feat(plan): plan 1.8, the plan gate, the design part and the plan-side checks
Wave 1, order 3 (P3 plan mode as the start, P2 design as its own part, P9
plan side). Contracts K3-K6 for the later merges.
- plan_version 1.8 (1.7 plans stay valid): Test first on every step, one
End-to-end ("run the thing") step, optional Interface in / Interface out /
Review focus fields, a ceremony class in the header (absent =
architectural), and for architectural plans a ## Design part: module map
without cycles, architecture rules that name their enforcer and the step
that proves them, a requirements check asked as keep / change.
- K3 lib/validators/plan-gate.mjs: the critic verdict is recorded INSIDE the
plan (Claude Code's plan approval hands a hook only the plan file's text),
bound to the hash taken before the reviewers read it; 0 go / 1 stop / 2
could not run; at most 3 critic rounds, then stop and ask.
- K5 lib/verification/arch-rules.mjs: runs the target repo's own
dependency-cruiser or import-linter (installs nothing; 2 = not measured,
never green); --prove --rule shows pass -> fail on a planted violation of
that rule -> pass. Verified against dependency-cruiser 18.5.0 and
import-linter 2.15; madge 8.0.0 left out (missed a planted .mjs cycle).
- K6 lib/verification/plan-scope-sweep.mjs: changed files no step names.
- /trekplan Phases 8-10: design part, bounded critic loop, the gate before
"execute", a plan-mode path through ExitPlanMode (not yet run end to end
inside plan mode). plan-critic 10 -> 12 dimensions; scope-guardian checks
the ceremony.
- Fixed on the way: plan-validator and plan-review-dedup used the
file://argv[1] main guard (no-op under a symlink or a space); CRLF plans
failed every manifest; the dedup dropped finding severity.
Borrowed, credited in place: obra/superpowers, Kiro (docs), addyosmani/
agent-skills, mattpocock/skills, GitHub Spec Kit, bashebr/ai-native-sdlc,
Boris Cherny's public threads.
Suite 1497 (1495 / 0 / 2), 86 new tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
parent
2c89a6b580
commit
f5ca220bc0
SSH key fingerprint:
SHA256:JakMjO6FTBBzN0Bhfj9saOoEjaFxlSdYuZQQpM/lF9Q
24 changed files with 3115 additions and 61 deletions
352
lib/validators/plan-gate.mjs
Normal file
352
lib/validators/plan-gate.mjs
Normal file
|
|
@ -0,0 +1,352 @@
|
|||
// lib/validators/plan-gate.mjs
|
||||
// K3 — the plan gate: may this plan go on to execution?
|
||||
//
|
||||
// node plan-gate.mjs <plan.md> [--json]
|
||||
// 0 = the plan may go on
|
||||
// 1 = stop: invalid plan (plan-validator --strict, plan_version >= 1.8),
|
||||
// no critic verdict, a malformed verdict, a verdict the plan has
|
||||
// moved away from (stale), or an open BLOCKER
|
||||
// 2 = could not run (missing/unreadable plan, bad arguments)
|
||||
//
|
||||
// node plan-gate.mjs --hash <plan.md>
|
||||
// Prints the sha256 of the plan's operative text. Phase 9 takes it BEFORE
|
||||
// the reviewers read the plan and hands it to --record.
|
||||
//
|
||||
// node plan-gate.mjs --record <plan.md> --reviewed <sha256> [--extra-round]
|
||||
// stdin = plan-review-dedup.mjs output. Writes this round's verdict into
|
||||
// the plan. 0 = written; 1 = refused: the round cap is spent (stop and ask
|
||||
// the operator; `--extra-round` only after their yes, and only then);
|
||||
// 2 = could not run: bad stdin, a reviewer missing from the dedup input,
|
||||
// the plan changed since --hash (the verdict would bless unreviewed text),
|
||||
// an unreadable plan or a failed write. Nothing is written on 1 or 2.
