.indexOf('>', open.index) + 1;
if (!i) return false;
const tag = SVG_STYLE_TAG.exec(text.slice(open.index, open.index + 4096));
if (!tag) return false;
let i = open.index + tag[0].length;
let css = '';
for (;;) {
const lt = t.indexOf('<', i);
const lt = text.indexOf('<', i);
if (lt === -1) return false;
css += t.slice(i, lt);
if (t.startsWith('<!--', lt)) {
i = t.indexOf('-->', lt + 4) + 3;
if (i === 2) return false;
} else if (t.startsWith('<![CDATA[', lt)) {
const end = t.indexOf(']]>', lt + 9);
css += text.slice(i, lt);
const [start, close] = text.startsWith('<!--', lt) ? [lt + 4, '-->'] : text.startsWith('<![CDATA[', lt) ? [lt + 9, ']]>'] : [-1, ''];
if (start !== -1) {
const end = text.indexOf(close, start);
if (end === -1) return false;
css += t.slice(lt + 9, end);
css += ' ' + text.slice(start, end) + ' ';
i = end + 3;
} else if (SVG_STYLE_CLOSE.test(t.slice(lt, lt + 256))) {
} else if (SVG_STYLE_CLOSE.test(text.slice(lt, lt + 256))) {
at = lt + 1;
break;
} else return false;
}
if (!cssDrawing(css)) return false;
if (css.includes('&') || !cssDrawing(css)) return false;
}
for (let at = t.indexOf('<'); at !== -1; at = t.indexOf('<', at + 1)) {
if (SVG_MARKUP.test(t.slice(at, at + 512))) continue;
const tag = /^<\/?(?:[A-Za-z_][\w.-]*:)?([A-Za-z_][\w.-]*)/.exec(t.slice(at, at + 256));
const tag = SVG_TAG_NAME.exec(t.slice(at, at + 256));
if (!tag || !SVG_ELEMENTS.has(tag[1]!.toLowerCase())) return false;
}
for (const m of t.matchAll(/(?<![\w.:-])([A-Za-z_][\w.:-]*)\s*=\s*(?=(?:"([^"]*)"|'([^']*)'|([^\s"'<>=]+)))/g)) {
for (const m of t.matchAll(SVG_ATTR)) {
if (/[^\x00-\x7f]/.test(m[1]!)) return false;
const local = m[1]!.slice(m[1]!.lastIndexOf(':') + 1).toLowerCase();
const value = m[2] ?? m[3] ?? m[4] ?? '';
if (local.startsWith('on') || ((local === 'href' || local === 'src') && !SVG_HREF.test(value)) || (local === 'style' && !cssDrawing(value))) return false;
@@ -153,7 +164,7 @@ export function svgDrawing(text: string): boolean {
/** A hash of the constants and patterns that decide a depth, not of this file's bytes: a comment edit keeps it, a threshold edit changes it. */
export const DEPTH_RULES = createHash('sha256').update(JSON.stringify({
RISK_RULES, SMALL_DIFF_LINES, RED_TEAM_LINES, LARGE_DIFF_LINES, FULL_PASS_CAP, LIGHT_FIXED, BATCH_SCOPED, SCOPED_RANGE_LINES, ROLE_EFFORT, SVG_ELEMENTS: [...SVG_ELEMENTS],
patterns: Object.entries({ GATE_PATH, AGENT_PROSE, TEST_PATH, RELEASE_METADATA, DOC_PATH: [DOC_PATH], DATA_PATH: [DATA_PATH], DATA_CONFIG, SVG_MARKUP: [SVG_MARKUP], SVG_REFUSED, SVG_HREF: [SVG_HREF], SVG_STYLE_OPEN: [SVG_STYLE_OPEN], SVG_STYLE_CLOSE: [SVG_STYLE_CLOSE], SVG_CSS_REF: [SVG_CSS_REF], SVG_CSS_REFUSED: [SVG_CSS_REFUSED] }).map(([k, v]) => [k, v.map(r => `/${r.source}/${r.flags}`)]),
patterns: Object.entries({ GATE_PATH, AGENT_PROSE, TEST_PATH, RELEASE_METADATA, DOC_PATH: [DOC_PATH], DATA_PATH: [DATA_PATH], DATA_CONFIG, SVG_MARKUP: [SVG_MARKUP], SVG_REFUSED, SVG_HREF: [SVG_HREF], SVG_STYLE_OPEN: [SVG_STYLE_OPEN], SVG_STYLE_CLOSE: [SVG_STYLE_CLOSE], SVG_STYLE_TAG: [SVG_STYLE_TAG], SVG_TAG_NAME: [SVG_TAG_NAME], SVG_ATTR: [SVG_ATTR], SVG_CSS_REF: [SVG_CSS_REF], SVG_CSS_REFUSED: [SVG_CSS_REFUSED] }).map(([k, v]) => [k, v.map(r => `/${r.source}/${r.flags}`)]),
})).digest('hex').slice(0, 16);
/**
+3
View File
@@ -272,6 +272,9 @@ export function bindReview(rec: Record<string, any>, token: string, env = proces
if (env.GSTACK_STAMP_TREE) rec.tree = env.GSTACK_STAMP_TREE;
if (env.GSTACK_STAMP_DIRTY) rec.dirty = env.GSTACK_STAMP_DIRTY === 'true';
stampGateFindings(rec.findings, env.GSTACK_STAMP_SESSION_KIND);
if (rec.skill === 'review-fix-check' && (typeof rec.completed !== 'boolean' || typeof rec.converged !== 'boolean')) {
throw new Error('review-log: a review-fix-check row must carry completed and converged as true or false (review-army item 6 fields; without them gstack-review-read marks the row UNVERIFIED and finish cannot count it)');
}
// A /ship pass whose findings were all fixed, or deferred or noted below the floor, left no open defect (review-army
// item 6), so a claimed issues_found is recorded clean with zero counts and Step 11.5 can bind it (GST-430). Only the
// logger-stamped action counts (never a caller's status), and the findings must account for every claimed issue.
+67 -14
View File
@@ -122,20 +122,38 @@ function select(i: ExcludeInput, patterns: readonly string[]): { byPattern: Map<
return { byPattern, files };
}
/** Executable image magic: ELF, Mach-O (thin, universal and 64-bit universal; 0xcafebabe is also a Java class), PE, WebAssembly and a script's #!. */
const NATIVE_MAGIC = [[0x7f, 0x45, 0x4c, 0x46], [0xfe, 0xed, 0xfa, 0xce], [0xfe, 0xed, 0xfa, 0xcf], [0xce, 0xfa, 0xed, 0xfe], [0xcf, 0xfa, 0xed, 0xfe], [0xca, 0xfe, 0xba, 0xbe], [0xca, 0xfe, 0xba, 0xbf], [0x4d, 0x5a], [0x00, 0x61, 0x73, 0x6d], [0x23, 0x21]];
/**
* The matched paths git calls binary whose bytes hold a NUL in their first 8,000 (git's own test), read from the head blob,
* the file on a live tree, or the old blob for a deletion. A branch's .gitattributes can make git call a text file binary;
* only real bytes make it binary content here.
* The matched paths git calls binary by their bytes alone: `git diff --numstat` with attributes read from the empty tree, so
* a branch's .gitattributes (`diff`, `-binary`, `text`) cannot make an executable image read as text and skip the
* magic check. null when git cannot run it (a git without --attr-source); the caller then reads every matched file's bytes.
*/
function binaryContent(i: ExcludeInput, files: ReadonlyMap<string, Matched>, claimed: ReadonlySet<string>, unread: Set<string>): Set<string> {
function bytesBinary(i: ExcludeInput, patterns: readonly string[]): string[] | null {
const empty = git(i.root, ['hash-object', '-t', 'tree', '--stdin'], '');
if (!empty.ok) return null;
const range = i.head ? [i.mergeBase, i.head] : [i.mergeBase];
const r = git(i.root, [`--attr-source=${empty.out.trim()}`, 'diff', '--numstat', '-z', '--no-renames', ...range, '--', ...patterns.map(p => `:(top,glob)${p}`)]);
return r.ok ? r.out.split('\0').filter(l => l.startsWith('-\t-\t')).map(l => l.slice(4)) : null;
}
const nativeImage = (b: Buffer) => NATIVE_MAGIC.some(m => m.every((x, k) => b[k] === x));
/**
* The claimed paths (git calls them binary under the branch's attributes or under none: bytesBinary) whose bytes hold a NUL
* in their first 8,000 (git's own test), read from the head blob, the file on a live tree, or the old blob for a deletion.
* A branch's .gitattributes can make git call a text file binary; only real bytes make it binary content here. A binary file that starts with executable image magic also goes in `native`.
*/
function binaryContent(i: ExcludeInput, files: ReadonlyMap<string, Matched>, claimed: ReadonlySet<string>, unread: Set<string>, native: Set<string>): Set<string> {
const out = new Set<string>();
const sniff = (p: string, b: Buffer) => { if (!b.includes(0)) return; out.add(p); if (nativeImage(b)) native.add(p); };
const real = (sha: string) => /^[0-9a-f]{40,64}$/.test(sha) && !/^0+$/.test(sha);
const bySha = new Map<string, string[]>();
for (const p of claimed) {
const m = files.get(p);
if (!m) continue;
if (!i.head && m.newMode !== '000000') {
try { const fd = fs.openSync(path.join(i.root, p), 'r'); const buf = Buffer.alloc(8000); const n = fs.readSync(fd, buf, 0, 8000, 0); fs.closeSync(fd); if (buf.subarray(0, n).includes(0)) out.add(p); } catch { unread.add(p); }
try { const fd = fs.openSync(path.join(i.root, p), 'r'); const buf = Buffer.alloc(8000); const n = fs.readSync(fd, buf, 0, 8000, 0); fs.closeSync(fd); sniff(p, buf.subarray(0, n)); } catch { unread.add(p); }
continue;
}
const sha = real(m.newSha) ? m.newSha : real(m.oldSha) ? m.oldSha : '';
@@ -144,8 +162,36 @@ function binaryContent(i: ExcludeInput, files: ReadonlyMap<string, Matched>, cla
for (const [sha, paths] of bySha) {
const r = spawnSync('git', ['cat-file', 'blob', sha], { cwd: i.root, timeout: 30_000, maxBuffer: 8000 });
const read = r.status === 0 || (r.error as NodeJS.ErrnoException | undefined)?.code === 'ENOBUFS';
const binary = read && (r.stdout ?? Buffer.alloc(0)).subarray(0, 8000).includes(0);
for (const p of paths) if (!read) unread.add(p); else if (binary) out.add(p);
const head = (r.stdout ?? Buffer.alloc(0)).subarray(0, 8000);
for (const p of paths) if (!read) unread.add(p); else sniff(p, head);
}
return out;
}
/**
* The added lines of matched files git calls binary only because of their attributes (the bytes hold no NUL): the
* collector's added lines are empty for them, so the secret-line rule reads `git diff --text` instead, which no
* .gitattributes, textconv or external diff driver changes. One git diff per KEEP_ARG_MAX_BYTES of literal pathspecs;
* git prints one section per path in byte order, so a chunk whose sections do not line up with its sorted paths, or a
* git failure, puts every path of that chunk in `unread`.
*/
function attributeBinaryLines(i: ExcludeInput, paths: readonly string[], unread: Set<string>): Record<string, string[]> {
const out: Record<string, string[]> = {};
const range = i.head ? [i.mergeBase, i.head] : [i.mergeBase];
const chunks: string[][] = [];
let bytes = Infinity;
for (const p of [...paths].sort((a, b) => Buffer.compare(Buffer.from(a), Buffer.from(b)))) {
const n = argBytes([p]);
if (bytes + n > KEEP_ARG_MAX_BYTES) { chunks.push([]); bytes = 0; }
chunks[chunks.length - 1]!.push(p);
bytes += n;
}
for (const chunk of chunks) {
const r = git(i.root, ['diff', '--text', '--no-textconv', '--no-ext-diff', '--no-renames', '--src-prefix=a/', '--dst-prefix=b/', '-U0', ...range, '--', ...chunk.map(literal)]);
const sections = r.ok ? r.out.split(/^(?=diff --git )/m).filter(s => s.startsWith('diff --git ')) : [];
const aligned = sections.length === chunk.length && chunk.every((p, k) => /["\\\x00-\x1f\x7f]/.test(p) || sections[k]!.startsWith(`diff --git a/${p} b/${p}\n`));
if (!aligned) { for (const p of chunk) unread.add(p); continue; }
chunk.forEach((p, k) => { let hunk = false; out[p] = sections[k]!.split('\n').filter(l => (hunk ||= l.startsWith('@@')) && l.startsWith('+')).map(l => l.slice(1)); });
}
return out;
}
@@ -159,17 +205,21 @@ function contentKind(i: ExcludeInput, p: string, binary: ReadonlySet<string>): C
return classifyPath(p, i.classRoot) === 'docs' ? 'docs' : null;
}
/** Source and script extensions: a NUL in one does not make it excludable binary content. */
const CODE_EXT = /\.(?:[cm]?[jt]sx?|py|rb|go|rs|java|kts?|scala|swift|exs?|php|cs|c|h|cc|cpp|hpp|mm?|dart|pl|pm|lua|sh|bash|zsh|fish|ps1|bat|cmd|vue|svelte|html?|wasm)$/i;
/**
* The only extensions a binary file may be excludable content under: images, fonts, media, documents, compressed archives
* and data blobs. Any other binary (source with a NUL, a native library or executable, a key store, a pickle, no
* extension) is code; so is one that starts with executable image magic whatever its extension.
*/
const BINARY_CONTENT_EXT = /\.(?:png|jpe?g|gif|webp|bmp|ico|avif|tiff?|heic|heif|woff2?|ttf|otf|eot|mp3|mp4|m4a|wav|ogg|oga|ogv|webm|mov|avi|mkv|flac|aac|opus|pdf|docx|xlsx|pptx|odt|ods|odp|gz|tgz|zip|tar|bz2|xz|zst|7z|bin|dat|parquet|arrow|feather|avro|orc|npy|npz|h5|hdf5|sqlite3?|db|safetensors)$/i;
/** Why a matched file is kept by its path, mode and lines alone; null for content the policy may exclude. A binary file readers would only see as "Binary files differ" is content unless its directory, mode or gate tags keep it. */
function pathReason(i: ExcludeInput, p: string, m: Matched, binary: ReadonlySet<string>): KeptReason | null {
/** Why a matched file is kept by its path, mode and lines alone; null for content the policy may exclude. A binary file readers would only see as "Binary files differ" is content unless its directory, mode, gate tags, extension or magic keep it. */
function pathReason(i: ExcludeInput, p: string, m: Matched, binary: ReadonlySet<string>, native: ReadonlySet<string>): KeptReason | null {
if (/^\.(gstack|github)\//.test(p)) return 'policy-dir';
if (isGatePath(p)) return 'gate-path';
if (m.oldMode === '100755' || m.newMode === '100755') return 'executable';
if ([m.oldMode, m.newMode].some(x => x === '120000' || x === '160000')) return 'link';
if (i.gateTagged?.has(p)) return 'gate-tagged';
if (binary.has(p)) return CODE_EXT.test(p) ? 'code' : null;
if (binary.has(p)) return BINARY_CONTENT_EXT.test(p) && !native.has(p) ? null : 'code';
if (isAgentProse(p)) return 'agent-prose';
if (isDataConfig(p)) return 'config';
if (contentKind(i, p, binary) === null) return 'code';
@@ -227,9 +277,12 @@ export function reviewExclude(i: ExcludeInput): Exclusion {
const sel = select(i, patterns);
if ('error' in sel) return { state: 'refused', code: 'REVIEW_DIFF_UNREADABLE', lines: [`${i.context}: git could not list the files review_exclude_paths matches (${sel.error}); fix: git fetch origin ${base}, then rerun once; nothing was planned (REVIEW_DIFF_UNREADABLE)`] };
const unread = new Set<string>();
const binary = binaryContent(i, sel.files, new Set(i.perPath.filter(e => e.binary && sel.files.has(e.path)).map(e => e.path)), unread);
const native = new Set<string>();
const claimed = i.perPath.filter(e => e.binary && sel.files.has(e.path)).map(e => e.path);
const binary = binaryContent(i, sel.files, new Set([...claimed, ...(bytesBinary(i, patterns) ?? sel.files.keys())]), unread, native);
const text = { ...i, added: { ...i.added, ...attributeBinaryLines(i, claimed.filter(p => !binary.has(p) && !unread.has(p)), unread) } };
const reasons = new Map<string, KeptReason | null>();
for (const [p, m] of sel.files) reasons.set(p, unread.has(p) ? 'consumer-scan-unavailable' : pathReason(i, p, m, binary));
for (const [p, m] of sel.files) reasons.set(p, unread.has(p) ? 'consumer-scan-unavailable' : pathReason(text, p, m, binary, native));
const candidates = new Set([...reasons].filter(([, r]) => r === null).map(([p]) => p));
const named = grepFixed(i, [...candidates].filter(p => !p.includes('\n')), candidates, [i.mergeBase]);
for (const p of candidates) {
+5 -2
View File
@@ -55,9 +55,11 @@ export const SECURITY_LINE = [
/\b(?:includes|startsWith|endsWith|indexOf)\s*\(\s*['"`]\.\.[/\\]{0,2}['"`]|[!=]==?\s*['"`]\.\.['"`]|['"`]\.\.['"`]\s*[!=]==?|\\\.\\\./,
/(?<![.\w])eval\s*\(/, /(?:^|[;&|])\s*eval\s+["'$]/, /\bnew\s+Function\s*\(/,
/\b(?:spawn|spawnSync|execFile|execFileSync)\s*\(\s*['"](?:\/bin\/)?(?:ba|z)?sh['"]/, /\bBun\.spawn(?:Sync)?\s*\(\s*\[\s*['"](?:\/bin\/)?(?:ba|z)?sh['"]/,
/\brmSync\s*\((?:[^)]{0,500}recursive\s*:\s*true|[^)]{501})/, /\brm\s+-(?=[a-z]*r)(?=[a-z]*f)[a-z]*[rf]\b/,
/\brmSync\s*\((?:.{0,500}?recursive\s*:\s*true|[^)]{501})/, /\brm\s*\((?:.{0,500}?recursive\s*:\s*true|[^)]{501})/, /\brm\s+-(?=[a-z]*r)(?=[a-z]*f)[a-z]*[rf]\b/,
/\b(?:innerHTML|outerHTML)\s*\+?=/, /\bdangerouslySetInnerHTML\b/, /\bv-html\b/, /\{@html\b/, /\binsertAdjacentHTML\s*\(/, /\bdocument\.write\s*\(/,
];
/** A recursive delete split across lines (a formatter puts `recursive: true` on its own line), read over a file's changed lines joined. */
export const SECURITY_JOINED = [/\brm(?:Sync)?\s*\((?:[\s\S]{0,500}?recursive\s*:\s*true|[^)]{501})/];
export const SVG_PATH = /\.svg$/i;
/** SVG text that can run code, read decoded and again with whitespace removed (java&#9;script: is javascript:); a file's changed lines are read joined, so a split attribute still counts. */
export const SVG_ACTIVE = [/<(?:[\w.-]+:)?(?:script|foreignObject)\b/i, /(?:^|[\s"'/])on[a-z]+\s*=/i, /(?:java|vb)script:|data:text\/html/i];
@@ -81,7 +83,7 @@ const namesMoney = (t: readonly string[], words: readonly string[]) => t.some(x
export const RISK_RULES = createHash('sha256').update(JSON.stringify({
RISK_CLASSES, MONEY_PATH_TOKENS, AUTH_PATH_TOKENS, AUTH_PATH_PREFIXES, KEY_NAME, SECURITY_PATH_TOKENS, DATA_PATH_TOKENS,
MONEY_LINE_TOKENS, MONEY_PAIRS,
patterns: Object.entries({ MONEY_LINE, SECRET_LINE, SECRET_PATH: [SECRET_PATH], SECURITY_LINE, DATA_LINE, SVG_PATH: [SVG_PATH], SVG_ACTIVE }).map(([k, v]) => [k, v.map(r => '/' + r.source + '/' + r.flags)]),
patterns: Object.entries({ MONEY_LINE, SECRET_LINE, SECRET_PATH: [SECRET_PATH], SECURITY_LINE, SECURITY_JOINED, DATA_LINE, SVG_PATH: [SVG_PATH], SVG_ACTIVE }).map(([k, v]) => [k, v.map(r => '/' + r.source + '/' + r.flags)]),
})).digest('hex').slice(0, 16);
export const tokens = (p: string) => p.replace(/([a-z0-9])([A-Z])/g, '$1 $2').toLowerCase().split(/[^a-z0-9]+/).filter(Boolean);
@@ -113,6 +115,7 @@ export function matchRisk(i: RiskInput): Map<RiskClass, string> {
if (!why.has('security') && SECURITY_LINE.some(r => r.test(line))) hit('security', 'security line: ' + p);
if (!why.has('data') && DATA_LINE.some(r => r.test(line))) hit('data', 'data line: ' + p);
}
if (!why.has('security') && [i.added?.[p], i.removed?.[p]].some(l => l && SECURITY_JOINED.some(r => r.test(l.join('\n'))))) hit('security', 'security lines: ' + p);
}
return new Map(RISK_CLASSES.filter(c => why.has(c)).map(c => [c, why.get(c)!]));
}
+89 -11
View File
@@ -42,7 +42,11 @@ import * as path from 'node:path';
import { sha256Hex, writeReceipt } from './egress-receipt';
import { runBashSync } from './script-invocation';
export interface RunProfile { workflow: string; events: string[]; jobs: string[]; artifact: string; max_age_hours?: number; os?: string }
/**
* `skippable` (GST-1190): named jobs the workflow may skip by path; a skip counts only in a pull_request run whose `gate` job
* succeeded, with the workflow, the declared `scripts` and any path the gate's steps name matching origin/<base> byte for byte.
