/** * autoplan-shadow-reader — the read side of shadow mode (lib/autoplan-shadow.ts, * docs/unattended.md#shadow-mode). `exportShadow` is `gstack-autoplan shadow * --out | --runs `, the export gstack-evals reads for taste-replay * R2: one row per item of each answered gate revision, with the shadow's * status (recorded, malformed, failed with its code, missing, retrospective, or * off when the run never turned shadow on), the shadow fields when present, and * the owner's answer with its provenance (owner_mode, owner_named, carried, * answered_role, rec_hash at record and at answer). It computes no agreement * number, only counts. A revision whose plan approval is unanswered or answered * c (interrogate) refuses SHADOW_BEFORE_ANSWER (a courtesy for honest tools, * not access control: the journaled sha256 is the tamper evidence), and a * CEO-only run CEO_ONLY_NO_GATE. Read-only. `validateShadow` is the private * check `gstack-artifact validate ` adds: each record's schema and * its journaled sha256. */ import { createHash } from 'node:crypto'; import * as fs from 'node:fs'; import * as path from 'node:path'; import { previousGate } from './autoplan-gate'; import { readDecisions, readJournal, readState } from './autoplan-state'; import type { ResultCodeName } from './result-codes'; import type { ValidationError } from './headless-artifacts'; import { RECORD_RE, SHADOW_ACTIONS, recordFile, shadowDir } from './autoplan-shadow'; export type ShadowStatus = 'recorded' | 'malformed' | 'failed' | 'missing' | 'retrospective' | 'off'; export interface ShadowExportRow { run: string; gate_rev: number; item: string; decision: string; item_class: string | null; ceo_eligible: boolean; status: ShadowStatus; code?: string; shadow: Record | null; owner: { chosen: string | null; status: string; owner_mode: string | null; owner_named: boolean | null; carried: boolean; carried_from: string | null; answered_role: string | null; answered_by: string | null; rec_hash_at_answer: string | null; rec_hash_at_record: string | null }; } export interface ShadowExport { rows: ShadowExportRow[]; counts: Record>; refused: Array<{ run: string; gate_rev?: number; code: ResultCodeName; detail: string }> } const readRows = (file: string) => fs.readFileSync(file, 'utf8').split('\n').filter(Boolean).map(l => JSON.parse(l) as Record); /** Every answered revision of every run, deduped by run folder, gate_rev and item (two runs can share a run id). */ export function exportShadow(dirs: string[]): ShadowExport { const rows = new Map(); const refused: ShadowExport['refused'] = []; for (const dir of dirs) { let state; try { state = readState(dir); } catch (e: any) { refused.push({ run: dir, code: e.code ?? 'RUN_NOT_INITIALIZED', detail: `${dir}: ${e.detail ?? e.message}` }); continue; } if (state.ceo_only) { refused.push({ run: state.run, code: 'CEO_ONLY_NO_GATE', detail: `${dir} is CEO-only run ${state.run}: no gate, nothing answered` }); continue; } const key = fs.realpathSync(dir); const decisions = readDecisions(dir).filter(r => r.gate_rev >= 1); const journal = readJournal(dir); for (const rev of [...new Set(decisions.map(r => r.gate_rev))].sort((a, b) => a - b)) { const revRows = decisions.filter(r => r.gate_rev === rev); const p1 = revRows.find(r => (r.label ?? '').toLowerCase() === 'p1'); if (!p1?.chosen || p1.status === 'pending' || p1.chosen === 'c') { refused.push({ run: state.run, gate_rev: rev, code: 'SHADOW_BEFORE_ANSWER', detail: `${dir} gate_rev ${rev}: the plan approval is ${p1?.chosen === 'c' ? 'answered c (interrogate)' : 'unanswered'}; the export waits for the owner` }); continue; } const record = fs.existsSync(recordFile(dir, rev)) ? new Map(readRows(recordFile(dir, rev)).map(r => [String(r.item), r])) : undefined; const failed = journal.find(r => r.event === 'shadow_failed' && r.gate_rev === rev); const gate = previousGate(dir, rev); for (const d of revRows) { const item = (d.label ?? d.id).toLowerCase(); const s = record?.get(item); const status: ShadowStatus = s ? (s.retrospective ? 'retrospective' : s.status) : failed ? 'failed' : state.shadow?.enabled ? 'missing' : 'off'; const gateItem = gate?.items.find(i => i.id === item); const row: ShadowExportRow = { run: state.run, gate_rev: rev, item, decision: d.id, item_class: s?.item_class ?? gateItem?.item_class ?? null, ceo_eligible: s?.ceo_eligible ?? ((s?.item_class ?? gateItem?.item_class) !== 'finding' && (item === 'p1' || d.phase === 'ceo')), status, ...(status === 'failed' ? { code: String(failed!.code) } : {}), shadow: s ? Object.fromEntries(['action', 'pick', 'confidence', 'cites', 'reason', 'malformed', 'rule_pick', 'audit', 'model', 'prompt_version', 'prompt_sha256', 'rec_hash', 'rec_rev', 'recorded_at', 'retrospective'].filter(k => k in s).map(k => [k, s[k]])) : null, owner: { chosen: d.chosen ?? null, status: d.status, owner_mode: d.owner_mode ?? null, owner_named: d.owner_named ?? null, carried: !!