|
||||
//
|
||||
// Where the verdict lives: INSIDE the plan, as the `## Critic Verdict` section
|
||||
// with one fenced json block. Claude Code's plan approval hands a hook the
|
||||
// plan file's text (`tool_input.plan`, `planFilePath`) and nothing next to
|
||||
// it, so a verdict in a sibling file would be invisible there.
|
||||
//
|
||||
// What the verdict is bound to: sha256 of the plan's operative text — the plan
|
||||
// minus its review log, with CRLF and line-end whitespace normalized. The
|
||||
// review log is the `> **Plan quality:` header line and the level-2 sections
|
||||
// titled exactly Critic Verdict, Revisions, Plan Quality Score and
|
||||
// "Adversarial Pass 2 (v5.1.1 high-effort)" — each only up to its first
|
||||
// other heading, so a step or section placed under a log heading is hashed.
|
||||
// Edit a step, the design or the verification after the critic ran, and the
|
||||
// verdict is stale until the next round.
|
||||
//
|
||||
// The loop is bounded (Kiro's bounded review loop): MAX_ROUNDS critic rounds,
|
||||
// then stop and ask; never silently on. Unrated findings count as blockers.
|
||||
// The verdict is a trace of a process, not a signature: anything that can
|
||||
// write the plan can write the block.
|
||||
|
||||
import { readFileSync, existsSync, realpathSync, writeFileSync, renameSync, rmSync } from 'node:fs';
|
||||
import { createHash } from 'node:crypto';
|
||||
import { fileURLToPath } from 'node:url';
|
||||
import { validatePlanContent } from './plan-validator.mjs';
|
||||
import { splitSections, fenceTracker, extractPlanVersion, versionAtLeast } from '../parsers/plan-schema.mjs';
|
||||
import { normalizeSeverity } from '../review/plan-review-dedup.mjs';
|
||||
|
||||
export const MAX_ROUNDS = 3;
|
||||
export const VERDICT_SCHEMA = 'voyage.plan-verdict/1';
|
||||
export const VERDICT_HEADING = '## Critic Verdict';
|
||||
export const REQUIRED_REVIEWERS = Object.freeze(['plan-critic', 'scope-guardian']);
|
||||
// Level-2 sections that are review log, not plan: excluded from the hash.
|
||||
export const REVIEW_LOG_SECTIONS = Object.freeze([
|
||||
'Critic Verdict', 'Revisions', 'Plan Quality Score', 'Adversarial Pass 2 (v5.1.1 high-effort)',
|
||||
]);
|
||||
// The quality line alone; a line that also carries `plan_version`/`ceremony` stays hashed.
|
||||
const QUALITY_LINE = /^> \*\*Plan quality:[^`]*$/;
|
||||
|
||||
const isLogTitle = t => t !== null && REVIEW_LOG_SECTIONS.includes(t);
|
||||
const norm = t => String(t).replace(/\r\n?/g, '\n');
|
||||
|
||||
/** A log section counts as log only up to its first other heading (outside code). */
|
||||
function logPart(sectionText) {
|
||||
const lines = sectionText.split('\n');
|
||||
const fence = fenceTracker();
|
||||
for (let i = 1; i < lines.length; i++) {
|
||||
if (fence.feed(lines[i])) continue;
|
||||
if (/^#{1,6} /.test(lines[i])) return { log: lines.slice(0, i).join('\n'), rest: lines.slice(i).join('\n') };
|
||||
}
|
||||
return { log: sectionText, rest: '' };
|
||||
}
|
||||
|
||||
/** The plan minus its review log, normalized (LF, no line-end whitespace, one trailing newline). */
|
||||
export function operativeText(text) {
|
||||
const kept = splitSections(norm(text), 2)
|
||||
.map(s => (isLogTitle(s.title) ? logPart(s.text).rest : s.text))
|
||||
.join('\n');
|
||||
return kept
|
||||
.split('\n')
|
||||
.filter(l => !QUALITY_LINE.test(l))
|
||||
.map(l => l.replace(/[ \t]+$/, ''))
|
||||
.join('\n')
|
||||
.replace(/\n{3,}/g, '\n\n')
|
||||
.trim() + '\n';
|
||||
}
|
||||
|
||||
export function planHash(text) {
|
||||
return createHash('sha256').update(operativeText(text)).digest('hex');
|
||||
}
|
||||
|
||||
/**
|
||||
* @returns {{found: boolean, placeholder: boolean, verdict: object|null, error: string|null}}
|
||||
* placeholder = a `## Critic Verdict` heading with no fenced block at all
|
||||
* (copied prose, not a verdict): --record replaces it, the gate stops on it.