*/
export interface RunProfile { workflow: string; events: string[]; jobs: string[]; artifact: string; max_age_hours?: number; os?: string; skippable?: { jobs: string[]; gate: string; scripts: string[] } }
export interface RemoteGateRuns { profiles: string[]; inventory: string; skipped?: string }
export interface ApiRun {
id: number; run_attempt: number; status: string; conclusion: string | null; event: string; head_sha: string;
@@ -50,7 +54,14 @@ export interface ApiRun {
repository?: { full_name: string } | null; head_repository?: { full_name: string } | null;
}
export interface ApiJob { name: string; conclusion: string | null; completed_at?: string | null }
export interface TreeRecord { run_id?: unknown; run_attempt?: unknown; event?: unknown; sha?: unknown; head_sha?: unknown; tree?: unknown; reused?: unknown; reuse?: unknown; os?: unknown; files?: unknown; tests?: unknown }
export interface TreeRecord { run_id?: unknown; run_attempt?: unknown; event?: unknown; sha?: unknown; head_sha?: unknown; tree?: unknown; reused?: unknown; reuse?: unknown; os?: unknown; files?: unknown; tests?: unknown; skipped_files?: unknown }
/** What a red run's reader needs beyond RunDeps (lib/pregate/flaky-exception.ts): every file of one artifact, and a pull request's head. */
export interface RunArtifactDeps {
/** Every file of the run's one unexpired artifact `name`, path → text; throws when it is missing, ambiguous or too large. */
artifactFiles(id: number, name: string): Record<string, string>;
pull(n: number): { head_sha: string; state: string; head_repo: string | null };
}
/** Everything verifyRun reads, injectable so every refusal has a fixture. */
export interface RunDeps {
@@ -62,18 +73,21 @@ export interface RunDeps {
}
export type RunRefusal = 'run_ref_invalid' | 'other_repository' | 'fork_run' | 'workflow_mismatch' | 'event_mismatch' | 'run_in_progress'
| 'run_not_success' | 'job_missing' | 'job_ambiguous' | 'job_not_success' | 'artifact_missing' | 'artifact_mismatch'
| 'run_not_success' | 'job_missing' | 'job_ambiguous' | 'job_skipped' | 'job_not_success' | 'artifact_missing' | 'artifact_mismatch'
| 'artifact_reused' | 'tree_mismatch' | 'newer_run_vetoes' | 'run_changed' | 'run_stale' | 'os_mismatch' | 'api_error';
export interface RunEvidence { run_id: number; run_attempt: number; head_sha: string; tree: string; profile: string; event: string; finished_at: string; files: string[]; tests: number | null }
/** `skipped`: the files whose every test skipped on the runner (tree.json `skipped_files`, GST-1190); undefined when an older tree.json does not say. */
export interface RunEvidence { run_id: number; run_attempt: number; head_sha: string; tree: string; profile: string; event: string; finished_at: string; files: string[]; tests: number | null; skipped?: string[]; skipped_jobs?: string[] }
export type RunVerdict = { ok: true; evidence: RunEvidence } | { ok: false; reason: RunRefusal; detail: string };
const SHA = /^[0-9a-f]{40}$/;
const RUN_EVENTS = ['push', 'workflow_dispatch', 'pull_request'];
const PROFILE_KEYS = ['workflow', 'events', 'jobs', 'artifact', 'max_age_hours', 'os'];
const PROFILE_KEYS = ['workflow', 'events', 'jobs', 'artifact', 'max_age_hours', 'os', 'skippable'];
const WORKFLOW_FILE = /^[A-Za-z0-9_.-]+\.ya?ml$/;
const NAME = /^[A-Za-z0-9_.\/-]{1,128}$/;
export const DEFAULT_MAX_AGE_HOURS = 24;
const MAX_ARTIFACT_BYTES = 8 * 1024 * 1024;
/** Shard logs are the whole test stream, so a red run's reader allows more than a tree.json. */
const MAX_LOG_ARTIFACT_BYTES = 64 * 1024 * 1024;
/** A numeric run id, or a run URL on this repository (`…/actions/runs/<id>[/attempts/<n>][/job/<j>]`). */
export function parseRunRef(ref: string, repo: string): number | null {
@@ -104,6 +118,14 @@ export function runProfileErrors(profiles: unknown, gateRuns: unknown): Array<{
if (typeof p.artifact !== 'string' || !NAME.test(p.artifact)) errors.push({ path: `${at}.artifact`, message: 'must be the artifact name holding tree.json' });
if ('max_age_hours' in p && (!Number.isInteger(p.max_age_hours) || (p.max_age_hours as number) < 1 || (p.max_age_hours as number) > 168)) errors.push({ path: `${at}.max_age_hours`, message: 'must be an integer from 1 to 168' });
if ('os' in p && (typeof p.os !== 'string' || !/^(Linux|macOS|Windows)$/.test(p.os))) errors.push({ path: `${at}.os`, message: 'must be Linux, macOS or Windows' });
if ('skippable' in p) {
const s = p.skippable as Record<string, unknown> | null;
const names = (v: unknown) => Array.isArray(v) && v.length > 0 && v.length <= 32 && v.every(x => typeof x === 'string' && !!x.trim() && x.length <= 128);
const repoPath = (x: unknown) => typeof x === 'string' && !!x && !x.startsWith('/') && !x.split('/').includes('..') && !/[\0\r\n]/.test(x);
if (!s || typeof s !== 'object' || Array.isArray(s) || Object.keys(s).some(k => !['jobs', 'gate', 'scripts'].includes(k)) || !names(s.jobs) || typeof s.gate !== 'string' || !s.gate.trim()
|| (s.jobs as string[]).includes(s.gate as string) || !Array.isArray(s.scripts) || !s.scripts.length || !s.scripts.every(repoPath))
errors.push({ path: `${at}.skippable`, message: 'must be { jobs: [job names], gate: <the job that decides the skip>, scripts: [every repo file the gate reads] }, the gate not among the jobs' });
}
}
}
if (gateRuns !== undefined && gateRuns !== null) {
@@ -126,13 +148,13 @@ export function runProfileErrors(profiles: unknown, gateRuns: unknown): Array<{
const fail = (reason: RunRefusal, detail: string): RunVerdict => ({ ok: false, reason, detail });
/** When a run's newest attempt started. A run vetoes the candidate when it started at or after the second the candidate was created: a red re-run of an older run counts, and re-running the candidate does not hide a failure in between. */
const started = (r: ApiRun): string => r.run_started_at ?? r.created_at ?? '';
const normalPath = (f: string): string => f.replace(/^(\.\/)+/, '');
export const normalPath = (f: string): string => f.replace(/^(\.\/)+/, '');
/**
* The verdict for run `runRef` under `profile`. `vetoHeads` are further heads
* whose newer runs of the same workflow can veto (the consumer's own head).
*/
export function verifyRun(deps: RunDeps, o: { repo: string; runRef: string; profileName: string; profile: RunProfile; vetoHeads?: string[]; now: Date }): RunVerdict {
export function verifyRun(deps: RunDeps, o: { repo: string; runRef: string; profileName: string; profile: RunProfile; vetoHeads?: string[]; now: Date; unchanged?: (p: RunProfile) => string | null }): RunVerdict {
const id = parseRunRef(o.runRef, o.repo);
if (id === null) return fail('run_ref_invalid', `${JSON.stringify(o.runRef)} is not a run id or a run URL on ${o.repo}`);
const p = o.profile;
@@ -153,10 +175,22 @@ export function verifyRun(deps: RunDeps, o: { repo: string; runRef: string; prof
const maxAge = p.max_age_hours ?? DEFAULT_MAX_AGE_HOURS;
if (!Number.isFinite(finished)) return fail('api_error', `${at} has no completion time`);
if (o.now.getTime() - finished > maxAge * 3600_000) return fail('run_stale', `${at} finished ${finishedAt}, more than ${maxAge} h before now`);
const skippedJobs: string[] = [];
for (const name of p.jobs) {
const hits = jobs.filter(j => j.name === name);
if (hits.length === 0) return fail('job_missing', `${at} has no job named ${name}`);
if (hits.length > 1) return fail('job_ambiguous', `${at} has ${hits.length} jobs named ${name}`);
if (hits[0]!.conclusion === 'skipped') {
const skip = p.skippable;
if (!skip?.jobs.includes(name)) return fail('job_skipped', `${at}'s ${name} job was skipped`);
if (run.event !== 'pull_request') return fail('job_skipped', `${at}'s ${name} job was skipped in a ${run.event} run; only a pull_request run's gate diffs against the PR base, so only its skip counts`);
const gate = jobs.filter(j => j.name === skip.gate);
if (gate.length !== 1 || gate[0]!.conclusion !== 'success') return fail('job_skipped', `${at}'s ${name} job was skipped and its gate job ${skip.gate} ${gate.length === 1 ? `concluded ${gate[0]!.conclusion}` : gate.length ? 'is ambiguous' : 'is missing'}`);
const changed = o.unchanged ? o.unchanged(p) : 'no base copy of the workflow was compared';
if (changed) return fail('job_skipped', `${at}'s ${name} job was skipped, but ${changed}, so this branch decided the skip`);
skippedJobs.push(name);
continue;
}
if (hits[0]!.conclusion !== 'success') return fail('job_not_success', `${at}'s ${name} job concluded ${hits[0]!.conclusion}`);
}
const heads = [...new Set([run.head_sha, ...(o.vetoHeads ?? []).filter(h => SHA.test(h))])];
@@ -186,11 +220,13 @@ export function verifyRun(deps: RunDeps, o: { repo: string; runRef: string; prof
if (p.os && !oses.includes(p.os)) return fail('os_mismatch', `${at} ran on ${oses.filter(Boolean).map(String).join(', ') || 'an unrecorded OS'}; ${o.profileName} requires ${p.os}`);
if (rec.files !== undefined && (!Array.isArray(rec.files) || rec.files.some(f => typeof f !== 'string'))) return fail('artifact_mismatch', `${at}'s ${p.artifact} files is not a list of paths`);
if (rec.tests !== undefined && (!Number.isInteger(rec.tests) || (rec.tests as number) < 0)) return fail('artifact_mismatch', `${at}'s ${p.artifact} tests is not a non-negative integer`);
if (rec.skipped_files !== undefined && (!Array.isArray(rec.skipped_files) || rec.skipped_files.some(f => typeof f !== 'string'))) return fail('artifact_mismatch', `${at}'s ${p.artifact} skipped_files is not a list of paths`);
const files = ((rec.files ?? []) as string[]).map(normalPath);
const tests = rec.tests === undefined ? null : (rec.tests as number);
const changed = runChanged(deps, id, run.run_attempt);
if (changed) return fail('run_changed', changed);
return { ok: true, evidence: { run_id: id, run_attempt: run.run_attempt, head_sha: run.head_sha, tree: headTree, profile: o.profileName, event: run.event, finished_at: finishedAt!, files, tests } };
const skipped = rec.skipped_files === undefined ? {} : { skipped: (rec.skipped_files as string[]).map(normalPath) };
return { ok: true, evidence: { run_id: id, run_attempt: run.run_attempt, head_sha: run.head_sha, tree: headTree, profile: o.profileName, event: run.event, finished_at: finishedAt!, files, tests, ...skipped, ...(skippedJobs.length ? { skipped_jobs: skippedJobs } : {}) } };
} catch (e: any) {
return fail('api_error', `reading run ${id} failed: ${String(e?.message ?? e).slice(0, 300)}`);
}
@@ -210,13 +246,21 @@ export function runInventory(command: string, cwd: string, env?: Record<string,
}
/** The files in `inventory` that no verified run's tree.json lists. */
export function uncoveredFiles(inventory: string[], evidence: RunEvidence[]): string[] {
const ran = new Set(evidence.flatMap(e => e.files.map(normalPath)));
/** A run's files minus the ones whose every test skipped there (GST-1190): only those ran anything. Older tree.json: all files. */
export const ranFiles = (e: RunEvidence): string[] => { const skipped = new Set(e.skipped ?? []); return e.files.map(normalPath).filter(f => !skipped.has(f)); };
/**
* The inventory files no run covers. `skipped: 'uncovered'` (a carry, which drops tests on a machine where they might run)
* also counts a file every test of which skipped in the run; `'covered'` (the remote gate, a Linux runner like CI) does not.