(d.carried_from || d.carried_gate_from), carried_from: d.carried_from ?? d.carried_gate_from ?? null, answered_role: d.answered_role ?? null, answered_by: d.answered_by ?? null, rec_hash_at_answer: d.rec_hash ?? null, rec_hash_at_record: s?.rec_hash ?? null }, }; rows.set(`${key}|${rev}|${item}`, row); } } } const all = [...rows.values()]; const count = (key: (r: ShadowExportRow) => string) => all.reduce>((c, r) => { const k = key(r); c[k] = (c[k] ?? 0) + 1; return c; }, {}); return { rows: all, refused, counts: { status: count(r => r.status), item_class: count(r => r.item_class ?? 'unknown'), owner_mode: count(r => r.owner.owner_mode ?? 'none'), ceo_eligible: count(r => String(r.ceo_eligible)) } }; } export function renderShadowExport(r: ShadowExport): string[] { const counts = Object.entries(r.counts).map(([k, c]) => `SHADOW_COUNTS: ${k} ${Object.entries(c).map(([v, n]) => `${v}=${n}`).join(' ') || 'none'}`); return [...r.rows.map(row => `SHADOW_ROW: ${JSON.stringify(row)}`), ...counts, `SHADOW_EXPORT: rows=${r.rows.length} refused=${r.refused.length} (no agreement number: gstack-evals computes it)`]; } /** The private check behind gstack-artifact validate: each shadow record matches its journaled sha256 and its rows the schema. */ export function validateShadow(runDir: string): ValidationError[] { const dir = shadowDir(runDir); const journal = readJournal(runDir); const missing = (present: Set): ValidationError[] => [...new Set(journal.filter(r => r.event === 'shadow_recorded').map(r => Number(r.gate_rev)))] .filter(rev => !present.has(rev)) .map(rev => ({ code: 'ARTIFACT_STALE', path: `shadow/${path.basename(recordFile(runDir, rev))}`, message: `the journal records gate_rev ${rev}, but its shadow record is missing` })); let st: fs.Stats; try { st = fs.lstatSync(dir); } catch { return missing(new Set()); } if (!st.isDirectory()) return [{ code: 'ARTIFACT_PATH_ESCAPE', path: 'shadow', message: 'the shadow folder is a symlink or not a directory' }]; const errors: ValidationError[] = []; const present = new Set(); for (const name of fs.readdirSync(dir)) { const m = RECORD_RE.exec(name); if (!m) continue; const rev = Number(m[1]); present.add(rev); const rel = `shadow/${name}`; const bytes = fs.readFileSync(path.join(dir, name)); const recorded = journal.filter(r => r.event === 'shadow_recorded' && r.gate_rev === rev).at(-1); const sha = createHash('sha256').update(bytes).digest('hex'); if (!recorded) errors.push({ code: 'ARTIFACT_STALE', path: rel, message: `no shadow_recorded journal row names gate_rev ${rev}` }); else if (recorded.sha256 !== sha) errors.push({ code: 'ARTIFACT_STALE', path: rel, message: `sha256 ${sha.slice(0, 12)} is not the journaled ${String(recorded.sha256).slice(0, 12)}` }); bytes.toString('utf8').split('\n').filter(Boolean).forEach((line, i) => { let row: Record; try { row = JSON.parse(line); } catch { errors.push({ code: 'ARTIFACT_SCHEMA', path: `${rel}:${i + 1}`, message: 'not JSON' }); return; } const problem = rowProblem(row, rev); if (problem) errors.push({ code: 'ARTIFACT_SCHEMA', path: `${rel}:${i + 1}`, message: problem }); }); } return [...errors, ...missing(present)]; } function rowProblem(r: Record, rev: number): string | undefined { if (r.schema_version !== 1 || r.gate_rev !== rev) return `schema_version 1 and gate_rev ${rev} expected`; for (const k of ['run', 'decision', 'item', 'model', 'prompt_sha256', 'recorded_at']) if (typeof r[k] !== 'string' || !r[k]) return `${k} must be a non-empty string`; if (typeof r.retrospective !== 'boolean' || typeof r.audit !== 'boolean' || typeof r.ceo_eligible !== 'boolean') return 'retrospective, audit and ceo_eligible must be booleans'; if (r.status === 'malformed') return r.pick === null && !r.audit ? undefined : 'a malformed row carries no pick and no audit mark'; if (r.status !== 'recorded') return `status ${JSON.stringify(r.status)} is not recorded or malformed`; if (!SHADOW_ACTIONS.includes(r.action)) return 'action must be decide, send_back or hold'; if (typeof r.pick !== 'string' || !Array.isArray(r.item_frozen?.options) || !r.item_frozen.options.some((o: { key?: string }) => o.key === r.pick)) return 'pick must be one of the frozen item\'s options'; if (typeof r.confidence !== 'number' || !(r.confidence >= 0 && r.confidence <= 1)) return 'confidence must be a number from 0 to 1'; if (!Array.isArray(r.cites) || typeof r.reason !== 'string' || r.reason.length > 400) return 'cites must be a list and reason a string of at most 400 characters'; if (r.audit && r.action !== 'decide') return 'only a decide row carries the audit mark'; return undefined; }