|
||||
*/
|
||||
export function extractVerdict(text) {
|
||||
const secs = splitSections(norm(text), 2).filter(s => s.title === 'Critic Verdict');
|
||||
const none = { found: false, placeholder: false, verdict: null };
|
||||
if (secs.length === 0) return { ...none, error: null };
|
||||
if (secs.length > 1) return { ...none, found: true, error: `${secs.length} "## Critic Verdict" sections; expected one` };
|
||||
const log = logPart(secs[0].text).log;
|
||||
const m = log.match(/```json[ \t]*\n([\s\S]*?)\n[ \t]*```/);
|
||||
if (!m) {
|
||||
const placeholder = !/^\s{0,3}(```|~~~)/m.test(log);
|
||||
return { ...none, found: true, placeholder, error: 'no ```json block in ## Critic Verdict' };
|
||||
}
|
||||
let v;
|
||||
try { v = JSON.parse(m[1]); } catch (e) { return { ...none, found: true, error: `verdict JSON does not parse: ${e.message}` }; }
|
||||
const okShape = v && typeof v === 'object' && v.schema === VERDICT_SCHEMA && Number.isInteger(v.round) && v.round >= 1 &&
|
||||
Number.isInteger(v.max_rounds) && v.max_rounds >= 1 && v.max_rounds <= MAX_ROUNDS &&
|
||||
Number.isInteger(v.extra_rounds ?? 0) && (v.extra_rounds ?? 0) >= 0 &&
|
||||
typeof v.plan_sha256 === 'string' && Array.isArray(v.findings) &&
|
||||
v.findings.every(f => f && typeof f === 'object' && typeof f.severity === 'string');
|
||||
if (!okShape) {
|
||||
return { ...none, found: true, error: `verdict block is not ${VERDICT_SCHEMA} (schema, round, max_rounds <= ${MAX_ROUNDS}, extra_rounds, plan_sha256, findings with severity)` };
|
||||
}
|
||||
return { found: true, placeholder: false, verdict: v, error: null };
|
||||
}
|
||||
|
||||
function verdictFor(counts) {
|
||||
if (counts.blocker > 0) return 'BLOCK';
|
||||
if (counts.major > 0) return 'REVISE';
|
||||
return 'PASS';
|
||||
}
|
||||
|
||||
function renderSection(v) {
|
||||
return [
|
||||
VERDICT_HEADING,
|
||||
'',
|
||||
`<!-- Written by lib/validators/plan-gate.mjs --record (round ${v.round} of ${v.max_rounds}` +
|
||||
`${v.extra_rounds ? ` + ${v.extra_rounds} extra` : ''}). Do not edit by hand: the gate re-hashes the plan and reads this block. -->`,
|
||||
'',
|
||||
'```json',
|
||||
JSON.stringify(v, null, 2),
|
||||
'```',
|
||||
'',
|
||||
].join('\n');
|
||||
}
|
||||
|
||||
/** Replace (or append) the `## Critic Verdict` section; anything after its log part is kept. */
|
||||
function placeSection(text, section) {
|
||||
const secs = splitSections(norm(text), 2);
|
||||
const idx = secs.findIndex(s => s.title === 'Critic Verdict');
|
||||
if (idx === -1) return norm(text).replace(/\s*$/, '\n\n') + section;
|
||||
const parts = secs.map(s => s.text);
|
||||
const { rest } = logPart(parts[idx]);
|
||||
parts[idx] = section.replace(/\n$/, '') + (rest ? `\n\n${rest}` : '');
|
||||
return parts.join('\n').replace(/\s*$/, '\n');
|
||||
}
|
||||
|
||||
/** Both reviewers must have been present in the dedup input; an absent one is not a clean review. */
|
||||
function missingReviewers(dedup) {
|
||||
const present = new Set(
|
||||
(Array.isArray(dedup.sources) ? dedup.sources : []).filter(s => s && s.present === true).map(s => s.agent),
|
||||
);
|
||||
return REQUIRED_REVIEWERS.filter(r => !present.has(r));
|
||||
}
|
||||
|
||||
/**
|
||||
* Record one critic round into the plan text.