*/
export function uncoveredFiles(inventory: string[], evidence: RunEvidence[], skipped: 'uncovered' | 'covered' = 'covered'): string[] {
const ran = new Set(evidence.flatMap(e => (skipped === 'uncovered' ? ranFiles(e) : e.files.map(normalPath))));
return inventory.filter(f => !ran.has(normalPath(f)));
}
/** What a run check reads: the repository, the verifier's dependencies, the base branch's profiles and the clock. */
export interface RunSource { repo: string; deps: RunDeps; profiles: Record<string, RunProfile>; now: Date }
/** `unchanged` (findCarry only): null when a profile's workflow and gate inputs match the base, else which file differs. */
export interface RunSource { repo: string; deps: RunDeps; profiles: Record<string, RunProfile>; now: Date; unchanged?: (p: RunProfile) => string | null }
export type FromRunVerdict = { ok: true; evidence: RunEvidence; receipt: string } | { ok: false; reason: string; detail: string };
/**
@@ -298,3 +342,37 @@ export function ghRunDeps(repo: string, repoRoot: string): RunDeps {
},
};
}
/** The gh-backed artifact and pull request reads for `repo` (lib/pregate/flaky-exception.ts): one unexpired artifact by name, every regular file in it under a byte cap. */
export function ghArtifactDeps(repo: string): RunArtifactDeps {
const all = <T>(p: string, key: string): T[] => gh(['api', '--paginate', p, '--jq', `.${key}[]`]).split('\n').filter(Boolean).map(l => JSON.parse(l) as T);
return {
artifactFiles: (id, name) => {
const listed = all<{ name: string; size_in_bytes: number; expired: boolean }>(`repos/${repo}/actions/runs/${id}/artifacts?name=${encodeURIComponent(name)}&per_page=100`, 'artifacts').filter(a => a.name === name && !a.expired);
if (listed.length !== 1) throw new Error(`run ${id} has ${listed.length} unexpired artifacts named ${name}`);
if (!(listed[0]!.size_in_bytes <= MAX_LOG_ARTIFACT_BYTES)) throw new Error(`artifact ${name} is ${listed[0]!.size_in_bytes} bytes, over ${MAX_LOG_ARTIFACT_BYTES}`);
const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'gstack-run-evidence-'));
try {
gh(['run', 'download', String(id), '--repo', repo, '--name', name, '--dir', dir], 300_000);
const out: Record<string, string> = {};
let bytes = 0;
const walk = (rel: string) => {
for (const entry of fs.readdirSync(path.join(dir, rel), { withFileTypes: true })) {
const child = rel ? `${rel}/${entry.name}` : entry.name;
if (entry.isDirectory()) { walk(child); continue; }
if (!entry.isFile()) continue;
bytes += fs.statSync(path.join(dir, child)).size;
if (bytes > MAX_LOG_ARTIFACT_BYTES) throw new Error(`artifact ${name} unpacks to more than ${MAX_LOG_ARTIFACT_BYTES} bytes`);
out[child] = fs.readFileSync(path.join(dir, child), 'utf8');
}
};
walk('');
return out;
} finally { fs.rmSync(dir, { recursive: true, force: true }); }
},
pull: n => {
const p = JSON.parse(gh(['api', `repos/${repo}/pulls/${n}`])) as { state: string; head: { sha: string; repo: { full_name: string } | null } };
return { head_sha: p.head.sha, state: p.state, head_repo: p.head.repo?.full_name ?? null };
},
};
}
+18 -13
View File
@@ -10,6 +10,7 @@ import * as fs from 'node:fs';
import * as os from 'node:os';
import * as path from 'node:path';
import { fragmentSlug } from './changelog-fragments';
import { laneMapping, recordCiResult, type ShipLaneMapping } from './ci-carry';
import { isConfigUnavailable, loadConfigFromBase } from './pregate/config';
import { parseUnifiedDiff, versionAtRef } from './restamp';
import { branchSides, stampOnlyLine } from './stamp-only';
@@ -113,7 +114,8 @@ export function releasePaths(policy: Record<string, any> | null, cwd: string, re
* added since the merge base with origin/<base>, and not added by the base since): after a lost slot and a merge of the
* base its stale stamp is older than `ref`'s VERSION, so it must be at or newer than the fork point's version instead,
* the bound gstack-restamp's stamp_paths stage uses (GST-895; the fork version itself since GST-963). A branch line is
* named by its position, and with a kept VERSION only such a line may move (GST-979, GST-980).
* named by its position, and with a kept VERSION only such a line may move (GST-979, GST-980); on it a declared
* `stamp_placeholders` value may also become the new version inside a `stamp_patterns` template (GST-1026).
*/
export function stampedFiles(policy: Record<string, any> | null, cwd: string, ref: string, base: string): string[] {
const globs = Array.isArray(policy?.stamp_files) ? policy.stamp_files.filter((s: unknown): s is string => typeof s === 'string') : [];
@@ -123,11 +125,13 @@ export function stampedFiles(policy: Record<string, any> | null, cwd: string, re
let after = '';
try { after = extractVersion(fs.readFileSync(path.join(cwd, before.rel), 'utf8'), before.rel); } catch { return []; }
if (!after) return [];
const templates = Array.isArray(policy?.stamp_patterns) ? policy.stamp_patterns.filter((s: unknown): s is string => typeof s === 'string') : [];
const strings = (v: unknown): string[] => (Array.isArray(v) ? v.filter((s): s is string => typeof s === 'string') : []);
const templates = strings(policy?.stamp_patterns);
const placeholders = strings(policy?.stamp_placeholders);
const diff = git(cwd, ['diff', '--unified=0', '--no-color', '--no-ext-diff', '--no-textconv', '--no-renames', ref]);
if (diff.status !== 0) return [];
const side = branchSides(cwd, `origin/${base}`, 'HEAD', before.version, ref);
const stamp = (rel: string, removed: string, added: string, at: number): boolean => stampOnlyLine(removed, added, before.version, after, templates, side(rel), at);
const stamp = (rel: string, removed: string, added: string, at: number): boolean => stampOnlyLine(removed, added, before.version, after, templates, side(rel), at, placeholders);
return parseUnifiedDiff(diff.stdout).filter(f => f.status === 'M' && !f.binary && !f.modeChanged && matchesAny(globs, f.path) && f.hunks.length > 0 &&
f.hunks.every(h => h.removed.length === h.added.length && h.removed.every((line, i) => stamp(f.path, line, h.added[i]!, h.removedAt[i]!)))).map(f => f.path);
}
@@ -171,15 +175,15 @@ export function laneConcurrency(mode: string | undefined, cpus = os.availablePar
return cpus <= 4 ? 1 : Infinity;
}
export interface LaneResult { label: string; exit: number | null; testsRan: string; log: string; carried?: string; reused?: string; carryNone?: string }
export interface LaneResult { label: string; exit: number | null; testsRan: string; log: string; carried?: string; reused?: string; carryNone?: string; notOwed?: string; here?: string }
/** The base branch's `ship_lanes_from` entry for each lane whose command is a key, byte for byte (the ledger binds its sha256). */
export function mappedLanes(cwd: string, base: string, list: { label: string; command: string }[]): Map<string, { profiles: string[]; inventory: string }> {
/** The base branch's `ship_lanes_from` entry for each lane whose command is a key or a declared alias, byte for byte (the ledger binds its sha256). */
export function mappedLanes(cwd: string, base: string, list: { label: string; command: string }[]): Map<string, ShipLaneMapping> {
const loaded = loadConfigFromBase(cwd, base);
const map = new Map<string, { profiles: string[]; inventory: string }>();
const map = new Map<string, ShipLaneMapping>();
if (isConfigUnavailable(loaded) || loaded.errors.length) return map;
const from = loaded.config.ship_lanes_from;
for (const l of list) if (Object.prototype.hasOwnProperty.call(from, l.command)) map.set(l.label, from[l.command]!);
for (const l of list) { const m = laneMapping(from, l.command); if (m) map.set(l.label, m); }
return map;
}
@@ -189,9 +193,9 @@ export function mappedLanes(cwd: string, base: string, list: { label: string; co
*/
export function carryLane(cwd: string, base: string, l: { label: string; command: string }): LaneResult {
const rec = bin(cwd, 'gstack-evidence', ['record-ci', '--label', l.label, '--base', base, '--', l.command], 600_000);
const got = / tests_ran=(\d+) log=- source=ci runs=(\S+)/.exec(rec.stderr);
if (rec.status === 0 && got) return { label: l.label, exit: 0, testsRan: got[1]!, log: '-', carried: got[2]! };
return { label: l.label, exit: null, testsRan: 'unknown', log: '-', carryNone: /^CARRY_NONE: (.+)$/m.exec(rec.stderr)?.[1] ?? `record_failed: ${firstLine(rec.stderr)}` };
const got = recordCiResult(rec.stderr);
if (rec.status === 0 && got.runs) return { label: l.label, exit: 0, testsRan: got.files ? `${got.tests} (files)` : got.tests!, log: '-', carried: got.runs, ...(got.notOwed ? { notOwed: got.notOwed } : {}), ...(got.here ? { here: got.here } : {}) };
return { label: l.label, exit: null, testsRan: 'unknown', log: '-', carryNone: got.none ?? `record_failed: ${firstLine(rec.stderr)}` };
}
/**
@@ -201,7 +205,7 @@ export function carryLane(cwd: string, base: string, l: { label: string; command
*/
export async function runLanes(cwd: string, list: { label: string; command: string }[], concurrency = Infinity, opts: { base?: string; reuse?: boolean } = {}): Promise<LaneResult[]> {
const out: LaneResult[] = [];
const mapped = opts.base ? mappedLanes(cwd, opts.base, list) : new Map<string, { profiles: string[]; inventory: string }>();
const mapped = opts.base ? mappedLanes(cwd, opts.base, list) : new Map<string, ShipLaneMapping>();
let next = 0;
const worker = async () => {
while (next < list.length) {
@@ -222,7 +226,8 @@ export async function runLanes(cwd: string, list: { label: string; command: stri
/** Step 5's lane verdict: a lane passes on its runner count, never exit 0 alone; a carried lane names its CI runs. */
export function laneLine(l: LaneResult): { line: string; ok: boolean } {
if (l.carried) return { line: `LANE: ${l.label} carried run=${l.carried} tests_ran=${l.testsRan}`, ok: true };
if (l.notOwed) return { line: `LANE: ${l.label} not owed: ${l.notOwed} run=${l.carried}`, ok: true };
if (l.carried) return { line: `LANE: ${l.label} carried run=${l.carried} tests_ran=${l.testsRan}${l.here ? ` ${l.here}` : ''}`, ok: true };
const zero = l.exit === 0 && l.testsRan === '0';
const state = l.exit !== 0 ? 'FAIL' : zero ? 'ZERO-RUN' : l.testsRan === 'unknown' ? 'pass (count unavailable)' : 'pass';
return { line: `LANE: ${l.label} ${state} exit=${l.exit} tests_ran=${l.testsRan} log=${l.log}${l.reused ? ` reused=${l.reused}` : ''}`, ok: l.exit === 0 && !zero };
+35
View File
@@ -0,0 +1,35 @@
/**
* Validation of the stamp policy keys (stamp_patterns templates and stamp_placeholders), split out of ship-policy.ts
* to keep that module under the module-size ratchet.
*/
import type { PolicyError } from './ship-policy';
import { parseVersion } from './version-source';
const MAX_TEMPLATES = 16;
const MAX_TEMPLATE = 128;
const MAX_PLACEHOLDERS = 16;
const MAX_PLACEHOLDER = 32;
export const VERSION_SLOT = '{version}';
export function templateErrors(value: unknown, out: PolicyError[]): void {
if (!Array.isArray(value)) { out.push({ code: 'POLICY_INVALID', path: 'stamp_patterns', message: 'must be an array of literal templates' }); return; }
if (value.length > MAX_TEMPLATES) out.push({ code: 'POLICY_CONTAINMENT', path: 'stamp_patterns', message: `more than ${MAX_TEMPLATES} templates` });
value.forEach((t, i) => {
const at = `stamp_patterns[${i}]`;
if (typeof t !== 'string') { out.push({ code: 'POLICY_INVALID', path: at, message: 'must be a string' }); return; }
if (t.split(VERSION_SLOT).length !== 2) out.push({ code: 'POLICY_INVALID', path: at, message: `${JSON.stringify(t)} must contain exactly one ${VERSION_SLOT}` });
else if (t.length > MAX_TEMPLATE || /[\0\r\n]/.test(t)) out.push({ code: 'POLICY_CONTAINMENT', path: at, message: `must be one line of at most ${MAX_TEMPLATE} characters` });
});
}
export function placeholderErrors(value: unknown, out: PolicyError[]): void {
if (!Array.isArray(value)) { out.push({ code: 'POLICY_INVALID', path: 'stamp_placeholders', message: 'must be an array of literal placeholders' }); return; }
if (value.length > MAX_PLACEHOLDERS) out.push({ code: 'POLICY_CONTAINMENT', path: 'stamp_placeholders', message: `more than ${MAX_PLACEHOLDERS} placeholders` });
value.forEach((p, i) => {
const at = `stamp_placeholders[${i}]`;
if (typeof p !== 'string') { out.push({ code: 'POLICY_INVALID', path: at, message: 'must be a string' }); return; }
if (!p.trim() || p.length > MAX_PLACEHOLDER || /[\0\r\n]/.test(p)) out.push({ code: 'POLICY_CONTAINMENT', path: at, message: `must be one non-empty line of at most ${MAX_PLACEHOLDER} characters` });
else if (p.includes(VERSION_SLOT)) out.push({ code: 'POLICY_INVALID', path: at, message: `${JSON.stringify(p)} must not contain ${VERSION_SLOT}` });
else if (parseVersion(p)) out.push({ code: 'POLICY_INVALID', path: at, message: `${JSON.stringify(p)} is a version, not a placeholder` });
});
}
+9 -14
View File
@@ -18,6 +18,8 @@ import * as path from 'node:path';
import { defaultBaseBranch, noDefaultBaseLine } from './default-base';
import { RESULT_CODES, type ResultCodeName } from './result-codes';
import { cmpVersion, parseVersion } from './version-source';
import { placeholderErrors, templateErrors } from './ship-policy-stamps';
export { VERSION_SLOT } from './ship-policy-stamps';
export const SHIP_POLICY_PATH = '.gstack/ship-policy.json';
export const SHIP_POLICY_SCHEMA = 'https://github.com/garrytan/gstack/docs/ship-policy.md#schema-v1';
@@ -64,6 +66,8 @@ export interface ShipPolicy {
stamp_files: string[];
/** Literal stamp templates with one `{version}` slot, beside the built-in `(was X)`, `since: X`, `vX`; writer and classifier share them. */
stamp_patterns: string[];
/** Placeholders (`TBD`) a restamp may replace with the new version inside a `stamp_patterns` template, on the branch's own lines only. */
stamp_placeholders: string[];
/** The repo's own restamp, run first in the staging worktree with the target version; gstack's version writer is skipped. */
restamp_command: string | null;
/** Migration numbering: `{ dir, id }` where `id` is a regex with one capture group over the file name. */
@@ -161,6 +165,7 @@ export const DEFAULT_POLICY: ShipPolicy = {
maintainers: [],
stamp_files: [],
stamp_patterns: [],
stamp_placeholders: [],
restamp_command: null,
migrations: null,
gate_command: null,
@@ -193,6 +198,7 @@ export const POLICY_EXPLANATIONS: Record<PolicyKey, string> = {
maintainers: 'GitHub logins treated as maintainers by /review contributor mode; empty uses repository push permission.',
stamp_files: 'The repo’s stamp rule: globs of the only files that may differ by a stamp (version file, manifests, plugin trees, the PR’s own since: rows). Empty keeps gstack’s per-file rules; set, it applies before every rule.',
stamp_patterns: 'Literal templates with exactly one {version} slot (at most 16, 128 characters each), for example `"version": "{version}"` or `since: \'{version}\'`; the restamp writer and the tree receipt use them beside (was X), since: X and vX.',
stamp_placeholders: 'Literal placeholders (at most 16, one line, 32 characters each, never `{version}` or a version), for example `["TBD"]`: a line the branch added that reads a placeholder in a stamp_patterns template (`since: \'TBD\'`) may become that template at the new version, and finish, the receipt carry and the tree receipt read it as a stamp. Empty (default) accepts no placeholder.',
restamp_command: 'The repo’s own restamp, run first in gstack-restamp’s staging worktree (a clean checkout of HEAD) with the target version as its argument and in GSTACK_RESTAMP_VERSION; gstack’s version writer is skipped, output confinement still applies. Needs a clean caller tree and --allow-repo-commands.',
migrations: '{ dir, id }: the migration directory and a regex with one capture group for the number in a file name; `gstack-pregate regate` reports MIGRATION_COLLISION against the base branch and every open PR.',
gate_command: 'The repo’s full pre-merge gate command. gstack-ship-queue gate-ahead and ready print it for the caller to run (in the background or in CI); gstack never executes it. null when the repo has none.',
@@ -236,6 +242,7 @@ export const BUMP_NEVER_KEY_VERSION = '1.91.353.0';
export const BASE_RED_KEY_VERSION = '1.91.374.0';
/** `train.gates` and `train.land` are values of the older train key, so compatErrors checks them by value, as for bump: never. */
export const TRAIN_GATES_KEY_VERSION = '1.92.0.0';
export const STAMP_PLACEHOLDERS_KEY_VERSION = '1.94.16.0';
export const REVIEW_EXCLUDE_KEY_VERSION = '1.93.33.0';
export const POLICY_KEY_INTRODUCED: Record<PolicyKey, string> = {
bump: '1.91.73.0', queue_mode: '1.91.73.0', release_tool: '1.91.73.0', release_outputs: '1.91.73.0', stamp_paths: '1.91.73.0',
@@ -245,6 +252,7 @@ export const POLICY_KEY_INTRODUCED: Record<PolicyKey, string> = {
train: TRAIN_KEYS_VERSION, review_scope: REVIEW_SCOPE_KEY_VERSION, docs: DOCS_CAPS_KEY_VERSION,
changelog_fragments: CHANGELOG_FRAGMENTS_KEY_VERSION,
base_red: BASE_RED_KEY_VERSION, review_exclude_paths: REVIEW_EXCLUDE_KEY_VERSION,
stamp_placeholders: STAMP_PLACEHOLDERS_KEY_VERSION,
};
const newer = (a: string, b: string): boolean => cmpVersion(parseVersion(a)!, parseVersion(b)!) > 0;
@@ -274,9 +282,6 @@ const MAX_PATTERN = 256;
const MAX_REGEX = 512;
const MAX_COMMAND = 512;
const SHA_RE = /^[0-9a-f]{7,64}$/;
const MAX_TEMPLATES = 16;
const MAX_TEMPLATE = 128;
export const VERSION_SLOT = '{version}';
function hasDotDot(p: string): boolean {
return p.split(/[\\/]+/).some(seg => seg === '..');
@@ -343,17 +348,6 @@ function excludeErrors(value: unknown, out: PolicyError[]): void {
});
}
function templateErrors(value: unknown, out: PolicyError[]): void {
if (!Array.isArray(value)) { out.push({ code: 'POLICY_INVALID', path: 'stamp_patterns', message: 'must be an array of literal templates' }); return; }
if (value.length > MAX_TEMPLATES) out.push({ code: 'POLICY_CONTAINMENT', path: 'stamp_patterns', message: `more than ${MAX_TEMPLATES} templates` });
value.forEach((t, i) => {
const at = `stamp_patterns[${i}]`;
if (typeof t !== 'string') { out.push({ code: 'POLICY_INVALID', path: at, message: 'must be a string' }); return; }
if (t.split(VERSION_SLOT).length !== 2) out.push({ code: 'POLICY_INVALID', path: at, message: `${JSON.stringify(t)} must contain exactly one ${VERSION_SLOT}` });
else if (t.length > MAX_TEMPLATE || /[\0\r\n]/.test(t)) out.push({ code: 'POLICY_CONTAINMENT', path: at, message: `must be one line of at most ${MAX_TEMPLATE} characters` });
});
}
function migrationErrors(value: unknown, out: PolicyError[]): void {
const m = value as Record<string, unknown> | null;
if (!m || typeof m !== 'object' || Array.isArray(m)) { out.push({ code: 'POLICY_INVALID', path: 'migrations', message: 'must be { dir, id } or null' }); return; }
@@ -497,6 +491,7 @@ export function validatePolicy(raw: unknown, opts: { gstackVersion?: string } =
if ('release_outputs' in obj) listErrors('release_outputs', obj.release_outputs, errors, globError);
if ('stamp_files' in obj) listErrors('stamp_files', obj.stamp_files, errors, globError);
if ('stamp_patterns' in obj) templateErrors(obj.stamp_patterns, errors);
if ('stamp_placeholders' in obj) placeholderErrors(obj.stamp_placeholders, errors);
if ('restamp_command' in obj && obj.restamp_command !== null) {
const t = obj.restamp_command;
if (typeof t !== 'string' || !t.trim()) errors.push({ code: 'POLICY_INVALID', path: 'restamp_command', message: 'must be a non-empty string or null' });
+50 -3
View File
@@ -8,7 +8,12 @@
import * as fs from 'node:fs';
import * as os from 'node:os';
import * as path from 'node:path';
import { findCarry, laneMapping, mappingCoverage } from './ci-carry';
import { defaultBaseBranch } from './default-base';
import { isConfigUnavailable, loadConfigFromHead } from './pregate/config';
import { buildRunCheck } from './pregate/run-check';
import { classifyDiff } from './review-depth';
import { wtreeFingerprint } from './script-invocation';
import { audit } from './ship-audit';
import { bin, carryNoneLine, firstLine, flagReader, gh, git, hostedOrigin, laneConcurrency, laneLine, lanes, Report, runLanes, shipPolicy, short } from './ship-bin-common';
@@ -19,6 +24,11 @@ export const PREFLIGHT_HELP = `gstack-ship-preflight — /ship Steps 1-3 and Ste
--lanes serial|parallel how Step 5 runs its lanes; unset, one at a time on 4 or fewer cores, all at once above
--policy-from head Step 5's pregate reads the branch's .gstack/pregate.json, which may only add test lanes
(a branch that adds the lane for a file it touches); anything else is PREGATE_HEAD_POLICY_LOOSENS
gstack-ship-preflight --tests-only --base <base> --policy-from head --simulate [--allow-repo-commands] --lane <label>=<command>...