|
||||
* @param {string} text - the plan
|
||||
* @param {{findings: Array<object>, sources: Array<{agent: string, present: boolean}>}} dedup
|
||||
* plan-review-dedup.mjs output
|
||||
* @param {{reviewed?: string, extraRound?: boolean, maxRounds?: number, now?: string}} [opts]
|
||||
* reviewed = planHash of the text the reviewers read (from --hash)
|
||||
* @returns {{code: 0|1|2, text?: string, verdict?: object, message: string}}
|
||||
*/
|
||||
export function recordVerdict(text, dedup, opts = {}) {
|
||||
if (!dedup || typeof dedup !== 'object' || !Array.isArray(dedup.findings)) {
|
||||
return { code: 2, message: 'record input must be plan-review-dedup output: {"findings": [...], "sources": [...]}' };
|
||||
}
|
||||
if (dedup.findings.some(f => !f || typeof f !== 'object')) {
|
||||
return { code: 2, message: 'record input has a finding that is not an object' };
|
||||
}
|
||||
const missing = missingReviewers(dedup);
|
||||
if (missing.length > 0) {
|
||||
return { code: 2, message: `reviewer(s) missing from the dedup input: ${missing.join(', ')} — an absent review is not a clean one; rerun the round` };
|
||||
}
|
||||
const current = planHash(text);
|
||||
if (typeof opts.reviewed !== 'string' || !/^[0-9a-f]{64}$/.test(opts.reviewed)) {
|
||||
return { code: 2, message: '--reviewed <sha256> is required: take it with --hash before the reviewers read the plan' };
|
||||
}
|
||||
if (opts.reviewed !== current) {
|
||||
return { code: 2, message: 'the plan changed after the reviewers read it (--reviewed differs from the plan now); rerun the round on the current plan' };
|
||||
}
|
||||
const prev = extractVerdict(text);
|
||||
if (prev.error && !prev.placeholder) return { code: 2, message: `existing verdict unreadable: ${prev.error}` };
|
||||
const maxRounds = Math.min(prev.verdict?.max_rounds ?? opts.maxRounds ?? MAX_ROUNDS, MAX_ROUNDS);
|
||||
const prevExtra = prev.verdict?.extra_rounds ?? 0;
|
||||
const round = (prev.verdict?.round ?? 0) + 1;
|
||||
const capped = round > maxRounds + prevExtra;
|
||||
if (opts.extraRound && !capped) {
|
||||
return { code: 2, message: `--extra-round refused: round ${round} is within the cap; it is only for a round after the cap, after the operator's yes` };
|
||||
}
|
||||
if (capped && !opts.extraRound) {
|
||||
return {
|
||||
code: 1,
|
||||
message: `round ${round} refused: ${maxRounds} critic rounds${prevExtra ? ` (+${prevExtra} extra)` : ''} are spent. ` +
|
||||
'Stop and ask the operator: one more round (rerun with --extra-round after their yes), or stop here.',
|
||||
};
|
||||
}
|
||||
const extra = prevExtra + (capped ? 1 : 0);
|
||||
const findings = dedup.findings.map((f) => {
|
||||
const sev = normalizeSeverity(f.severity);
|
||||
const out = {
|
||||
id: f.id ?? null,
|
||||
severity: sev ?? 'blocker',
|
||||
rule_key: f.rule_key ?? null,
|
||||
file: f.file ?? null,
|
||||
line: f.line ?? null,
|
||||
text: typeof f.text === 'string' ? f.text : '',
|
||||
raised_by: Array.isArray(f.raised_by) ? f.raised_by : [],
|
||||
};
|
||||
if (sev === null) out.severity_defaulted = true;
|
||||
return out;
|
||||
});
|
||||
const counts = { blocker: 0, major: 0, minor: 0 };
|
||||
for (const f of findings) counts[f.severity] += 1;
|
||||
const verdict = {
|
||||
schema: VERDICT_SCHEMA,
|
||||
round,
|
||||
max_rounds: maxRounds,
|
||||
extra_rounds: extra,
|
||||
recorded_at: opts.now ?? new Date().toISOString(),
|
||||
plan_sha256: current,
|
||||
verdict: verdictFor(counts),
|
||||
counts,
|
||||
findings,
|
||||
};
|
||||
return { code: 0, text: placeSection(text, renderSection(verdict)), verdict, message: `round ${round} recorded: ${verdict.verdict}` };
|
||||
}
|
||||
|
||||
/**
|
||||
* K3 — decide whether the plan may go on.
|
||||
* @returns {{code: 0|1, status: 'pass'|'stop', reasons: object[], warnings: object[], verdict: object|null, validation: object}}
|
||||
*/
|
||||
export function gatePlan(rawText) {
|
||||
const text = norm(rawText);
|
||||
const reasons = [];
|
||||
const warnings = [];
|
||||
const validation = validatePlanContent(text, { strict: true });
|
||||
if (!validation.valid) {
|
||||
reasons.push({ code: 'PLAN_GATE_INVALID_PLAN', message: `plan-validator --strict: ${validation.errors.map(e => e.code).join(', ')}`, errors: validation.errors });
|
||||
}
|
||||
const version = extractPlanVersion(text);
|
||||
if (!versionAtLeast(version, '1.8')) {
|
||||
reasons.push({ code: 'PLAN_GATE_VERSION', message: `plan_version ${version ?? '(none)'}: the gate passes only plans written to plan_version 1.8 or later` });
|
||||
}
|
||||
const ex = extractVerdict(text);
|
||||
const v = ex.verdict;
|
||||
if (!ex.found) {
|
||||
reasons.push({ code: 'PLAN_GATE_NO_VERDICT', message: 'No "## Critic Verdict": the plan has not been through the critic (run Phase 9).' });
|
||||
} else if (ex.error) {
|
||||
reasons.push({ code: 'PLAN_GATE_VERDICT_MALFORMED', message: ex.error });
|
||||
} else {
|
||||
if (v.plan_sha256 !== planHash(text)) {
|
||||
reasons.push({ code: 'PLAN_GATE_VERDICT_STALE', message: `The plan changed after critic round ${v.round}; run the next round.` });
|
||||
}
|
||||
const sev = f => normalizeSeverity(f.severity);
|
||||
const blockers = v.findings.filter(f => sev(f) !== 'major' && sev(f) !== 'minor');
|
||||
const majors = v.findings.filter(f => sev(f) === 'major');
|
||||
if (blockers.length > 0) {
|
||||
reasons.push({ code: 'PLAN_GATE_OPEN_BLOCKER', message: `${blockers.length} open BLOCKER(s) in critic round ${v.round}: ${blockers.map(b => b.text).join(' | ')}` });
|
||||
if (v.round >= v.max_rounds + (v.extra_rounds ?? 0)) {
|
||||
reasons.push({ code: 'PLAN_GATE_ROUNDS_EXHAUSTED', message: `All ${v.round} critic rounds are spent with a BLOCKER open: stop and ask the operator.` });
|
||||
}
|
||||
}
|
||||
if (majors.length > 0) {
|
||||
warnings.push({ code: 'PLAN_GATE_OPEN_MAJOR', message: `${majors.length} MAJOR finding(s) left open in round ${v.round}; show them to the operator.` });
|
||||
}
|
||||
}
|
||||
return { code: reasons.length === 0 ? 0 : 1, status: reasons.length === 0 ? 'pass' : 'stop', reasons, warnings, verdict: v ?? null, validation };
|
||||
}
|
||||
|
||||
// ---- CLI --------------------------------------------------------------------
|
||||
|
||||
function isMain() {
|
||||
try {
|
||||
return realpathSync(process.argv[1]) === realpathSync(fileURLToPath(import.meta.url));
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
function out(obj) {
|
||||
process.stdout.write(JSON.stringify(obj) + '\n');
|
||||
}
|
||||
|
||||
if (isMain()) {
|
||||
const args = process.argv.slice(2);
|
||||