a dry run that proves a new ship_lanes_from mapping: which lanes the branch's own map would carry from a verified
CI run at this tree. Every line starts SIMULATION:; no lane runs, no evidence or ledger row is written and no
obligation clears (a real carry reads origin/<base>'s map only). An inventory-mapped lane runs the branch's
inventory command, so it needs --allow-repo-commands. Exit 0 when every mapped lane would carry, else 1
gstack-ship-preflight --audit --base <base> --out <dir> the single reads and Steps 7-8's deterministic audits
(CONTEXT, TODOS, COVERAGE_MAP, Coverage, PLAN_*, READER: coverage=needed|none plan=needed|none)
@@ -33,14 +43,16 @@ const DIST = /(^|\/)(cmd\/[^/]+\/main\.go|bin\/[^/]+|Cargo\.toml|setup\.py|packa
function integrate(cwd: string, base: string, r: Report): boolean {
const branch = git(cwd, ['branch', '--show-current']).stdout.trim();
const remote = /^ref: refs\/heads\/(\S+)\s+HEAD$/m.exec(git(cwd, ['ls-remote', '--symref', 'origin', 'HEAD'], 60_000).stdout)?.[1];
const head = remote ?? git(cwd, ['symbolic-ref', '--short', 'refs/remotes/origin/HEAD']).stdout.trim().replace(/^origin\//, '');
r.add(`BRANCH: ${branch || '(detached)'} base=${base}`);
if (!branch || branch === base || branch === head) { r.stop(`on the base branch (${branch || 'detached HEAD'}): ship from a feature branch`); return false; }
if (!branch || branch === base) { r.stop(`on the base branch (${branch || 'detached HEAD'}): ship from a feature branch`); return false; }
const status = git(cwd, ['status', '--porcelain']).stdout.split('\n').filter(Boolean);
r.add(`STATUS: modified=${status.filter(l => !l.startsWith('??')).length} untracked=${status.filter(l => l.startsWith('??')).length} (uncommitted changes ship with this run)`);
const fetch = git(cwd, ['fetch', 'origin', `+refs/heads/${base}:refs/remotes/origin/${base}`], 300_000);
if (fetch.status !== 0) { r.stop(`git fetch origin ${base} failed: ${firstLine(fetch.stderr)}; restore access first`); return false; }
const remoteHead = git(cwd, ['ls-remote', '--symref', 'origin', 'HEAD'], 60_000);
if (remoteHead.status !== 0) { r.stop(`git ls-remote origin HEAD failed: ${firstLine(remoteHead.stderr)}; restore access first`); return false; }
const head = /^ref: refs\/heads\/(\S+)\s+HEAD$/m.exec(remoteHead.stdout)?.[1] ?? defaultBaseBranch(cwd).branch;
if (branch === head) { r.stop(`on the base branch (${branch}): ship from a feature branch`); return false; }
const stat = git(cwd, ['diff', forkPoint(cwd, base), '--shortstat']).stdout.trim();
const lines = [...stat.matchAll(/(\d+) (insertion|deletion)/g)].reduce((n, m) => n + Number(m[1]), 0);
const commits = git(cwd, ['rev-list', '--count', `origin/${base}..HEAD`]).stdout.trim();
@@ -161,6 +173,40 @@ async function tests(cwd: string, base: string, list: { label: string; command:
r.next('Step 6: select eval suites for prompt-related changes with the project\'s selector, then Step 7');
}
/**
* --simulate (GST-1190): HEAD's ship_lanes_from and run_profiles, findCarry against this tree, nothing else. It never calls
* gstack-evidence or gstack-pregate, so it writes no row and clears no obligation: a PR controls both its own map and the
* CI that would satisfy it, so head policy never carries. The branch's inventory commands run only with `allow`
* (--allow-repo-commands), as gstack-ci-carry find --simulate requires.
*/
function simulate(cwd: string, base: string, list: { label: string; command: string }[], allow: boolean): { text: string; exit: number } {
const lines: string[] = [];
const say = (s: string) => lines.push(`SIMULATION: ${s}`);
const done = (exit: number) => ({ text: lines.join('\n') + '\n', exit });
const loaded = loadConfigFromHead(cwd);
say(`policy ${isConfigUnavailable(loaded) ? 'unavailable' : loaded.label}; a dry run: only the branch's inventory commands run, and only with --allow-repo-commands; no evidence or ledger row is written, nothing clears (a carry reads origin/${base}'s map only)`);
if (isConfigUnavailable(loaded) || loaded.errors.length) { say(`refused: ${isConfigUnavailable(loaded) ? loaded.detail : loaded.errors.map(e => `${e.path}: ${e.message}`).join('; ')}`); return done(1); }
const check = buildRunCheck(cwd, null, allow, base, 'head');
if ('error' in check) { say(`refused: ${check.error}`); return done(1); }
const head = git(cwd, ['rev-parse', 'HEAD']).stdout.trim();
const tree = git(cwd, ['rev-parse', 'HEAD^{tree}']).stdout.trim();
const wtree = () => { const fp = wtreeFingerprint({ cwd }); return 'tree' in fp ? fp.tree : null; };
let mapped = 0;
let carried = 0;
for (const l of list) {
const m = laneMapping(loaded.config.ship_lanes_from, l.command);
if (!m) { say(`LANE: ${l.label} not mapped by the branch's ship_lanes_from; /ship would run it`); continue; }
mapped++;
if (m.inventory && !allow) { say(`CI_CARRY: none label=${l.label} (inventory_not_allowed: the branch's inventory command is a pin-file command; pass --allow-repo-commands)`); continue; }
const v = findCarry(check, { profiles: m.profiles, head, tree, wtree, ...mappingCoverage(m, cwd) });
if (!v.ok) { say(`CI_CARRY: none label=${l.label} (${v.reason}: ${v.detail})`); continue; }
carried++;
say(`LANE: ${l.label} would carry run=${v.runs.map(e => `${e.run_id}/${e.run_attempt}`).join(',')} tests_ran=${v.tests}${v.unit ? ` unit=${v.unit}` : ''}${v.remainder ? ` and would run ${v.remainder.length} skipped file(s) here with per_file` : ''}`);
}
say(`${carried} of ${mapped} mapped lane(s) would carry once origin/${base} holds this map`);
return done(mapped > 0 && carried === mapped ? 0 : 1);
}
export async function preflight(argv: readonly string[], cwd: string): Promise<{ text: string; exit: number }> {
const { flag, has } = flagReader(argv);
const base = flag('--base');
@@ -175,6 +221,7 @@ export async function preflight(argv: readonly string[], cwd: string): Promise<{
if (!['base', 'head'].includes(policyFrom)) return { text: 'gstack-ship-preflight: --policy-from needs base|head\n', exit: 2 };
if (!base || typeof list === 'string') return { text: `gstack-ship-preflight: ${typeof list === 'string' ? list : '--base <base> is required'}\n`, exit: 2 };
if (has('--tests-only') && !list.length) return { text: 'gstack-ship-preflight: --tests-only needs at least one --lane <label>=<command>\n', exit: 2 };
if (has('--simulate')) return has('--tests-only') && policyFrom === 'head' && list.length ? simulate(cwd, base, list, has('--allow-repo-commands')) : { text: 'gstack-ship-preflight: --simulate needs --tests-only --policy-from head and at least one --lane <label>=<command>\n', exit: 2 };
const mode = flag('--lanes');
const concurrency = has('--lanes') && mode === undefined ? '--lanes needs serial or parallel' : laneConcurrency(mode);
if (typeof concurrency === 'string') return { text: `gstack-ship-preflight: ${concurrency}\n`, exit: 2 };
+7
View File
@@ -29,6 +29,7 @@ import * as fs from 'node:fs';
import * as path from 'node:path';
import type { PregateFile } from './headless-artifacts';
import { CHECKS, pregateMirrorPath, readPregateFile, verifyPregate, PREGATE_OUT_REL } from './pregate/index';
import { exceptionConsumerProblem, flakyExceptionLines } from './pregate/flaky-exception';
import { ensureTmpExcluded, restamp, versionAtRef } from './restamp';
import type { ResultCodeName } from './result-codes';
import { extractReceipt } from './ship-receipt';
@@ -232,6 +233,7 @@ export function ready(ctx: QueueContext, o: { pr: string; pregate?: string; gate
const pregateFile = path.resolve(ctx.repoRoot, o.pregate ?? PREGATE_OUT_REL);
let pregateLine = `PREGATE: missing (${path.relative(ctx.repoRoot, pregateFile)}; run gstack-pregate run --allow-repo-commands)`;
let pregateOk = false;
let accepted: PregateFile | null = null;
const treeNow = wtreeFingerprint({ cwd: ctx.repoRoot });
// The mirror stands in for a record that is missing, unreadable, recorded for another tree, or current but
// waiting on a remote receipt another checkout holds; a record whose own checks failed or stayed incomplete (the diff's doing
@@ -242,6 +244,7 @@ export function ready(ctx: QueueContext, o: { pr: string; pregate?: string; gate
if (read.file) {
const v = verifyPregate(ctx.repoRoot, read.file, treeNow);
pregateOk = v.ok;
if (v.ok) accepted = read.file;
useMirror = !v.ok && v.code !== 'PREGATE_LANE_FAILED' && v.code !== 'PREGATE_INCOMPLETE' && v.code !== 'PREGATE_ENVIRONMENT';
pregateLine = `PREGATE: ${read.file.verdict} tree=${read.file.tree.slice(0, 12)} head=${(read.file.head ?? '').slice(0, 12)} checks=${read.file.checks.length}${v.ok ? '' : ` — ${v.message} (${v.code})`}`;
} else pregateLine = `PREGATE: unreadable (${read.errors[0]?.message ?? 'invalid'})`;
@@ -253,9 +256,12 @@ export function ready(ctx: QueueContext, o: { pr: string; pregate?: string; gate
const v = read?.file && !gap ? verifyPregate(ctx.repoRoot, read.file, treeNow) : null;
if (read?.file && v?.ok) {
pregateOk = true;
accepted = read.file;
pregateLine = `PREGATE: ${read.file.verdict} tree=${read.file.tree.slice(0, 12)} head=${(read.file.head ?? '').slice(0, 12)} checks=${read.file.checks.length} (from the state root's pregate/by-tree/${treeNow.tree.slice(0, 12)}.json, recorded by another checkout)`;
} else if (gap) pregateLine += `; the state root's mirror for this tree cannot stand in: ${gap}`;
}
const foreign = accepted && exceptionConsumerProblem(accepted, head, o.pr);
if (foreign) { pregateOk = false; accepted = null; pregateLine += ` — ${foreign} (PREGATE_REMOTE_UNCLEARED)`; }
const receipt = bindShippedReceipt(ctx.repoRoot, `origin/${ctx.base}`, info.body ?? '', o.pr, acceptedReceiptHeads(ctx, policy, head, info.body ?? ''));
const attempt = currentAttempt(ctx.repoRoot, o.pr);
const hint = conflictHint(info.mergeable, ci, info.headRefName, info.isCrossRepository === true, head);
@@ -268,6 +274,7 @@ export function ready(ctx: QueueContext, o: { pr: string; pregate?: string; gate
`ci: ${ci.line}`,
...(hint ? [hint] : []),
pregateLine,
...(accepted ? flakyExceptionLines(accepted) : []),
gateLine(policy, attempt),
receipt.line,
...(receipt.depthLine ? [receipt.depthLine] : []),
+3 -2
View File
@@ -224,7 +224,8 @@ function headInputs(root: string, baseRef: string, head: string, r?: ShipReceipt
/**
* The queue's exclusion check: the digest the receipt recorded (none when absent) against the one `head` has under the current
* base policy. Only the digest counts (the policy's patterns and the excluded list), so a base merge or a moved base that leaves
* both unchanged binds; a changed selection needs the review rerun, which receipt write alone cannot stand in for.
* both unchanged binds; a changed selection needs the review rerun (write checks the same digest and does not read the
* review ledger).
*/
export function excludeMismatch(x: Exclusion, r: ShipReceipt, head: string, base: string): string | null {
const expected = x.state === 'active' ? x.record.digest : 'none';
@@ -232,7 +233,7 @@ export function excludeMismatch(x: Exclusion, r: ShipReceipt, head: string, base
if (expected === recorded) return null;
const source = x.state === 'active' ? x.record.source : r.review_exclude?.source ?? base;
const h = head.slice(0, 12);
return `field=review_exclude expected=${expected} recorded=${recorded} source=${source}: the reviewers read another selection than ${base}'s review_exclude_paths gives ${h}; the receipt is unchanged; fix: rerun /ship's review on ${h} (gstack-review-dispatch pass --base ${base.replace(/^(refs\/remotes\/)?origin\//, '')}), then gstack-ship-receipt write --pr <n> --review-exclude <d> --body <body> with the digest gstack-review-dispatch verify --ledger <ledger> prints on REVIEW_EXCLUDE_RECORD:; a receipt write alone does not clear this`;
return `field=review_exclude expected=${expected} recorded=${recorded} source=${source}: the receipt records another selection than ${base}'s review_exclude_paths gives ${h}; the receipt is unchanged; fix: rerun /ship's review on ${h} (gstack-review-dispatch pass --base ${base.replace(/^(refs\/remotes\/)?origin\//, '')}), then gstack-ship-receipt write --pr <n> --review-exclude <d> --body <body> with the digest gstack-review-dispatch verify --ledger <ledger> prints on REVIEW_EXCLUDE_RECORD: (write checks the digest against the base policy and does not read the review ledger, so only the rerun shows the reviewers read this selection)`;
}
/** The receipt's recorded gate scan when it names this head; otherwise null (unavailable). */
const boundScan = (r: ShipReceipt, head: string): number | null => {
+32 -5
View File
@@ -6,6 +6,7 @@
* computes the P7 adoption census over merged PRs (deduplicated by run and
* thread id against an owner-maintained thread cohort).
*/
import { spawnSync } from 'node:child_process';
import { ARTIFACT_SCHEMAS, checkSchema, type ShipReceipt, type ValidationError } from './headless-artifacts';
import { approvalSummary } from './approval-citations';
import { FRAGMENT_README } from './changelog-fragments';
@@ -19,7 +20,7 @@ const KEY_ORDER: Array<keyof ShipReceipt | 'docs'> = [
'schema_version', 'pr', 'head', 'version', 'base', 'gated_tree', 'predecessor', 'tree', 'gate_reuse', 'gate', 'ci', 'ci_carried', 'spend_usd',
'session_kind', 'artifacts_consumed', 'run', 'thread', 'policy', 'queue_mode', 'preregistration', 'history',
'head_tree', 'product_patch_id', 'gstack_version', 'review_depth', 'pregate', 'requires_remote', 'issues', 'approval', 'docs', 'review_loop',
'review_loop_exception', 'review_targeted', 'review_exclude', 'carry', 'base_red',
'review_loop_exception', 'review_targeted', 'review_exclude', 'carry', 'base_red', 'flaky_exception',
];
/** Keys in the fixed order so identifiers lead; undefined values are dropped. */
@@ -177,6 +178,8 @@ export interface CarryPlanInput {
rangeDiff: string;
/** The base the receipt was reviewed against when it differs from the carry's (a stack parent that landed, #3570): productPatchId.from is then the recorded diff against it. */
retargetedFrom?: string;
/** A commit inside the reviewed range whose product change the base already holds (a stack parent that squash-merged under a receipt written against the base, GST-1199): productPatchId.from is then the range from it. */
stackParent?: string;
/** A clean scoped review of product paths whose only difference is base context next to unchanged hunks (lib/ship-patch.ts, GST-384): the carry reason it adds. */
scopedReview?: string;
/** Stamp files the branch changed as product whose only change since the reviewed head is a stamp on the branch's own rows, the base untouched, and without which the two product patches match (GST-980). */
@@ -189,9 +192,9 @@ const CARRY_RERUNS = ['CI at head (gstack-ship-queue ready requires it); test an
/**
* The file's text at the reviewed head and at HEAD, with the release version each head carries (from `VERSION`, else
* `package.json`) and the base policy's `stamp_patterns` templates.
* `package.json`) and the base policy's `stamp_patterns` templates and `stamp_placeholders`.
*/
export interface CarryChange { before: string | null; after: string | null; oldVersion?: string | null; newVersion?: string | null; templates?: readonly string[]; side?: BranchSide }
export interface CarryChange { before: string | null; after: string | null; oldVersion?: string | null; newVersion?: string | null; templates?: readonly string[]; side?: BranchSide; placeholders?: readonly string[] }
/**
* A policy-named stamp path's change is a stamp when the new text reads as the
@@ -214,7 +217,7 @@ export function stampOnlyText(change: CarryChange): boolean {
const now = after.split('\n');
if (was.length !== now.length) return false;
const changed = was.map((line, i) => i).filter(i => was[i] !== now[i]);
return changed.length > 0 && changed.every(i => stampOnlyLine(was[i]!, now[i]!, oldVersion, newVersion, change.templates ?? [], change.side, i + 1));
return changed.length > 0 && changed.every(i => stampOnlyLine(was[i]!, now[i]!, oldVersion, newVersion, change.templates ?? [], change.side, i + 1, change.placeholders ?? []));
}
/**
@@ -253,7 +256,8 @@ export function planCarry(input: CarryPlanInput): CarryPlan {
const named = input.changedProduct.slice(0, CARRY_NAMED_PATHS);
return { ok: false, code: 'RECEIPT_CARRY_PRODUCT_CHANGED', message: `product diff changed between ${input.from.slice(0, 12)} and ${input.head.slice(0, 12)}: ${named.join(', ') || 'patch-id differs'}${input.changedProduct.length > named.length ? ` (+${input.changedProduct.length - named.length} more)` : ''}` };
}
const reasons: string[] = input.retargetedFrom ? [`stack parent ${input.retargetedFrom} landed`] : [];
const parent = input.retargetedFrom ?? input.stackParent?.slice(0, 12);
const reasons: string[] = parent ? [`stack parent ${parent} landed`] : [];
if (input.mergeCommits > 0) reasons.push(input.mergeCommits === 1 ? 'base merge' : `${input.mergeCommits} base merges`);
if (input.scopedReview && input.productPatchId.from !== input.productPatchId.head) reasons.push(input.scopedReview);
if (input.stampedRows?.length && input.productPatchId.from !== input.productPatchId.head) reasons.push(`a stamp on the branch's own rows (${input.stampedRows.join(', ')})`);
@@ -271,6 +275,29 @@ export function planCarry(input: CarryPlanInput): CarryPlan {
};
}
/** How many commits of the reviewed range landedStackParent tries, newest first. */
export const STACK_PARENT_WALK = 100;
/**
* The newest commit inside the reviewed range (rangeBase..reviewed, the reviewed head itself excluded) whose product
* change the current base already holds: merging it into currentBase changes no product path. That is a stack parent
* that squash-merged (or merged) into the base after the receipt was written against the base (GST-1199). A commit with no
* product change of its own (a test-first commit) is no parent, since any base trivially holds it. Null when no
* commit within STACK_PARENT_WALK qualifies or git cannot tell; a conflicted product path is never held.
*/
export function landedStackParent(repoRoot: string, s: { rangeBase: string; reviewed: string; currentBase: string }, product: readonly string[]): string | null {
const run = (args: string[]) => spawnSync('git', args, { cwd: repoRoot, encoding: 'utf8', timeout: 120_000, maxBuffer: 64 * 1024 * 1024 });
const commits = run(['rev-list', '--topo-order', `--max-count=${STACK_PARENT_WALK + 1}`, `${s.rangeBase}..${s.reviewed}`]);
if (commits.status !== 0) return null;
for (const commit of commits.stdout.split('\n').filter(c => c && c !== s.reviewed)) {
if (run(['diff', '--quiet', s.rangeBase, commit, '--', ...product]).status !== 1) continue;
const merged = run(['merge-tree', '--write-tree', '--no-messages', s.currentBase, commit]);
const tree = merged.status === 0 || merged.status === 1 ? merged.stdout.split('\n')[0]?.trim() : undefined;
if (tree && run(['diff', '--quiet', s.currentBase, tree, '--', ...product]).status === 0) return commit;
}
return null;
}
export interface TreeCarryInput {
from: string; head: string;
/** HEAD^{tree}, and every tree the receipt already covers (its head_tree plus earlier carries' trees). */
+36 -4
View File
@@ -15,6 +15,8 @@ import { mergeChangelogs, normalizeChangelog } from './changelog-check';
import { versionAtRef } from './restamp';
import { classifyPath } from './review-depth';
import { comparePath, contextPacket, gateEditedPaths, isTestCode, PACKET_SECTION_CAP, readScopedReview, touchedPaths, type PatchSides } from './ship-patch';
import { locate, readRecord } from './approval-record';
import { reviewFreshness } from './review-evidence';
import { parseReviewLoopException, type ReviewLoopException } from './ship-receipt-depth';
import { parseReceiptDocs, type ReceiptDocs } from './ship-receipt-docs';
import { resolveStateRoot } from './state-root';
@@ -185,8 +187,14 @@ function harnessRepin(cwd: string, from: string, to: string, p: string): boolean
|| (e?.since === b.version && typeof e?.nonBehavioral === 'string' && e.nonBehavioral.trim().length >= 10));
}
/** Step 16 stage 2's route. With --reviewed-base, a path whose branch patch is the reviewed one counts as base (lib/ship-patch.ts). */
type Routed = Record<'behavior' | 'docs' | 'metadata' | 'tests' | 'base' | 'context' | 'accepted', string[]>;
/**
* Step 16 stage 2's route. With --reviewed-base, a path whose branch patch is the reviewed one counts as base (lib/ship-patch.ts).
* A code or test path the newest completed clean review-fix-check row read (inside its range_base..commit range, where
* range_base holds the reviewed tree, and unchanged since) counts as fix_checked when an approved review_loop_grant names a stop
* head inside range_base..commit: the owner's granted fix-check after Step 11 is a review (GST-1130). Authored docs keep the
* stage 2.2 recheck.
*/
type Routed = Record<'behavior' | 'docs' | 'metadata' | 'tests' | 'base' | 'context' | 'accepted' | 'fix_checked', string[]>;
function route(cwd: string, base: string, reviewed: string, reviewedBase: string | undefined, release: string[], opts: { docsRechecked: boolean; contextReviewed?: string; loop?: ReviewLoopException }, r: Report): Routed | null {
const now = bin(cwd, 'gstack-wtree', [], 60_000).stdout.trim();
if (!/^[0-9a-f]{40,64}$/.test(now)) { r.stop('gstack-wtree printed no tree; the stage 2 comparison is blocked'); return null; }
@@ -206,7 +214,7 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
sides = { reviewedBase: rb, reviewed, currentBase: cb, current: now };
touched = new Set([...a, ...b]);
}
const by: Routed = { behavior: [], docs: [], metadata: [], tests: [], base: [], context: [], accepted: [] };
const by: Routed = { behavior: [], docs: [], metadata: [], tests: [], base: [], context: [], accepted: [], fix_checked: [] };
for (const p of [...changedSet].sort()) {
const verdict = sides ? (touched.has(p) ? comparePath(cwd, sides, p) : 'same') : 'changed';
by[verdict === 'same' ? 'base' : release.includes(p) || harnessRepin(cwd, reviewed, now, p) ? 'metadata' : sides && isTestCode(p) ? 'tests' : classifyPath(p, cwd) === 'docs' ? 'docs' : verdict === 'context' ? 'context' : 'behavior'].push(p);
@@ -221,6 +229,30 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
by.context = by.context.filter(p => !packetBody.over.includes(p));
by.tests = by.tests.filter(p => !packetBody.over.includes(p));
}
const fixRow = bin(cwd, 'gstack-review-read', [], 60_000).stdout.split('\n---CONFIG---')[0].split('\n')
.flatMap(l => { try { const v = JSON.parse(l); return v?.skill === 'review-fix-check' ? [v] : []; } catch { return []; } }).at(-1);
const commitOf = (c: unknown) => typeof c === 'string' && /^[0-9a-f]{7,64}$/.test(c) ? git(cwd, ['rev-parse', '--verify', '--quiet', `${c}^{commit}`]).stdout.trim() : '';
const onBranch = (c: string) => !!c && git(cwd, ['merge-base', '--is-ancestor', c, 'HEAD']).status === 0;
const fixCommit = commitOf(fixRow?.commit_full), fixFrom = commitOf(fixRow?.range_base);
const fresh = fixRow && reviewFreshness(fixRow, fixRow.wtree);
const ancestor = (a: string, d: string) => git(cwd, ['merge-base', '--is-ancestor', a, d]).status === 0;
const grants = fixRow ? (loc => 'file' in loc ? readRecord(loc).rows : [])(locate({ cwd, env: process.env })) : [];
const grant = grants.find(g => g.outcome === 'approved' && `${g.ask} ${g.scope}`.includes('review_loop_grant')
&& (`${g.ask} ${g.scope}`.match(/\b[0-9a-f]{7,40}\b/g) ?? []).some(s => { const c = commitOf(s); return !!c && !!fixFrom && !!fixCommit && ancestor(fixFrom, c) && ancestor(c, fixCommit); }));
const notCounted = !fixRow ? '' : fresh?.status !== 'CURRENT' ? fresh?.reason ?? 'not a review row'
: fixRow.tree !== fixRow.wtree || !onBranch(fixCommit) ? 'the row was logged on content no branch commit holds'
: !onBranch(fixFrom) ? 'the row names no range_base commit on this branch'
: git(cwd, ['rev-parse', `${fixFrom}^{tree}`]).stdout.trim() !== reviewed ? `range_base ${fixFrom.slice(0, 12)} does not hold the reviewed tree`
: !grant ? 'no approved review_loop_grant names a stop head inside its range_base..commit' : '';
if (notCounted) r.add(`FIX_CHECK: not counted (${notCounted})`);
const counted = !!fixRow && !notCounted;
if (counted) {
const range = touchedPaths(cwd, fixFrom, fixCommit), after = touchedPaths(cwd, fixCommit, now);
if (!range || !after) { r.stop(`git could not diff the fix-check range ${fixFrom.slice(0, 12)}..${fixCommit.slice(0, 12)} against the reviewed and current trees: stage 2 is blocked`); return null; }
by.fix_checked = [...by.behavior, ...by.tests, ...by.context].filter(p => range.has(p) && !after.has(p)).sort();
for (const k of ['behavior', 'tests', 'context'] as const) by[k] = by[k].filter(p => !by.fix_checked.includes(p));
r.add(`FIX_CHECK: reviewed ${by.fix_checked.length ? short(by.fix_checked) : 'none'} (fix-check at ${fixCommit.slice(0, 12)}, range from ${fixFrom.slice(0, 12)}, grant ${grant!.id})`);
}
const loop = opts.loop;
if (loop) {
const head = loop.fix_check.head;
@@ -235,7 +267,7 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
by.behavior = by.behavior.filter(p => !by.accepted.includes(p));
by.docs = by.docs.filter(p => !by.accepted.includes(p));
}
r.add(`ROUTE: reviewed=${reviewed.slice(0, 12)} current=${now.slice(0, 12)} behavior=${by.behavior.length} docs=${by.docs.length} metadata=${by.metadata.length} tests=${by.tests.length} base=${sides ? by.base.length : 'unknown'} context=${by.context.length}${loop ? ` accepted=${by.accepted.length}` : ''}`);
r.add(`ROUTE: reviewed=${reviewed.slice(0, 12)} current=${now.slice(0, 12)} behavior=${by.behavior.length} docs=${by.docs.length} metadata=${by.metadata.length} tests=${by.tests.length} base=${sides ? by.base.length : 'unknown'} context=${by.context.length}${counted ? ` fix_checked=${by.fix_checked.length}` : ''}${loop ? ` accepted=${by.accepted.length}` : ''}`);
if (loop) r.add(`REVIEW_LOOP_EXCEPTION: accepted ${by.accepted.length ? short(by.accepted) : 'none'} (changed since fix-check ${loop.fix_check.head.slice(0, 12)}; accepted by ${loop.accepted_by} ${loop.date} via ${loop.via})`);
if (by.behavior.length) { r.stop(`behavior, tests or build inputs changed since review (${short(by.behavior.sort())}): insert 5–11.5 → 12–14 → 16 before Step 17 (stage 2.1)`); return null; }
const scoped = [...by.context, ...by.tests].sort();
+596
View File
@@ -0,0 +1,596 @@
/**
* skill-contract — the skill contract checker (GST-1088, lane A).
*
* A targeted skill declares the outputs that must exist before it is done, in
* its SKILL.md frontmatter (schema frozen in plan-v2):
*
* contract: { version: 1, items: [ { id, kind, ... } ] }
*
* checkContract() is a pure function of a Claude Code transcript (the JSONL a
* hook payload's transcript_path names): the model never self-records. Waivers
* are read the same way, from the `gstack-skill-contract waive` Bash calls in
* that transcript, and count as misses for follow. The Stop hook
* (hosts/claude/hooks/skill-contract-stop-hook.ts) and `gstack-skill-contract`
* (check, report, waive) are its two callers; evals can reuse `check`.
* Records live under $GSTACK_STATE_ROOT/skill-contracts/<session id>.json,
* written atomically. Docs: docs/skill-routing-and-contracts.md.
*/
import * as fs from 'fs';
import * as path from 'path';
import { atomicWriteSync } from './fs-atomic';
import { readFrontmatter, scanInstalledSkills } from './skill-route';
import { resolveStateRoot } from './state-root';
export const TRANSCRIPT_MAX_BYTES = 32 * 1024 * 1024;
export const ITEM_KINDS = ['report_labels', 'check_after_last_edit', 'selected_specialists_dispatched', 'agent_dispatch'] as const;
export type ItemKind = typeof ITEM_KINDS[number];
const ITEM_ID = /^[a-z0-9][a-z0-9_.-]{0,63}$/;
const MAX_ITEMS = 20;
/** Stable exit codes of gstack-skill-contract; each has an anchor in docs/skill-routing-and-contracts.md. */
export const EXIT = { MET: 0, UNMET: 1, USAGE: 2, STATE_ERROR: 3, UNKNOWN_SHAPE: 4, NO_CONTRACT: 5, NOT_LOADED: 6 } as const;
export const STATE_ERROR_DOC = 'docs/skill-routing-and-contracts.md#state-error';
export interface ContractItem {
id: string;
kind: ItemKind;
heading?: string;
labels?: string[];
match?: string;
marker?: string;
empty_marker?: string;
fix?: string;
summary?: string;
}
export interface Contract { version: 1; items: ContractItem[] }
export class StateError extends Error {
constructor(readonly file: string, readonly why: string) { super(`${file}: ${why}`); }
}
export class UnknownShape extends Error {}
export function stateErrorLine(file: string, why: string): string {
return `gstack-skill-contract: SKILL_CONTRACT_STATE_ERROR ${file}: ${why}; stop allowed, miss not recorded (see ${STATE_ERROR_DOC})`;
}
const isObject = (v: unknown): v is Record<string, any> => typeof v === 'object' && v !== null && !Array.isArray(v);
const errText = (e: unknown) => (e as NodeJS.ErrnoException)?.code ?? (e instanceof Error ? e.message : String(e));
const escapeRe = (s: string) => s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
/** The `contract:` key and its indented lines, or null when the frontmatter has none. */
export function contractBlock(frontmatter: string): string | null {
const lines = frontmatter.split('\n');
const start = lines.findIndex(l => /^contract:/.test(l));
if (start < 0) return null;
let end = start + 1;
while (end < lines.length && (!lines[end].trim() || /^\s/.test(lines[end]))) end++;
return lines.slice(start, end).join('\n');
}
export function parseContract(frontmatter: string): { contract: Contract } | { error: string } | null {
const block = contractBlock(frontmatter);
if (block === null) return null;
let doc: unknown;
try {
doc = Bun.YAML.parse(block);
} catch (e) {
return { error: `contract block is not valid YAML (${errText(e)})` };
}
return validateContract(isObject(doc) ? doc.contract : undefined);
}
const STRING_FIELDS = ['heading', 'match', 'marker', 'empty_marker', 'fix', 'summary'] as const;
const REGEX_FIELDS = ['match'] as const;
/**
* Every default matcher string the checker uses, in one table: edit it here. A contract item's own marker or
* empty_marker replaces its row for that item. Markers are literal text in which a whole-word N is a number.
*/
export const DEFAULT_MATCHERS = {
/** selected_specialists_dispatched: the selection line. The names are the rest of the line; N = 0 selects none. */
specialistSelection: 'Dispatching N specialists:',
/** selected_specialists_dispatched: a line that means no specialist was selected. */
specialistsSkipped: 'specialists skipped',
/** Any contract: a line of the turn's final message, starting with this, means the skill ended the run early (met). A final STATUS: BLOCKED does too. */
earlyExit: 'Nothing to review',
} as const;
/**
* A marker as a pattern: literal text, case-insensitive, any run of spaces matching any whitespace, and a whole-word
* N matching a number (the first one captured as count). The names are the rest of that line.
*/
export function markerPattern(marker: string): RegExp {
const literal = (s: string) => s.split(/\s+/).map(escapeRe).join('\\s+');
const body = marker.split(/\bN\b/).map(literal).map((s, i) => (i === 0 ? s : `${i === 1 ? '(?<count>\\d+)' : '\\d+'}${s}`)).join('');
return new RegExp(`${body}\\s*(?<names>[^\\n]*)`, 'i');
}
export function validateContract(raw: unknown): { contract: Contract } | { error: string } {
if (!isObject(raw)) return { error: 'contract must be a mapping' };
if (raw.version !== 1) return { error: `contract version ${JSON.stringify(raw.version)} is not supported (expected 1)` };
if (!Array.isArray(raw.items) || !raw.items.length || raw.items.length > MAX_ITEMS) {
return { error: `contract items must be a list of 1 to ${MAX_ITEMS} items` };
}
const items: ContractItem[] = [];
for (const [n, item] of raw.items.entries()) {
if (!isObject(item)) return { error: `items[${n}] must be a mapping` };
const id = item.id;
if (typeof id !== 'string' || !ITEM_ID.test(id)) return { error: `items[${n}].id must match ${ITEM_ID}` };
if (items.some(i => i.id === id)) return { error: `duplicate item id ${id}` };
if (!(ITEM_KINDS as readonly string[]).includes(item.kind)) {
return { error: `${id}: unknown kind ${JSON.stringify(item.kind)} (expected one of ${ITEM_KINDS.join(', ')})` };
}
const out: ContractItem = { id, kind: item.kind };
for (const field of STRING_FIELDS) {
if (item[field] === undefined) continue;
if (typeof item[field] !== 'string' || !item[field].trim()) return { error: `${id}: ${field} must be a non-empty string` };
out[field] = item[field].trim();
}
for (const field of REGEX_FIELDS) {
if (out[field] === undefined) continue;
try { new RegExp(out[field] as string, 'i'); } catch { return { error: `${id}: ${field} is not a valid regular expression` }; }
}
if (out.kind === 'report_labels') {
if (!out.heading) return { error: `${id}: report_labels needs a heading` };
if (!Array.isArray(item.labels) || !item.labels.length || item.labels.some((l: unknown) => typeof l !== 'string' || !l.trim())) {
return { error: `${id}: report_labels needs labels, a list of non-empty strings` };
}
out.labels = item.labels.map((l: string) => l.trim());
}
if (out.kind === 'agent_dispatch' && !out.match) return { error: `${id}: agent_dispatch needs match` };
items.push(out);
}
return { contract: { version: 1, items } };
}
/** One plain phrase per item, for the routing hint's "Done means" variant. */
export function itemSummary(item: ContractItem): string {
if (item.summary) return item.summary;
switch (item.kind) {
case 'report_labels': return `a ${item.heading} with ${(item.labels ?? []).join(', ')}`;
case 'check_after_last_edit': return 'a test or repro run after your last edit';
case 'selected_specialists_dispatched': return 'one Agent dispatch per selected specialist';
case 'agent_dispatch': return `an Agent dispatch matching /${item.match}/`;
}
}
export type TranscriptEvent =
| { kind: 'prompt' | 'meta' | 'text'; line: number; text: string }
| { kind: 'tool_use'; line: number; id: string; name: string; input: Record<string, any> }
| { kind: 'tool_result'; line: number; id: string; text: string; isError: boolean };
export interface Transcript { events: TranscriptEvent[]; sessionId: string | null }
const META_PREFIX = /^\s*<(?:local-command-stdout|local-command-stderr|system-reminder|task-notification|user-prompt-submit-hook|bash-input|bash-stdout|bash-stderr)\b/;
function blockText(content: unknown): string {
if (typeof content === 'string') return content;
if (!Array.isArray(content)) return '';
return content.filter(b => isObject(b) && typeof b.text === 'string').map(b => b.text).join('\n');
}
/**
* Main-thread events of a Claude Code transcript JSONL. Sidechain (subagent) records are skipped; unparseable
* lines (a partial last line) are skipped. Throws UnknownShape when the file is not a transcript this parser knows.
*/
export function parseTranscript(text: string): Transcript {
const events: TranscriptEvent[] = [];
let sessionId: string | null = null;
let lines = 0;
let objects = 0;
let messages = 0;
for (const [idx, raw] of text.split('\n').entries()) {
if (!raw.trim()) continue;
lines++;
let rec: unknown;
try { rec = JSON.parse(raw); } catch { continue; }
if (!isObject(rec)) continue;
objects++;
if (rec.type !== 'user' && rec.type !== 'assistant') continue;
const line = idx + 1;
if (!isObject(rec.message)) throw new UnknownShape(`line ${line}: ${rec.type} record without a message object`);
const content = rec.message.content;
if (typeof content !== 'string' && !Array.isArray(content)) throw new UnknownShape(`line ${line}: message content is neither text nor a block list`);
messages++;
if (rec.isSidechain === true) continue;
if (typeof rec.sessionId === 'string') sessionId = rec.sessionId;
const blocks = typeof content === 'string' ? [{ type: 'text', text: content }] : content;
for (const block of blocks) {
if (!isObject(block)) continue;
if (rec.type === 'assistant') {
if (block.type === 'text' && typeof block.text === 'string') events.push({ kind: 'text', line, text: block.text });
if (block.type === 'tool_use' && typeof block.name === 'string') {
events.push({ kind: 'tool_use', line, id: String(block.id ?? ''), name: block.name, input: isObject(block.input) ? block.input : {} });
}
} else if (block.type === 'tool_result') {
events.push({ kind: 'tool_result', line, id: String(block.tool_use_id ?? ''), text: blockText(block.content), isError: block.is_error === true });
} else if (block.type === 'text' && typeof block.text === 'string') {
events.push({ kind: rec.isMeta === true || META_PREFIX.test(block.text) ? 'meta' : 'prompt', line, text: block.text });
}
}
}
if (lines > 0 && objects === 0) throw new UnknownShape('no line is a JSON object');
if (objects > 0 && messages === 0) throw new UnknownShape('no user or assistant message records');
return { events, sessionId };
}
/**
* The transcript with the Stop payload's last_assistant_message as its final text, when the file does not end on it
* yet: Claude Code can fire Stop before it writes the turn's last message.
*/
export function withFinalMessage(t: Transcript, message: unknown): Transcript {
if (typeof message !== 'string' || !message.trim()) return t;
const lastPrompt = t.events.map(e => e.kind).lastIndexOf('prompt');
const lastText = t.events.slice(lastPrompt + 1).filter(e => e.kind === 'text').pop() as { text: string } | undefined;
if (lastText?.text.trim() === message.trim()) return t;
const line = (t.events[t.events.length - 1]?.line ?? 0) + 1;
return { ...t, events: [...t.events, { kind: 'text', line, text: message }] };
}
export function readTranscriptFile(file: string): string {
try {
const size = fs.statSync(file).size;
if (size > TRANSCRIPT_MAX_BYTES) throw new StateError(file, `transcript is ${size} bytes, over the ${TRANSCRIPT_MAX_BYTES} byte read limit`);
return fs.readFileSync(file, 'utf8');
} catch (e) {
if (e instanceof StateError) throw e;
throw new StateError(file, `cannot read transcript (${errText(e)})`);
}
}
export const canonicalSkill = (name: string) => name.trim().toLowerCase().replace(/^\//, '').replace(/^gstack:/, '');
export function sameSkill(a: string, b: string): boolean {
const x = canonicalSkill(a);
const y = canonicalSkill(b);
return x === y || x === `gstack-${y}` || `gstack-${x}` === y;
}
export interface SkillLoad { skill: string; at: number; line: number; via: 'skill-tool' | 'slash-command' | 'read' }
/** Every skill load in order: a Skill tool call, a typed /command, or a Read of a SKILL.md under a skills/ directory (an installed skill, not a checkout's source). */
export function skillLoads(t: Transcript): SkillLoad[] {
const loads: SkillLoad[] = [];
t.events.forEach((e, at) => {
if (e.kind === 'tool_use' && e.name === 'Skill') {
const name = e.input.skill ?? e.input.command;
if (typeof name === 'string' && name.trim()) loads.push({ skill: canonicalSkill(name.split(/\s/)[0]), at, line: e.line, via: 'skill-tool' });
} else if (e.kind === 'tool_use' && e.name === 'Read' && typeof e.input.file_path === 'string') {
const m = /(?:^|[\\/])skills[\\/](?:.*[\\/])?([^\\/]+)[\\/]SKILL\.md$/.exec(e.input.file_path);
if (m) loads.push({ skill: canonicalSkill(m[1]), at, line: e.line, via: 'read' });
} else if (e.kind === 'prompt') {
const m = /<command-name>\s*\/?([^<\s]+)\s*<\/command-name>/.exec(e.text);
if (m) loads.push({ skill: canonicalSkill(m[1]), at, line: e.line, via: 'slash-command' });
}
});
return loads;
}
export const STATUS_RE = /^[ \t>*_-]*(?:\*\*|__)?(?:STATUS|Status)(?:\*\*|__)?[ \t]*(?:\*\*|__)?:[ \t]*(?:\*\*|__)?[ \t]*(DONE_WITH_CONCERNS|DONE|BLOCKED|NEEDS_CONTEXT)\b/m;
const EDIT_TOOLS = /^(?:Edit|Write|MultiEdit|NotebookEdit)$/;
const AGENT_TOOLS = /^(?:Agent|Task)$/;
const CHECK_RUNNER = /^(?:pytest|jest|vitest|mocha|rspec|phpunit|ctest|tox|nox|tsc|make|cargo|go|bun|bunx|npm|npx|pnpm|yarn|node|deno|python3?|ruby|rake|mvn|gradlew?|dotnet|swift|xcodebuild|curl|mix|bazel|just|dart|flutter|rails)$/i;
const CHECK_SCRIPT = /(?:^|[-_.])(?:tests?|spec|repro|check|verify|lint)(?:[-_.]|$)/i;
const COMMAND_WRAPPER = /^(?:bundle|uv|poetry|pipenv|docker|podman|xvfb-run|bash|sh|zsh)$/;
const COMMAND_PREFIX = /^(?:env|sudo|time|nice|timeout|command|exec|\w+=\S*|-\S*|\d+[smh]?)$/;
const isCheckProgram = (word: string) => { const name = word.split(/[\\/]/).pop() ?? ''; return CHECK_RUNNER.test(name) || (name !== 'test' && CHECK_SCRIPT.test(name)); };
const SAFE_SPECIALIST = /^[A-Za-z0-9][A-Za-z0-9 _.-]{0,39}$/;
/**
* A test or repro run: some segment of the command (split on && || ; | and newlines) starts with a test runner or a
* test, check or repro script, directly or behind a wrapper (bundle exec, uv run, docker compose run, xvfb-run, bash).
*/
export function isCheckCommand(cmd: string): boolean {
return cmd.split(/&&|\|\||[;|\n]/).some(segment => {
const words = segment.trim().replace(/^[({]\s*/, '').split(/\s+/);
while (words.length > 1 && COMMAND_PREFIX.test(words[0])) words.shift();
if (isCheckProgram(words[0])) return true;
return COMMAND_WRAPPER.test(words[0].split(/[\\/]/).pop() ?? '') && words.slice(1, 6).some(isCheckProgram);
});
}
const WAIVE_RE = /gstack-skill-contract["']?\s+waive\s+["']?([A-Za-z0-9][A-Za-z0-9_.-]*)["']?/;
type ToolUse = Extract<TranscriptEvent, { kind: 'tool_use' }>;
type ToolResult = Extract<TranscriptEvent, { kind: 'tool_result' }>;
const isToolUse = (e: TranscriptEvent): e is ToolUse => e.kind === 'tool_use';
const bashCommand = (e: TranscriptEvent) => (isToolUse(e) && e.name === 'Bash' ? String(e.input.command ?? '') : null);
const agentText = (e: ToolUse) => [e.input.description, e.input.prompt, e.input.subagent_type, e.input.name].filter(v => typeof v === 'string').join('\n');
const cut = (s: string, n = 80) => (s.length > n ? `${s.slice(0, n - 3)}...` : s).replace(/\s+/g, ' ');
function labelValue(text: string, label: string): string | null {
const re = new RegExp(`^[ \\t>*_|-]*(?:\\*\\*|__)?${escapeRe(label)}[ \\t]*(?:\\*\\*|__)?[ \\t]*:[ \\t]*(?:\\*\\*|__)?(.*)$`, 'im');
const m = re.exec(text);
return m ? m[1].replace(/(?:\*\*|__)\s*$/, '').replace(/^`+|`+$/g, '').trim() : null;
}
export function isPlaceholder(value: string): boolean {
const v = value.trim();
return !v || /^\[[^\]]*\]$/.test(v) || /^<[^>]*>$/.test(v) || /^[A-Z_]+(?:\s*\|\s*[A-Z_]+)+$/.test(v) ||
/^(?:tbd|tba|todo|xxx|fixme|placeholder|\.{2,}|…|-+|—|\?+)$/i.test(v);
}
function parseNames(raw: string): string[] {
const end = raw.search(/\.\s|\.$|\s(?:Skipped|Gated)\s*:/i);
return (end < 0 ? raw : raw.slice(0, end)).replace(/\([^)]*\)/g, ' ').split(/,|;|\band\b/)
.map(n => n.replace(/[[\]()"'`*]/g, '').trim()).filter(n => n && !/^none$/i.test(n));
}
interface ItemCheck { met: boolean; detail: string; fix: string }
function checkItem(item: ContractItem, run: TranscriptEvent[]): ItemCheck {
const texts = run.filter(e => e.kind === 'text').map(e => (e as { text: string }).text);
const agents = run.filter((e): e is ToolUse => isToolUse(e) && AGENT_TOOLS.test(e.name));
switch (item.kind) {
case 'report_labels': {
const heading = item.heading as string;
const labels = item.labels ?? [];
const headingRe = new RegExp(escapeRe(heading), 'i');
const source = [...texts].reverse().find(t => headingRe.test(t));
if (!source) return { met: false, detail: `no ${heading} found`, fix: item.fix ?? `print the ${heading} with real values for ${labels.join(', ')}` };
const after = source.slice(source.search(headingRe));
const bad = labels.filter(label => { const v = labelValue(after, label); return v === null || isPlaceholder(v); });
return bad.length
? { met: false, detail: `${heading} lacks a real value for ${bad.join(', ')}`, fix: item.fix ?? `fill ${bad.join(', ')} in the ${heading} with real values` }
: { met: true, detail: `${heading} has all ${labels.length} labels`, fix: '' };
}
case 'check_after_last_edit': {
let lastEdit = -1;
run.forEach((e, i) => { if (isToolUse(e) && EDIT_TOOLS.test(e.name)) lastEdit = i; });
if (lastEdit < 0) return { met: true, detail: 'no edits since the skill loaded', fix: '' };
const re = item.match ? new RegExp(item.match, 'i') : null;
const check = run.slice(lastEdit + 1).map(bashCommand).find(cmd => {
if (cmd === null || /gstack-skill-contract/.test(cmd)) return false;
const bare = cmd.replace(/"(?:[^"\\]|\\.)*"|'[^']*'/g, '""');
return re ? re.test(bare) : isCheckCommand(bare);
});
return check
? { met: true, detail: `ran "${cut(check)}" after the last edit`, fix: '' }
: { met: false, detail: `no test or repro command after the last ${(run[lastEdit] as ToolUse).name}`, fix: item.fix ?? 'run the test or repro again after your last edit and show its result' };
}
case 'selected_specialists_dispatched': {
const marker = markerPattern(item.marker ?? DEFAULT_MATCHERS.specialistSelection);
const empty = markerPattern(item.empty_marker ?? DEFAULT_MATCHERS.specialistsSkipped);
const bashIds = new Set(run.filter(e => bashCommand(e) !== null).map(e => (e as ToolUse).id));
const sources = run.filter(e => e.kind === 'text' || (e.kind === 'tool_result' && bashIds.has(e.id))).map(e => (e as { text: string }).text).reverse();
let names: string[] | null = null;
for (const src of sources) {
const m = marker.exec(src);
const parsed = m ? (m.groups?.count === '0' ? [] : parseNames(m.groups?.names ?? '')) : null;
if (parsed && parsed.every(n => SAFE_SPECIALIST.test(n))) { names = parsed; break; }
if (empty.test(src)) { names = []; break; }
}
if (names === null) return { met: false, detail: 'no specialist selection line found', fix: item.fix ?? 'print the specialist selection line, then dispatch one Agent per selected specialist' };
if (!names.length) return { met: true, detail: 'no specialists selected', fix: '' };
const missing = names.filter(n => {
const forms = [n, n.replace(/\s+/g, '-'), n.replace(/\s+/g, '_')].map(f => f.toLowerCase());
return !agents.some(a => forms.some(f => agentText(a).toLowerCase().includes(f)));
});
return missing.length
? { met: false, detail: `selected ${names.join(', ')}; no Agent dispatch for ${missing.join(', ')}`, fix: item.fix ?? `dispatch an Agent for ${missing.join(', ')}` }
: { met: true, detail: `dispatched ${names.join(', ')}`, fix: '' };
}
case 'agent_dispatch': {
if (item.marker) {
const marker = markerPattern(item.marker);
let markedAt = -1;
run.forEach((e, i) => { if (e.kind === 'text' && marker.test(e.text)) markedAt = i; });
if (markedAt >= 0 && run.slice(markedAt + 1).some(e => isToolUse(e) && AGENT_TOOLS.test(e.name))) {
return { met: true, detail: `printed "${item.marker}", then an Agent dispatch`, fix: '' };
}
}
const re = new RegExp(item.match as string, 'i');
return agents.some(a => re.test(agentText(a)))
? { met: true, detail: `an Agent dispatch matches /${item.match}/`, fix: '' }
: { met: false, detail: `no Agent dispatch matches /${item.match}/`, fix: item.fix ?? `dispatch the Agent pass matching /${item.match}/` };
}
}
}
/** Waivers in a run: item id -> reason, from waive Bash calls whose result (if any) is not an error. */
function waivers(run: TranscriptEvent[], skill: string, results: Map<string, ToolResult>): Map<string, string> {
const out = new Map<string, string>();
for (const e of run) {
const cmd = bashCommand(e);
const m = cmd === null ? null : WAIVE_RE.exec(cmd);
if (!m || !cmd) continue;
const reason = /--reason[= ]\s*(?:"([^"]*)"|'([^']*)'|(\S+))/.exec(cmd);
const text = reason ? (reason[1] ?? reason[2] ?? reason[3] ?? '').trim() : '';
const skillFlag = /--skill[= ]\s*["']?([^\s"']+)/.exec(cmd);
if (!text || (skillFlag && !sameSkill(skillFlag[1], skill)) || results.get((e as ToolUse).id)?.isError) continue;
out.set(m[1], text);
}
return out;
}
export interface ItemResult { id: string; kind: ItemKind; status: 'met' | 'missing' | 'waived'; detail: string; fix: string; reason?: string }
export interface CheckResult {
skill: string;
version: 1;
loaded: boolean;
load: SkillLoad | null;
items: ItemResult[];
missing: string[];
waived: string[];
/** every item met or waived: nothing to block on */
met: boolean;
/** every item met with no waiver: what follow-rate metrics count */
follow: boolean;
statusLine: string | null;
pendingQuestion: boolean;
pendingAgents: string[];
}
/**
* The contract against one run of the skill: from its load (the last one, or the one at loadLine) to its next load.
* Pure: same transcript, same result.
*/
export function checkContract(t: Transcript, skill: string, contract: Contract, opts: { loadLine?: number } = {}): CheckResult {
const events = t.events;
const loads = skillLoads(t).filter(l => sameSkill(l.skill, skill));
const load = (opts.loadLine === undefined ? loads[loads.length - 1] : loads.find(l => l.line === opts.loadLine)) ?? null;
const next = load ? loads.find(l => l.at > load.at) : undefined;
const run = load ? events.slice(load.at + 1, next?.at) : [];
const results = new Map<string, ToolResult>();
for (const e of run) if (e.kind === 'tool_result') results.set(e.id, e);
const waived = load ? waivers(run, skill, results) : new Map<string, string>();
const lastPrompt = events.map(e => e.kind).lastIndexOf('prompt');
const turn = events.slice(lastPrompt + 1).filter(e => e.kind === 'text' || e.kind === 'tool_use');
const final = turn[turn.length - 1];
const finalText = [...turn].reverse().find(e => e.kind === 'text') as { text: string } | undefined;
const statusLine = finalText ? STATUS_RE.exec(finalText.text)?.[1] ?? null : null;
const exitMarker = new RegExp(`^[ \\t>*_-]*${markerPattern(DEFAULT_MATCHERS.earlyExit).source}`, 'im');
const endedEarly = !!finalText && run.includes(finalText as TranscriptEvent) && exitMarker.test(finalText.text);
const earlyExit = !load ? null : statusLine === 'BLOCKED' ? 'STATUS: BLOCKED' : endedEarly ? DEFAULT_MATCHERS.earlyExit : null;
const items: ItemResult[] = load ? contract.items.map(item => {
if (earlyExit) return { id: item.id, kind: item.kind, status: 'met', detail: `the skill ended the run early (${earlyExit})`, fix: '' };
const r = checkItem(item, run);
if (r.met) return { id: item.id, kind: item.kind, status: 'met', detail: r.detail, fix: '' };
const reason = waived.get(item.id);
return reason ? { id: item.id, kind: item.kind, status: 'waived', detail: r.detail, fix: r.fix, reason } : { id: item.id, kind: item.kind, status: 'missing', detail: r.detail, fix: r.fix };
}) : [];
const lastParagraph = (finalText?.text ?? '').trim().split(/\n\s*\n/).pop() ?? '';
const pendingQuestion = statusLine === 'NEEDS_CONTEXT' || (final !== undefined && isToolUse(final) && /AskUserQuestion$/.test(final.name)) ||
(final !== undefined && final.kind === 'text' && /\?[\s*_)"']*$/.test(lastParagraph));
const pendingAgents = run.flatMap((e, i) => {
if (!isToolUse(e) || !AGENT_TOOLS.test(e.name)) return [];
const result = results.get(e.id);
if (!result) return [e.id];
const resultAt = run.indexOf(result);
const ids = [e.id, ...[...result.text.matchAll(/\b(?:agent|task)[_ -]?id["']?\s*[:=]\s*["']?([\w-]+)/gi)].map(m => m[1])].filter(Boolean);
const named = ids.map(id => new RegExp(`(?:^|[^\\w-])${escapeRe(id)}(?:[^\\w-]|$)`));
const notified = run.slice(resultAt + 1).some(n => n.kind === 'meta' && named.some(re => re.test(n.text)));
return e.input.run_in_background === true && i < resultAt && !notified ? [e.id] : [];
});
const missing = items.filter(i => i.status === 'missing').map(i => i.id);
const waivedIds = items.filter(i => i.status === 'waived').map(i => i.id);
return {
skill, version: 1, loaded: !!load, load, items, missing, waived: waivedIds,
met: !!load && missing.length === 0, follow: !!load && missing.length === 0 && waivedIds.length === 0,
statusLine, pendingQuestion, pendingAgents,
};
}
/** The one agent-facing block text (gate rev 2, E1). It never names the off switch: that is for humans, in the docs and setup. */
export function blockMessage(result: CheckResult): string {
const fixes = result.items.filter(i => i.status === 'missing').map(i => `${i.id}: ${i.fix}`).join('; ');
return `SKILL_CONTRACT_UNMET ${result.skill} (contract v1): missing ${result.missing.join(', ')}. Fix: ${fixes}. ` +
'If an item truly does not apply, run gstack-skill-contract waive <item> --reason "<why>" (this session only; counted as a miss for follow). This blocks once.';
}
export type ContractLookup =
| { skill: string; file: string; contract: Contract }
| { skill: string; file: string | null; error: string; reason: 'not-found' | 'no-contract' | 'invalid' };
/** The contract of an installed skill (by frontmatter name or directory), or of an explicit SKILL.md. */
export function lookupContract(skill: string, skillsDir: string, skillFile?: string): ContractLookup {
let file: string | null = skillFile ?? null;
let frontmatter: string | null = null;
if (file) {
try { frontmatter = readFrontmatter(file); } catch (e) { return { skill, file, error: `cannot read ${file} (${errText(e)})`, reason: 'not-found' }; }
} else {
const found = scanInstalledSkills(skillsDir).skills.find(s => sameSkill(s.name, skill) || sameSkill(path.basename(path.dirname(s.file)), skill));
if (found) { file = found.file; frontmatter = found.frontmatter; }
}
if (!file) return { skill, file, error: `no installed skill named ${skill} under ${skillsDir}`, reason: 'not-found' };
const parsed = frontmatter === null ? null : parseContract(frontmatter);
if (parsed === null) return { skill, file, error: `${file} declares no contract`, reason: 'no-contract' };
return 'error' in parsed ? { skill, file, error: `${file}: ${parsed.error}`, reason: 'invalid' } : { skill, file, contract: parsed.contract };
}
export type Outcome = 'met' | 'met_waived' | 'blocked' | 'met_after_block' | 'miss';
export interface RunRecord { skill: string; load_line: number; outcome: Outcome; missing: string[]; waived: string[]; blocked_at?: string; updated_at: string }
export interface SessionRecord { version: 1; session_id: string; transcript_path: string; runs: Record<string, RunRecord> }
export function recordsDir(env: NodeJS.ProcessEnv = process.env): string {
return path.join(resolveStateRoot(env), 'skill-contracts');
}
export function recordPath(sessionId: string, env: NodeJS.ProcessEnv = process.env): string {
const safe = sessionId.replace(/[^A-Za-z0-9_-]/g, '_').slice(0, 128);
if (!safe) throw new StateError(recordsDir(env), 'hook payload has no session_id');
return path.join(recordsDir(env), `${safe}.json`);
}
export const runKey = (skill: string, load: SkillLoad) => `${canonicalSkill(skill)}@${load.line}`;
export function readRecord(file: string): SessionRecord | null {
let text: string;
try {
text = fs.readFileSync(file, 'utf8');
} catch (e) {
if ((e as NodeJS.ErrnoException)?.code === 'ENOENT' || (e as NodeJS.ErrnoException)?.code === 'ENOTDIR') return null;
throw new StateError(file, `cannot read record (${errText(e)})`);
}
let rec: unknown;
try { rec = JSON.parse(text); } catch { throw new StateError(file, 'corrupt record (not JSON)'); }
if (!isObject(rec) || rec.version !== 1 || typeof rec.session_id !== 'string' || !isObject(rec.runs)) {
throw new StateError(file, 'corrupt record (unexpected shape)');
}
return rec as SessionRecord;
}
export function writeRecord(file: string, rec: SessionRecord): void {
try {
fs.mkdirSync(path.dirname(file), { recursive: true, mode: 0o700 });
atomicWriteSync(file, `${JSON.stringify(rec, null, 2)}\n`, { mode: 0o600 });
} catch (e) {
throw new StateError(file, `cannot write record (${errText(e)})`);
}
}
/**
* Block at most once per skill run: an unmet contract blocks unless this run was already blocked, which makes it a miss.
* A run recorded met, met_waived or met_after_block is final: later unrelated turns in the session never reopen it.
*/
export function nextOutcome(prev: RunRecord | undefined, result: CheckResult): { outcome: Outcome; block: boolean } {
if (prev && (prev.outcome === 'met' || prev.outcome === 'met_waived' || prev.outcome === 'met_after_block')) return { outcome: prev.outcome, block: false };
if (result.met) return { outcome: prev?.blocked_at ? 'met_after_block' : result.waived.length ? 'met_waived' : 'met', block: false };
return prev?.blocked_at ? { outcome: 'miss', block: false } : { outcome: 'blocked', block: true };
}
/** Count one unknown transcript shape in $GSTACK_STATE_ROOT/hooks/skill-contract.canary.json; returns the new count, 0 when it cannot be written. */
export function bumpCanary(why: string, env: NodeJS.ProcessEnv = process.env): number {
try {
const file = path.join(resolveStateRoot(env), 'hooks', 'skill-contract.canary.json');
let count = 0;
try { count = Number(JSON.parse(fs.readFileSync(file, 'utf8')).count) || 0; } catch { /* first one */ }
fs.mkdirSync(path.dirname(file), { recursive: true, mode: 0o700 });
atomicWriteSync(file, `${JSON.stringify({ count: count + 1, last_cause: why, updated_at: new Date().toISOString() })}\n`, { mode: 0o600 });
return count + 1;
} catch {
return 0;
}
}
export interface ReportRow { session: string; skill: string; outcome: Outcome | 'fixed_after_block'; missing: string[]; waived: string[]; transcript: string }
/** Every recorded run; a run still marked blocked is re-checked against its transcript when a recheck is given. */
export function reportRows(records: SessionRecord[], recheck?: (rec: SessionRecord, run: RunRecord) => CheckResult | null): ReportRow[] {
return records.flatMap(rec => Object.values(rec.runs).map(run => {
const row: ReportRow = { session: rec.session_id, skill: run.skill, outcome: run.outcome, missing: run.missing, waived: run.waived, transcript: rec.transcript_path };
if (run.outcome !== 'blocked' || !recheck) return row;
const again = recheck(rec, run);
if (!again) return row;
return again.met
? { ...row, outcome: 'fixed_after_block' as const, missing: [], waived: again.waived }
: { ...row, outcome: 'miss' as const, missing: again.missing, waived: again.waived };
}));
}
/** A miss for follow: blocked and never fixed, stopped again unmet, or met only through a waiver. */
export const isMiss = (row: ReportRow) => row.outcome === 'blocked' || row.outcome === 'miss' || row.waived.length > 0;
export function readAllRecords(env: NodeJS.ProcessEnv = process.env): { records: SessionRecord[]; errors: string[] } {
const dir = recordsDir(env);
const records: SessionRecord[] = [];
const errors: string[] = [];
let names: string[] = [];
try { names = fs.readdirSync(dir).filter(n => n.endsWith('.json')).sort(); } catch { return { records, errors }; }
for (const name of names) {
try {
const rec = readRecord(path.join(dir, name));
if (rec) records.push(rec);
} catch (e) {
errors.push(e instanceof StateError ? `${e.file}: ${e.why}` : String(e));
}
}
return { records, errors };
}
+305
View File
@@ -0,0 +1,305 @@
/**
* skill-route — deterministic prompt-to-skill routing (GST-1088, lane A).
*
* The UserPromptSubmit hook (hosts/claude/hooks/skill-route-hook.ts) and
* `gstack-skill-route --explain` score a prompt against the `triggers:` lists
* in the installed skills' SKILL.md frontmatter. No model call, no network:
* the same prompt and the same installed skills always give the same route.
*
* Also home to what both skill hooks share: the on/off switches
* (skill_routing, skill_contracts), the installed-skills directory, the
* bounded stdin reader and the trace log under $GSTACK_STATE_ROOT/hooks/.
* Docs: docs/skill-routing-and-contracts.md.
*/
import * as fs from 'fs';
import * as os from 'os';
import * as path from 'path';
import { atomicWriteSync } from './fs-atomic';
import { frontmatterList } from './skill-triggers';
import { readConfigKey, resolveStateRoot } from './state-root';
/** A route needs at least this share of a trigger's content words (2 of 3, 3 of 4; both of 2). */
export const ROUTE_THRESHOLD = 0.66;
/** Longer prompts (pasted logs, files) are never routed: their words match everything. */
export const PROMPT_MAX_CHARS = 8000;
const FRONTMATTER_MAX_BYTES = 16 * 1024;
const MAX_SKILLS = 256;
const MAX_TRIGGER_CHARS = 200;
const TRACE_MAX_BYTES = 1024 * 1024;
const FAILURE_SESSIONS_KEPT = 64;
export type SwitchName = 'skill_routing' | 'skill_contracts';
export const SWITCH_ENV: Record<SwitchName, string> = {
skill_routing: 'GSTACK_SKILL_ROUTING',
skill_contracts: 'GSTACK_SKILL_CONTRACTS',
};
export function normalizeSwitch(raw: string | null | undefined): 'on' | 'off' | null {
const v = String(raw ?? '').trim().toLowerCase();
if (['on', 'yes', 'y', 'true', '1'].includes(v)) return 'on';
if (['off', 'no', 'n', 'false', '0', 'skip'].includes(v)) return 'off';
return null;
}
export interface SwitchReading { on: boolean; source: string; invalid?: string }
/** env GSTACK_SKILL_ROUTING / GSTACK_SKILL_CONTRACTS > config key > default off. An unrecognized value is off. */
export function readSkillSwitch(name: SwitchName, env: NodeJS.ProcessEnv = process.env): SwitchReading {
const envName = SWITCH_ENV[name];
const raw = env[envName];
if (raw !== undefined && raw !== '') {
const v = normalizeSwitch(raw);
return v ? { on: v === 'on', source: `env ${envName}` } : { on: false, source: `env ${envName}`, invalid: raw };
}
const cfg = readConfigKey(name, env);
if (cfg === null) return { on: false, source: 'default' };
const v = normalizeSwitch(cfg);
return v ? { on: v === 'on', source: `config ${name}` } : { on: false, source: `config ${name}`, invalid: cfg };
}
/** Where Claude Code reads skills: GSTACK_INSTALLED_SKILLS_DIR, else ${CLAUDE_CONFIG_DIR:-~/.claude}/skills. */
export function installedSkillsDir(env: NodeJS.ProcessEnv = process.env): string {
if (env.GSTACK_INSTALLED_SKILLS_DIR) return env.GSTACK_INSTALLED_SKILLS_DIR;
return path.join(env.CLAUDE_CONFIG_DIR || path.join(env.HOME || os.homedir(), '.claude'), 'skills');
}
/** The frontmatter between the leading `---` lines, read from at most the first 16 KiB; null without one. */
export function readFrontmatter(file: string): string | null {
const fd = fs.openSync(file, 'r');
try {
const buf = Buffer.alloc(FRONTMATTER_MAX_BYTES);
const text = buf.subarray(0, fs.readSync(fd, buf, 0, FRONTMATTER_MAX_BYTES, 0)).toString('utf8').replace(/\r\n/g, '\n');
if (!text.startsWith('---\n')) return null;
const end = text.indexOf('\n---', 4);
return end < 0 ? null : text.slice(4, end);
} finally {
fs.closeSync(fd);
}
}
export function frontmatterName(frontmatter: string, fallback: string): string {
const m = /^name:\s*(.+)$/m.exec(frontmatter);
return m ? m[1].trim().replace(/^["']|["']$/g, '') : fallback;
}
export interface InstalledSkill { name: string; file: string; frontmatter: string; triggers: string[] }
/** Every installed skill with frontmatter, sorted by directory; `problems` names skills whose triggers the repo's reader cannot use. */
export function scanInstalledSkills(dir: string): { skills: InstalledSkill[]; problems: string[] } {
const skills: InstalledSkill[] = [];
const problems: string[] = [];
let entries: string[];
try {
entries = fs.readdirSync(dir).sort().slice(0, MAX_SKILLS);
} catch {
return { skills, problems: [`skills directory unreadable: ${dir}`] };
}
for (const entry of entries) {
const file = path.join(dir, entry, 'SKILL.md');
let frontmatter: string | null;
try {
frontmatter = readFrontmatter(file);
} catch {
continue;
}
if (frontmatter === null) continue;
const name = frontmatterName(frontmatter, entry);
let triggers: string[] = [];
if (/^triggers:/m.test(frontmatter)) {
triggers = frontmatterList(`---\n${frontmatter}\n---`, 'triggers');
if (!triggers.length || triggers.some(t => !t || t.length > MAX_TRIGGER_CHARS)) {
problems.push(`${name}: malformed triggers`);
triggers = [];
}
}
skills.push({ name, file, frontmatter, triggers });
}
return { skills, problems };
}
const STOPWORDS = new Set(('a an the this that these those it its i me my we our us you your he she they them their ' +
'is are was were be been being am do does did done to of in on at by for from with as into onto about and or but ' +
'so if then than just please can could would should will shall may might must some any all here there now ' +
'up out over get got let lets want need make sure ok okay hey hi').split(' '));
/** Lowercase, light stemming: plural s/es, -ing and -ed, with a doubled final consonant undone. */
export function stem(word: string): string {
let w = word.toLowerCase().replace(/'s$/, '');
if (w.length > 4 && /(?:xes|sses|shes|ches|zes)$/.test(w)) w = w.slice(0, -2);
else if (w.length > 3 && w.endsWith('s') && !w.endsWith('ss') && !w.endsWith('us')) w = w.slice(0, -1);
if (w.length > 5 && w.endsWith('ing')) w = w.slice(0, -3);
else if (w.length > 4 && w.endsWith('ed')) w = w.slice(0, -2);
if (/([b-df-hj-np-tv-z])\1$/.test(w) && !/(?:ll|ss|zz|ff)$/.test(w)) w = w.slice(0, -1);
return w;
}
const words = (text: string) => text.toLowerCase().match(/[a-z0-9]+(?:'[a-z]+)?/g) ?? [];
export function tokens(text: string): string[] {
return words(text).map(stem);
}
export function contentTokens(text: string): string[] {
return [...new Set(words(text).filter(w => !STOPWORDS.has(w)).map(stem))];
}
export interface RouteCandidate { skill: string; score: number; matched: number; trigger: string; file: string }
/** A lone content word specific enough to route on by itself: five or more letters, not a question word. */
const SPECIFIC_WORD = /^(?!(?:what|which|where|whose)$)[a-z0-9]{5,}$/;
/**
* Every skill whose best trigger shares a content word with the prompt, best first (score, matched words,
* name). A trigger's score is 1 when its whole phrase appears in the prompt, else the share of its content
* words (stopwords dropped, lightly stemmed) that the prompt contains. A trigger with one content word scores
* only on its whole phrase unless that word is specific (SPECIFIC_WORD).
*/
export function scorePrompt(prompt: string, skills: InstalledSkill[]): RouteCandidate[] {
const promptSeq = ` ${tokens(prompt).join(' ')} `;
const promptSet = new Set(contentTokens(prompt));
const out: RouteCandidate[] = [];
for (const skill of skills) {
let best: RouteCandidate | null = null;
for (const trigger of skill.triggers) {
const content = contentTokens(trigger);
if (!content.length) continue;
const matched = content.filter(w => promptSet.has(w)).length;
const phrase = promptSeq.includes(` ${tokens(trigger).join(' ')} `);
const weakSingle = content.length === 1 && !SPECIFIC_WORD.test(content[0]);
const score = phrase ? 1 : weakSingle ? 0 : Math.round((matched / content.length) * 100) / 100;
if (score > 0 && (!best || score > best.score || (score === best.score && matched > best.matched))) {
best = { skill: skill.name, score, matched, trigger, file: skill.file };
}
}
if (best) out.push(best);
}
return out.sort((a, b) => b.score - a.score || b.matched - a.matched || a.skill.localeCompare(b.skill));
}
/**
* The skills the hook may route to: plan-v2 decision 4's discovery targets. Run D's prompts showed routing
* every installed skill's triggers false-routes most non-fit prompts (docs/skill-routing-and-contracts.md).
*/
export const ROUTED_SKILLS = ['investigate', 'review', 'qa', 'verify'];
const isRouted = (name: string) => ROUTED_SKILLS.includes(name.toLowerCase().replace(/^gstack[-:]/, ''));
export type RouteDecision =
| { route: RouteCandidate; candidates: RouteCandidate[] }
| { route: null; reason: 'slash-command' | 'prompt-too-long' | 'empty' | 'below-threshold'; candidates: RouteCandidate[] };
export function routePrompt(prompt: string, skills: InstalledSkill[], threshold = ROUTE_THRESHOLD): RouteDecision {
const trimmed = prompt.trim();
if (!trimmed) return { route: null, reason: 'empty', candidates: [] };
if (trimmed.startsWith('/')) return { route: null, reason: 'slash-command', candidates: [] };
if (prompt.length > PROMPT_MAX_CHARS) return { route: null, reason: 'prompt-too-long', candidates: [] };
const candidates = scorePrompt(prompt, skills.filter(s => isRouted(s.name)));
const top = candidates[0];
return top && top.score >= threshold ? { route: top, candidates } : { route: null, reason: 'below-threshold', candidates };
}
export type HintVariant = 'name' | 'step' | 'contract';
export const HINT_VARIANTS: HintVariant[] = ['name', 'step', 'contract'];
/** The default line (d6 pilot); GSTACK_SKILL_ROUTE_HINT selects another variant for the pilot. */
export const DEFAULT_HINT_VARIANT: HintVariant = 'name';
export function hintVariant(env: NodeJS.ProcessEnv = process.env): HintVariant {
const v = (env.GSTACK_SKILL_ROUTE_HINT || '').trim().toLowerCase();
return (HINT_VARIANTS as string[]).includes(v) ? v as HintVariant : DEFAULT_HINT_VARIANT;
}
/** The one additionalContext line: its source, its off switch, the skill, and the ask to say gstack routed it. */
export function hintLine(route: RouteCandidate, extra = ''): string {
const s = `/${route.skill}`;
const trigger = route.trigger.replace(/["\s]+/g, ' ').trim();
return `gstack skill routing (UserPromptSubmit hook; matched the installed ${s} trigger "${trigger}"; off: gstack-config set skill_routing off): ` +
`this request fits ${s}. Run ${s} with the Skill tool before you start, and tell the user in one sentence that gstack routed this to ${s}.` +
(extra.trim() ? ` ${extra.replace(/\s+/g, ' ').trim()}` : '');
}
/** First sentence under a skill body's first "Step 1", "Phase 1" or "How to work" heading, plain text, at most 200 chars; '' without one. */
export function firstStep(skillMd: string): string {
const m = /^#{2,3}\s+(?:Step\s*0*1\b|Phase\s*0*1\b|How to work\b).*$/im.exec(skillMd);
if (!m) return '';
const para = skillMd.slice(m.index + m[0].length).split(/\n\s*\n/).map(p => p.trim())
.find(p => p && !p.startsWith('```') && !p.startsWith('#'));
if (!para) return '';
const plain = para.replace(/[*_`]/g, '').replace(/\s+/g, ' ');
const sentence = (/^.*?[.!?](?=\s|$)/.exec(plain)?.[0] ?? plain).trim();
return sentence.length > 200 ? `${sentence.slice(0, 197).replace(/\s+\S*$/, '')}...` : sentence;
}
export interface BoundedStdin { text: string; overCap: boolean }
/** A hook payload: returns at EOF, at the first complete JSON value, at the deadline, or as soon as it passes capBytes. */
export function readStdinBounded(capBytes: number, ms: number, stream: NodeJS.ReadStream = process.stdin): Promise<BoundedStdin> {
return new Promise((resolve) => {
let buf = '';
let bytes = 0;
let settled = false;
const done = (overCap: boolean) => {
if (settled) return;
settled = true;
clearTimeout(timer);
stream.removeAllListeners('data');
try { stream.destroy(); } catch { /* already closed */ }
resolve({ text: overCap ? '' : buf, overCap });
};
const timer = setTimeout(() => done(false), ms);
stream.on('data', (chunk: Buffer | string) => {
const b = typeof chunk === 'string' ? Buffer.from(chunk) : chunk;
bytes += b.length;
if (bytes > capBytes) { done(true); return; }
buf += b.toString('utf8');
try { JSON.parse(buf); done(false); } catch { /* keep reading */ }
});
stream.on('end', () => done(false));
stream.on('error', () => done(false));
});
}
export type TraceHook = 'skill-route' | 'skill-contract';
export function traceLogPath(hook: TraceHook, env: NodeJS.ProcessEnv = process.env): string {
return path.join(resolveStateRoot(env), 'hooks', `${hook}.log`);
}
const traceValue = (v: string | number | boolean) => {
const s = String(v).replace(/[\r\n\t]+/g, ' ');
return /[\s"=]/.test(s) ? JSON.stringify(s) : s;
};
/**
* Append one trace line to $GSTACK_STATE_ROOT/hooks/<hook>.log (rotated to .log.1 past 1 MiB). A failure also
* bumps the session's count in <hook>.failures.json and the line carries `failures=<n>`. Never throws, never
* writes stderr, never logs prompt text.
*/
export function appendHookTrace(hook: TraceHook, sessionId: string, fields: Record<string, string | number | boolean>,
failed = false, env: NodeJS.ProcessEnv = process.env): void {
try {
const log = traceLogPath(hook, env);
const dir = path.dirname(log);
fs.mkdirSync(dir, { recursive: true, mode: 0o700 });
const sid = sessionId.replace(/[^A-Za-z0-9_-]/g, '').slice(0, 64) || 'unknown';
const parts = [`session=${sid}`, ...Object.entries(fields).map(([k, v]) => `${k}=${traceValue(v)}`)];
if (failed) {
const counter = path.join(dir, `${hook}.failures.json`);
let sessions: Record<string, number> = {};
try {
const parsed = JSON.parse(fs.readFileSync(counter, 'utf8'));
if (parsed && typeof parsed.sessions === 'object' && parsed.sessions) sessions = parsed.sessions;
} catch { /* first failure, or a corrupt counter that starts over */ }
const n = (Number(sessions[sid]) || 0) + 1;
delete sessions[sid];
sessions[sid] = n;
const kept = Object.fromEntries(Object.entries(sessions).slice(-FAILURE_SESSIONS_KEPT));
atomicWriteSync(counter, `${JSON.stringify({ sessions: kept })}\n`, { mode: 0o600 });
parts.push(`failures=${n}`);
}
try {
if (fs.statSync(log).size > TRACE_MAX_BYTES) fs.renameSync(log, `${log}.1`);
} catch { /* no log yet */ }
fs.appendFileSync(log, `${new Date().toISOString()} ${parts.join(' ')}\n`, { mode: 0o600 });
} catch {
// tracing is best-effort; a hook never fails because its log did
}
}
+28 -4
View File
@@ -4,7 +4,8 @@
* lines the branch wrote itself. Split from restamp.ts (module-size ratchet).
*/
import { spawnSync } from 'node:child_process';
import { branchAddedLines, forkVersion, rewriteStamps, stampMentions } from './restamp';
import { branchAddedLines, escapeRegExp, forkVersion, rewriteStamps, stampMentions } from './restamp';
import { VERSION_SLOT } from './ship-policy';
import { cmpVersion, parseVersion, versionWidth } from './version-source';
/**
@@ -26,12 +27,20 @@ import { cmpVersion, parseVersion, versionWidth } from './version-source';
* stage's bound for its own lines (GST-895, GST-963): finish's stampedFiles,
* the tree receipt and the receipt carry all pass the same side. When the
* version does not move (a slot that keeps its claimed VERSION), only such a
* branch line may change: the stage moves nothing else then.
* branch line may change: the stage moves nothing else then. On such a branch
* line a declared placeholder (`stamp_placeholders`, GST-1026: gbrain's
* `since: 'TBD'`) may also become the new version, but only inside a
* `stamp_patterns` template with literal text around its slot, never a built-in
* form: with at least one such rewrite, the line must then be the added line,
* or one whose remaining stamp-form moves pass the fork-version bound.
*/
export function stampOnlyLine(removed: string, added: string, oldV: string, newV: string, templates: readonly string[] = [], side?: BranchSide, at?: number): boolean {
export function stampOnlyLine(removed: string, added: string, oldV: string, newV: string, templates: readonly string[] = [], side?: BranchSide, at?: number, placeholders: readonly string[] = []): boolean {
if (removed === added) return false;
if (oldV !== newV && stampOnlyWithin(removed, added, oldV, newV, templates)) return true;
return !!side && at !== undefined && side.at.has(at) && side.added.has(removed) && stampOnlyWithin(removed, added, side.fork, newV, templates);
if (!side || at === undefined || !side.at.has(at) || !side.added.has(removed)) return false;
if (stampOnlyWithin(removed, added, side.fork, newV, templates)) return true;
const filled = fillPlaceholders(removed, newV, templates, placeholders);
return filled.count > 0 && (filled.text === added || stampOnlyWithin(filled.text, added, side.fork, newV, templates));
}
/**
@@ -64,6 +73,21 @@ export function branchSides(repoRoot: string, baseRef: string, head: string, fal
};
}
/** Each declared placeholder rewritten to `newV` inside each template with literal text around its slot (a bare edge at a word boundary). */
function fillPlaceholders(removed: string, newV: string, templates: readonly string[], placeholders: readonly string[]): { text: string; count: number } {
let text = removed;
let count = 0;
for (const t of templates) {
const [pre, post] = t.split(VERSION_SLOT) as [string, string];
if (!pre && !post) continue;
for (const p of placeholders) {
const re = new RegExp(`${escapeRegExp(pre)}${pre ? '' : '(?<![A-Za-z0-9_.])'}${escapeRegExp(p)}${post ? '' : '(?![A-Za-z0-9_.])'}${escapeRegExp(post)}`, 'g');
text = text.replace(re, () => { count++; return pre + newV + post; });
}
}
return { text, count };
}
function stampOnlyWithin(removed: string, added: string, oldV: string, newV: string, templates: readonly string[]): boolean {
const floor = parseVersion(oldV);
const width = versionWidth(oldV);
+5 -4
View File
@@ -78,15 +78,16 @@ export function classifyFiles(repoRoot: string, fileDiffs: FileDiff[], gated: st
/**
* Every removed/added pair differs only by stamp rewrites to the new version:
* the old version, or a tentative one (newer than it) the stamp_paths stage
* moved (#3512), each in a stamp form (`stampOnlyLine`, bounded by the fork version on the branch's own lines).
* moved (#3512), each in a stamp form (`stampOnlyLine`, bounded by the fork version on the branch's own lines, where a
* declared placeholder may also move inside a stamp_patterns template, GST-1026).
*/
function stampOnlyHunks(hunks: FileDiff['hunks'], oldV: string, newV: string, templates: readonly string[], side?: BranchSide): { why: string; line: number } | null {
function stampOnlyHunks(hunks: FileDiff['hunks'], oldV: string, newV: string, templates: readonly string[], side?: BranchSide, placeholders: readonly string[] = []): { why: string; line: number } | null {
for (const h of hunks) {
if (h.removed.length !== h.added.length) return { why: `${h.removed.length} line(s) removed, ${h.added.length} added`, line: h.at };
for (let i = 0; i < h.removed.length; i++) {
const r = h.removed[i]!;
const a = h.added[i]!;
if (!stampOnlyLine(r, a, oldV, newV, templates, side, h.removedAt[i])) return { why: `${JSON.stringify(r.trim().slice(0, 60))} -> ${JSON.stringify(a.trim().slice(0, 60))}`, line: h.at + i };
if (!stampOnlyLine(r, a, oldV, newV, templates, side, h.removedAt[i], placeholders)) return { why: `${JSON.stringify(r.trim().slice(0, 60))} -> ${JSON.stringify(a.trim().slice(0, 60))}`, line: h.at + i };
}
}
return null;
@@ -130,7 +131,7 @@ export function classifyFile(repoRoot: string, f: FileDiff, gated: string, curre
const rule = matchesAny(policy.stamp_paths, f.path) ? 'stamp_paths' : matchesAny(policy.release_outputs, f.path) ? 'release_outputs' : matchesAny(policy.stamp_files, f.path) ? 'stamp_files' : null;
const own = rule ? side?.(f.path) : undefined;
if (oldV && newV && (oldV !== newV || own) && rule) {
const bad = stampOnlyHunks(f.hunks, oldV, newV, policy.stamp_patterns, own);
const bad = stampOnlyHunks(f.hunks, oldV, newV, policy.stamp_patterns, own, policy.stamp_placeholders);
return bad ? changed(rule, bad.why, bad.line) : stamp(rule);
}
return changed('other', `${f.hunks.length} hunk(s) outside every stamp rule`);