const usage = 'Usage: plan-gate.mjs <plan.md> [--json] | --hash <plan.md> | --record <plan.md> --reviewed <sha256> [--extra-round] < dedup.json';
|
||||
const known = new Set(['--json', '--hash', '--record', '--reviewed', '--extra-round']);
|
||||
const rIdx = args.indexOf('--reviewed');
|
||||
const reviewed = rIdx >= 0 ? args[rIdx + 1] : undefined;
|
||||
const positional = args.filter((a, i) => !a.startsWith('--') && !(rIdx >= 0 && i === rIdx + 1));
|
||||
const unknown = args.filter(a => a.startsWith('--') && !known.has(a));
|
||||
const planPath = positional[0];
|
||||
if (!planPath || positional.length > 1 || unknown.length > 0 || (rIdx >= 0 && (!reviewed || reviewed.startsWith('--')))) {
|
||||
process.stderr.write(`${usage}\n`);
|
||||
out({ code: 2, status: 'error', message: unknown.length ? `unknown argument: ${unknown.join(' ')}` : 'bad arguments' });
|
||||
process.exit(2);
|
||||
}
|
||||
let text;
|
||||
try {
|
||||
if (!existsSync(planPath)) throw new Error('file not found');
|
||||
text = readFileSync(planPath, 'utf-8');
|
||||
} catch (e) {
|
||||
out({ code: 2, status: 'error', plan: planPath, message: `cannot read plan: ${e.message}` });
|
||||
process.exit(2);
|
||||
}
|
||||
if (args.includes('--hash')) {
|
||||
process.stdout.write(`${planHash(text)}\n`);
|
||||
process.exit(0);
|
||||
}
|
||||
if (args.includes('--record')) {
|
||||
let dedup;
|
||||
try { dedup = JSON.parse(readFileSync(0, 'utf-8')); } catch (e) {
|
||||
out({ code: 2, recorded: false, message: `stdin is not JSON: ${e.message}` });
|
||||
process.exit(2);
|
||||
}
|
||||
const r = recordVerdict(text, dedup, { extraRound: args.includes('--extra-round'), reviewed });
|
||||
if (r.code === 0) {
|
||||
const tmp = `${planPath}.plan-gate.tmp`;
|
||||
try {
|
||||
writeFileSync(tmp, r.text);
|
||||
renameSync(tmp, planPath);
|
||||
} catch (e) {
|
||||
rmSync(tmp, { force: true });
|
||||
out({ code: 2, recorded: false, message: `cannot write the plan: ${e.message}` });
|
||||
process.exit(2);
|
||||
}
|
||||
}
|
||||
out({
|
||||
code: r.code, recorded: r.code === 0, message: r.message,
|
||||
round: r.verdict?.round ?? null, max_rounds: r.verdict?.max_rounds ?? null,
|
||||
verdict: r.verdict?.verdict ?? null, counts: r.verdict?.counts ?? null,
|
||||
});
|
||||
process.exit(r.code);
|
||||
}
|
||||
const g = gatePlan(text);
|
||||
if (args.includes('--json')) {
|
||||
process.stdout.write(JSON.stringify({ code: g.code, status: g.status, plan: planPath, reasons: g.reasons, warnings: g.warnings, verdict: g.verdict }, null, 2) + '\n');
|
||||
} else {
|
||||
process.stdout.write(
|
||||
`plan-gate: ${g.status.toUpperCase()} ${planPath}\n` +
|
||||
g.reasons.map(r => ` STOP [${r.code}] ${r.message}\n`).join('') +
|
||||
g.warnings.map(w => ` WARN [${w.code}] ${w.message}\n`).join(''),
|
||||
);
|
||||
}
|
||||
process.exit(g.code);
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue