diff --git a/bin/gstack-approvals b/bin/gstack-approvals index da97642a9..6d8006b64 100755 --- a/bin/gstack-approvals +++ b/bin/gstack-approvals @@ -5,7 +5,7 @@ * gstack-approvals add --ask --words --by --scope * [--pending ] [--outcome approved|rejected|pending|standing] * [--date YYYY-MM-DD] [--source ] [--branch ] [--json] - * gstack-approvals import --autoplan [--by ] [--branch ] [--json] + * gstack-approvals import --autoplan [--by ] [--branch | --stack ] [--json] * gstack-approvals render [--no-heading] [--branch ] * gstack-approvals list [--json] [--branch ] * gstack-approvals sync [--branch ] @@ -20,14 +20,15 @@ */ import * as path from 'node:path'; import { EXIT, exitTableHelp } from '../lib/headless-artifacts'; -import { appendRows, buildRow, importAutoplan, locate, readRecord, renderRecord, syncRecord, type RecordError, type StoreLocation } from '../lib/approval-record'; +import { appendRows, branchExists, buildRow, importAutoplan, locate, readRecord, renderRecord, syncRecord, type RecordError, type StoreLocation } from '../lib/approval-record'; import { resultLine } from '../lib/result-codes'; const HELP = `gstack-approvals — the per-thread approval record (capy/skill/gstack/captain.md) add --ask --words --by --scope [--pending ] [--outcome approved|rejected|pending|standing] [--date YYYY-MM-DD] [--source ] - import --autoplan [--by ] + import --autoplan [--by ] [--stack ] + --stack writes the same rows to every branch of a stack; each branch is checked first, and one unknown branch writes none render [--no-heading] the ## Approvals section for a PR body or report list [--json] sync re-mirror the record (GSTACK_APPROVALS_DIR, the Captain's drive approvals/, or the project-user Drive) @@ -40,7 +41,7 @@ Example: ${exitTableHelp()}`; -const VALUE_FLAGS = new Set(['--ask', '--words', '--by', '--scope', '--pending', '--outcome', '--date', '--source', '--branch', '--autoplan']); +const VALUE_FLAGS = new Set(['--ask', '--words', '--by', '--scope', '--pending', '--outcome', '--date', '--source', '--branch', '--autoplan', '--stack']); function parse(args: string[]): { flags: Record; bools: Set; error?: string } { const flags: Record = {}; @@ -64,11 +65,41 @@ function fail(cmd: string, e: RecordError): number { return EXIT.refused; } +/** `import --stack`: every branch checked first, then the same rows (ids are branch-independent) written to each branch's record. */ +function importStack(cmd: string, flags: Record, json: boolean): number { + if (cmd !== 'import' || !flags['--autoplan'] || flags['--branch'] !== undefined) { process.stderr.write(`gstack-approvals: --stack goes with import --autoplan , without --branch\n`); return EXIT.usage; } + const branches = [...new Set(flags['--stack']!.split(',').map(b => b.trim()).filter(Boolean))]; + if (!branches.length) { process.stderr.write(`gstack-approvals: --stack needs \n`); return EXIT.usage; } + const cwd = process.cwd(); + const missing = branches.filter(b => !branchExists(cwd, b)); + if (missing.length) return fail(cmd, { code: 'APPROVAL_RECORD_NO_BRANCH', message: `no local or origin branch ${missing.join(', ')}; nothing was written` }); + const runDir = path.resolve(flags['--autoplan']); + const results: Array<{ branch: string; added: string[]; skipped: number; file: string; mirror: string | null; error?: string }> = []; + let firstError: RecordError | undefined; + for (const branch of branches) { + const loc = locate({ cwd, env: process.env, branch }); + if ('code' in loc) return fail(cmd, loc); + const rows = importAutoplan(runDir, branch, flags['--by']); + if ('code' in rows) return fail(cmd, rows); + const res = appendRows(loc, rows); + if ('code' in res) { firstError ??= res; results.push({ branch, added: [], skipped: 0, file: loc.file, mirror: loc.mirror ?? null, error: res.message }); continue; } + if (res.mirrorError) firstError ??= res.mirrorError; + results.push({ branch, added: res.added.map(r => r.id), skipped: res.skipped, file: loc.file, mirror: loc.mirror ?? null, ...(res.mirrorError ? { error: res.mirrorError.message } : {}) }); + } + if (json) process.stdout.write(JSON.stringify({ branches: results }) + '\n'); + else { + for (const r of results) process.stdout.write(`APPROVAL_IMPORTED: branch=${r.branch} n=${r.added.length} skipped=${r.skipped} ${r.file}${r.mirror ? ` mirror=${r.mirror}` : ''}${r.error ? ` error=${r.error}` : ''}\n`); + process.stdout.write(`APPROVAL_IMPORTED_STACK: branches=${results.length} added=${results.reduce((n, r) => n + r.added.length, 0)}\n`); + } + return firstError ? fail(cmd, firstError) : EXIT.ok; +} + function main(argv: string[]): number { const [cmd, ...rest] = argv; if (!cmd || cmd === '--help' || cmd === '-h') { process.stdout.write(HELP + '\n'); return cmd ? EXIT.ok : EXIT.usage; } const { flags, bools, error } = parse(rest); if (error) { process.stderr.write(`gstack-approvals: ${error}\n${HELP}\n`); return EXIT.usage; } + if (flags['--stack'] !== undefined) return importStack(cmd, flags, bools.has('--json')); const loc = locate({ cwd: process.cwd(), env: process.env, branch: flags['--branch'] }); if ('code' in loc) return fail(cmd, loc); const json = bools.has('--json'); diff --git a/bin/gstack-autoplan b/bin/gstack-autoplan index 0b0a1c45b..d6558b886 100755 --- a/bin/gstack-autoplan +++ b/bin/gstack-autoplan @@ -19,13 +19,14 @@ * gstack-autoplan release --out --attempt drop a reservation for an attempt that never ran * gstack-autoplan close --out --phase

--deferred "" --by > close a phase accepted-with-findings * gstack-autoplan budget --out

[--spend-cap ] [--max-rounds ] --by > the owner's journaled cap change + * gstack-autoplan check --out --attempt --result --model | --unavailable the delta check before a carry * gstack-autoplan checkpoints --from list checkpoint generations, newest first, with hash ok|bad * gstack-autoplan contract print the coordinator contract every lane prompt inherits (plan E2) * * Logic: lib/autoplan-run.ts (next/submit/resume), lib/autoplan-gate.ts (close, gate, answer), * lib/autoplan-export.ts, lib/autoplan-prompts.ts, lib/spend-ledger.ts. Codes: lib/result-codes.ts. * Workflow commands end with GSTACK_RESULT: skill=autoplan status= run=; the run is never approved here. - * Mutating verbs (next, submit, answer, recommend, revise, close, budget, export, run) pass admission first (lib/autoplan-admission.ts) + * Mutating verbs (next, submit, answer, recommend, revise, close, budget, check, export, run) pass admission first (lib/autoplan-admission.ts) * and bring an older run directory forward (lib/autoplan-migrate.ts) under the run lock. */ import * as fs from 'node:fs'; @@ -33,6 +34,7 @@ import * as path from 'node:path'; import { amendmentSteps, answerGate, closeDeferred, finishGate, recommendGate } from '../lib/autoplan-gate'; import { gateCost, gateCostPayload, renderGateCost, type GateCost } from '../lib/autoplan-gate-cost'; import { attemptEstimate, changeBudget, escalateLine, liftStop } from '../lib/autoplan-budget'; +import { carryLines, submitCarryCheck } from '../lib/autoplan-carry'; import { ANSWERED_BY_RE } from '../lib/gate-list'; import { checkpointLines, checkpointRun, restoreCheckpoint, type CheckpointMoment } from '../lib/autoplan-checkpoint'; import { admit, touchInstallLease } from '../lib/autoplan-admission'; @@ -60,6 +62,9 @@ DRIVER: ${driverFile()} --estimate-usd: reserved per new attempt; unattended Capy runs reserve a default for host attempts so --spend-cap binds --scaffold (first next): a plan missing one of its two sections is reviewed from /plan-source.md with the heading added --light loads each phase's checklist plus the sections its scope needs (the unattended default); interactive kinds load everything + --no-carry-forward (first next): every later gate revision goes to the owner, even one that asks nothing new + --carry-check-cap (first next): the most one delta check may reserve before a carry (default 5.00) + the first next prints REVIEW_ESTIMATE: the most the review can reserve at this scope and round cap prepare the current phase; one ATTEMPT: line per voice; ends awaiting_result submit --out --phase

--voice native|outside --result | --record [--attest ] --model [--attempt ] [--runner ] [--usage-usd ] [--reconcile ] [--json] @@ -84,6 +89,9 @@ DRIVER: ${driverFile()} resume --out

[--from ] [--json] --from restores a hash-verified checkpoint into an empty --out first (a collection or root restores its newest valid generation; GSTACK_AUTOPLAN_CHECKPOINT_DIR set: every command leaving the run at rest publishes a generation under // and prints CHECKPOINT: on stderr) + check --out --attempt --result --model [--same-family ""] | --unavailable "" + bind the one-reviewer delta check a CARRY_CHECK: line asked for: nothing at High or above carries the owner's + last answer (CARRIED:); otherwise, or when unavailable, the gate goes to the owner as before checkpoints --from list checkpoint generations, newest first: run, gate_rev, time, hash ok|bad, current close --out --phase

--deferred "" --by > close the phase accepted-with-findings: open voices skipped (dispatched attempts settle unknown at their @@ -180,7 +188,10 @@ function doNext(out: string, rest: string[], json: boolean): number { const ui = rest.includes('--ui') ? true : rest.includes('--no-ui') ? false : undefined; const maxRounds = num(rest, '--max-rounds'); if (maxRounds !== undefined && (!Number.isInteger(maxRounds) || maxRounds < 1)) throw new RunError('ARTIFACT_SCHEMA', '--max-rounds needs an integer >= 1', 1); + const carryCheckCapUsd = num(rest, '--carry-check-cap'); + if (carryCheckCapUsd !== undefined && !(carryCheckCapUsd >= 0)) throw new RunError('ARTIFACT_SCHEMA', '--carry-check-cap needs a number >= 0', 1); initRun({ plan, out, cwd, planHeading: flag(rest, '--plan-heading'), outsideEnabled: !rest.includes('--no-outside'), spendCap: num(rest, '--spend-cap') ?? null, maxRounds, + carryForward: !rest.includes('--no-carry-forward'), carryCheckCapUsd, deadlineMin: num(rest, '--deadline'), light: rest.includes('--light'), scaffold: rest.includes('--scaffold'), scope: { ui, developerTool: rest.includes('--developer-tool'), agentPrimary: rest.includes('--agent-primary'), dx: rest.includes('--no-dx') ? false : undefined, ...(rest.includes('--docs-only') ? { docsOnly: true } : {}) } }, initLines); } const redispatch = rest.flatMap((a, i) => (a === '--redispatch' && rest[i + 1] ? [rest[i + 1]!] : [])); @@ -257,7 +268,7 @@ function doClose(out: string, rest: string[], json: boolean): number { const r = closeDeferred(state, phase as Phase, { reason: reason.trim(), by, cause: 'owner' }); moment = 'transition'; const lines = [...r.lines, ...(stopped ? liftStop(state, `${phase} closed deferred by ${by}`) : [])]; - if (!currentPhase(state)) { finishGate(state); lines.push(...gateLines(state, out)); } + if (!currentPhase(state)) { finishGate(state); lines.push(...carryLines(state), ...(state.status === 'gate_pending' ? gateLines(state, out) : [])); } else { state.status = 'running'; lines.push(`NEXT: gstack-autoplan next --out ${out}`); } writeState(state); lines.push(resultLineFor(state.status, out)); @@ -265,6 +276,19 @@ function doClose(out: string, rest: string[], json: boolean): number { return EXIT.ok; } +/** `check`: the delta reviewer's verdict before a carry (lib/autoplan-carry.ts). */ +function doCheck(out: string, rest: string[], json: boolean): number { + const unavailable = flag(rest, '--unavailable'); + const result = flag(rest, '--result'); + if (unavailable === undefined && (!result || !flag(rest, '--model'))) return usage('check needs --attempt and --result --model , or --unavailable ""'); + const r = submitCarryCheck({ out, attempt: flag(rest, '--attempt'), result: result ? path.resolve(result) : undefined, model: flag(rest, '--model'), unavailable, sameFamily: flag(rest, '--same-family') }); + moment = 'transition'; + const status = r.state.status; + const lines = [...r.lines, ...(status === 'gate_pending' ? gateLines(r.state, out) : []), ...(status === 'amend_pending' || status === 'running' ? [nextLine(r.state, out)] : []), resultLineFor(status, out)]; + emit(json, lines, { run: r.state.run, status, gate_rev: r.state.gate_rev ?? null, carry: r.state.carry ?? null, out }); + return EXIT.ok; +} + function doBudget(out: string, rest: string[], json: boolean): number { const state = readState(out); const lines = changeBudget(state, { spendCap: num(rest, '--spend-cap'), maxRounds: num(rest, '--max-rounds'), by: flag(rest, '--by') }); @@ -338,6 +362,7 @@ async function main(argv: string[]): Promise { if (cmd === 'revise') return mutating(out, json, () => doRevise(out, rest, json)); if (cmd === 'close') return mutating(out, json, () => doClose(out, rest, json)); if (cmd === 'budget') return mutating(out, json, () => doBudget(out, rest, json)); + if (cmd === 'check') return mutating(out, json, () => doCheck(out, rest, json)); if (cmd === 'run') { const admission = admit(process.env); if (admission.lines.length) process.stdout.write(admission.lines.join('\n') + '\n'); return await doRun(out, rest, json); } if (cmd === 'status') { const state = readState(out); @@ -416,7 +441,7 @@ async function main(argv: string[]): Promise { if (e instanceof RunError) { process.stderr.write(resultLine(`gstack-autoplan ${cmd}`, e.code, e.detail) + '\n'); if (e.after.lines?.length) process.stdout.write(e.after.lines.join('\n') + '\n'); - if (['next', 'submit', 'answer', 'resume', 'run', 'revise', 'recommend', 'close', 'budget'].includes(cmd)) process.stdout.write(resultLineFor(e.after.status ?? 'refused', out) + '\n'); + if (['next', 'submit', 'answer', 'resume', 'run', 'revise', 'recommend', 'close', 'budget', 'check'].includes(cmd)) process.stdout.write(resultLineFor(e.after.status ?? 'refused', out) + '\n'); return e.exit === 3 ? EXIT.refused : EXIT.fail; } process.stderr.write(`gstack-autoplan ${cmd}: ${e?.message ?? e}\n`); diff --git a/lib/approval-record.ts b/lib/approval-record.ts index ed8bcf738..38d798682 100644 --- a/lib/approval-record.ts +++ b/lib/approval-record.ts @@ -52,6 +52,11 @@ export type Env = NodeJS.ProcessEnv; const git = (cwd: string, args: string[]) => spawnSync('git', args, { cwd, encoding: 'utf8', timeout: 10_000 }); +/** A branch this checkout knows: a local head or an origin remote-tracking ref. */ +export function branchExists(cwd: string, branch: string): boolean { + return [`refs/heads/${branch}`, `refs/remotes/origin/${branch}`].some(ref => git(cwd, ['rev-parse', '--verify', '--quiet', ref]).status === 0); +} + /** One contained path segment for a branch: readable prefix plus a hash of the exact ref. */ export function branchKey(branch: string): string { const readable = branch.replace(/[^A-Za-z0-9._-]+/g, '-').replace(/^[.-]+/, '').slice(0, 60) || 'branch'; @@ -226,13 +231,15 @@ export function renderRecord(read: ReadResult, loc: StoreLocation, o: { heading? return [...head, ...lines, '', `Source: gstack-approvals record for ${cell(loc.branch)} (display only; not an approval citation).`, ''].join('\n'); } -interface DecisionLike { id?: string; title?: string; status?: string; chosen?: string | null; recommended?: string | null; gate_rev?: number; answered_at?: string; answered_by?: string; reply?: string; owner_reply?: string; carried_from?: string; label?: string; chosen_text?: string } +interface DecisionLike { id?: string; title?: string; status?: string; chosen?: string | null; recommended?: string | null; gate_rev?: number; answered_at?: string; answered_by?: string; reply?: string; owner_reply?: string; carried_from?: string; label?: string; chosen_text?: string; + carried_gate_from?: string; carried_gate_rev?: number; carried_answered_by?: string; carried_under?: string; carry_check?: string } const stripRef = (t: string) => t.replace(/\s*\[[^\]]*\]\s*$/, '').trim(); /** * Rows for every answered gate in an autoplan run: one row per answer event (gate_rev, time, answerer), covering - * the items it settled. Carried-forward answers were recorded when first given and are skipped. Ids are stable, + * the items it settled. Item answers carried from an earlier revision were recorded when first given and are skipped; + * a whole gate carried forward (lib/autoplan-carry.ts) gets its own row naming the revision its words came from. Ids are stable, * so a rerun (or a retry after an interrupted import) adds nothing twice. */ export function importAutoplan(runDir: string, branch: string, by?: string): ApprovalRow[] | RecordError { @@ -265,10 +272,16 @@ export function importAutoplan(runDir: string, branch: string, by?: string): App const changed = items.filter(r => r.chosen && r.recommended && r.chosen !== r.recommended); const stillPending = rows.filter(r => r.gate_rev === rev && r.status === 'pending'); const answerer = first.answered_by ?? 'unknown'; - const who = answerer === 'human' ? (by ? `${by}, typed at the gate` : 'the owner, typed at the gate') : `${by ?? 'the owner'}, relayed by ${answerer}`; - const scope = `the reviewed plan (sha256 ${planSha}), ${changed.length ? `${changed.length} recommendation(s) changed: ${changed.map(r => stripRef(r.title ?? '')).join('; ')}` : 'every recommendation taken'}`; + const carried = p1?.carried_gate_from ? p1 : undefined; + const relayed = carried?.carried_answered_by && carried.carried_answered_by !== 'human' ? `, relayed by ${carried.carried_answered_by}` : ''; + const who = carried ? `${by ?? 'the owner'} (words from gate_rev ${carried.carried_gate_rev}${relayed}), carried forward by gstack-autoplan` + : answerer === 'human' ? (by ? `${by}, typed at the gate` : 'the owner, typed at the gate') : `${by ?? 'the owner'}, relayed by ${answerer}`; + const unseen = items.filter(r => r.carried_under).map(r => stripRef(r.title ?? '')); + const checked = [...new Set(items.map(r => r.carry_check).filter((c): c is string => !!c))]; + const scope = [`the reviewed plan (sha256 ${planSha}), ${changed.length ? `${changed.length} recommendation(s) changed: ${changed.map(r => stripRef(r.title ?? '')).join('; ')}` : 'every recommendation taken'}`, + unseen.length ? `taken under "do it", not seen by the owner: ${unseen.join('; ')}` : '', checked.length ? `post-cap delta checked by ${checked.join(', ')}` : ''].filter(Boolean).join('; '); const built = buildRow({ - ask: `Approve the plan "${planTitle}" (autoplan run ${runId}, ${items.length + (p1 ? 1 : 0)} decisions)`.slice(0, CAPS.ask), + ask: `Approve the plan "${planTitle}" (autoplan run ${runId}, ${carried ? `gate_rev ${rev} carried forward from gate_rev ${carried.carried_gate_rev}, ` : ''}${items.length + (p1 ? 1 : 0)} decisions)`.slice(0, CAPS.ask), words: first.owner_reply ?? first.reply ?? '(no words recorded)', by: who.slice(0, CAPS.by), scope: scope.slice(0, CAPS.scope), diff --git a/lib/autoplan-carry.ts b/lib/autoplan-carry.ts new file mode 100644 index 000000000..061b3036d --- /dev/null +++ b/lib/autoplan-carry.ts @@ -0,0 +1,223 @@ +/** + * autoplan-carry — carry the owner's last answer to a later gate revision (GSTA-155, + * docs/unattended.md#carry-forward). When a gate_rev N+1 asks nothing the owner has not + * already decided, the runner answers it with the owner's recorded gate_rev N words instead + * of a second round trip, and records that it did (decisions.jsonl `carried_gate_from`, + * journal `gate_carried`, `/decisions.tsv`). It never invents an answer: every + * carried answer is the owner's own earlier words, and every refusal says which condition failed. + * + * A carry needs all of: carry not turned off (`--no-carry-forward`); gate_rev N approved (p1 a or b) + * with the owner's words recorded, and not itself a carry; every change accepted at N + * incorporated or settled; the plan file unchanged since the gate or the revise that took those + * changes, and the reviewed plan equal to it; no open urgent finding; and every new item already + * carried. Two kinds of new item do not block: items a "do it" covers (the owner's words are + * "do it": an auto item that changes no scope, cost or direction and is no round grant), and a + * capped-delta item recommending "accept", which first gets one reviewer on just the delta + * (`CARRY_CHECK:`, answered with `gstack-autoplan check`); a High or Critical finding there + * means no carry, and the gate goes to the owner as before. + */ +import { createHash } from 'node:crypto'; +import * as fs from 'node:fs'; +import * as path from 'node:path'; +import { attemptEstimate } from './autoplan-budget'; +import { lineDiff } from './autoplan-confirm'; +import { answerGate } from './autoplan-gate'; +import { driftLine } from './autoplan-plan-text'; +import { BOUNDARY, RESULT_FORMAT, REVIEWER_TRAILER } from './autoplan-prompts'; +import { parseReviewResult } from './autoplan-reconcile'; +import { FILES, RunError, journal, now, readDecisions, readJournal, readState, writeDecisions, writeState, type RunState } from './autoplan-state'; +import { appendTrailRow } from './decision-trail'; +import { decisionId, type GateItem, type GateList } from './gate-list'; +import { readJsonl } from './headless-artifacts'; +import { modelFamily } from './outside-voice-runner'; +import { release, reserve, settle } from './spend-ledger'; + +export const DEFAULT_CARRY_CHECK_CAP_USD = 5; +const HIGH = ['Critical', 'High']; +const DO_IT = /^\s*(?:(?!captain\b)[a-z][\w .'-]{0,40}:\s*)?do it[\s.!]*$/i; +/** The owner's verbatim words are a "do it" (optionally after their name), not a Captain's words naming one. */ +export const isDoIt = (words: string | undefined): boolean => !!words && DO_IT.test(words); + +export interface CarryCheck { + attempt: string; prompt: string; result: string; delta_sha256: string; items: string[]; estimate_usd: number; + status: 'dispatched' | 'passed' | 'failed' | 'unavailable'; model?: string; max_severity?: string; findings?: number; at: string; +} +export interface CarryRecord { + gate_rev: number; from_gate_rev?: number; verdict: 'carried' | 'refused' | 'check_pending'; + reason: string; words?: string; under_do_it?: string[]; check?: CarryCheck; lines: string[]; at: string; +} + +const sha256 = (text: string | Buffer) => createHash('sha256').update(text).digest('hex'); +const planFile = (state: RunState) => state.plan_input ?? state.plan_source; +const fileSha = (file: string) => (fs.existsSync(file) ? sha256(fs.readFileSync(file)) : undefined); +export const atGateCopy = (out: string, rev: number) => path.join(out, `plan-at-gate.r${rev}.md`); + +/** What finishGate journals beside gate_written: the plan file's bytes and a copy of the reviewed plan the owner is asked about. */ +export function gateEvidence(state: RunState, rev: number): { source_file_sha256?: string } { + if (fs.existsSync(state.active_plan)) fs.copyFileSync(state.active_plan, atGateCopy(state.out, rev)); + return { source_file_sha256: fileSha(planFile(state)) }; +} + +/** The lines the latest carry decision printed, for the command that wrote this gate revision. */ +export function carryLines(state: RunState): string[] { + const c = state.carry; + return c && c.gate_rev === state.gate_rev ? c.lines : []; +} + +function record(state: RunState, c: Omit): CarryRecord { + const full = { ...c, at: now() }; + state.carry = full; + journal(state.out, { event: c.verdict === 'carried' ? 'gate_carried' : c.verdict === 'check_pending' ? 'carry_check_dispatched' : 'carry_refused', gate_rev: c.gate_rev, from_gate_rev: c.from_gate_rev ?? null, reason: c.reason, ...(c.under_do_it?.length ? { under_do_it: c.under_do_it } : {}), ...(c.check ? { check: c.check.attempt, check_status: c.check.status } : {}) }); + appendTrailRow(path.join(state.out, 'decisions.tsv'), ['gate', c.verdict === 'carried' ? `carried gate_rev ${c.gate_rev} from gate_rev ${c.from_gate_rev}` : c.verdict === 'check_pending' ? `delta check before carrying gate_rev ${c.gate_rev}` : `no carry for gate_rev ${c.gate_rev}`, + c.reason, c.check ? path.basename(c.check.result) : 'decisions.jsonl', c.check?.status ?? c.verdict]); + return full; +} +const refuse = (state: RunState, rev: number, from: number | undefined, reason: string) => record(state, { gate_rev: rev, from_gate_rev: from, verdict: 'refused', reason, lines: [`CARRY: no (${reason})`] }); + +interface Source { id: string; gate_rev: number; words: string; by?: string } +type Plan = { source: Source; doIt: GateItem[]; deltas: GateItem[] } | { reason: string; from?: number }; + +/** Every condition but the delta check; returns why not, or what the carry takes. */ +function assess(state: RunState, prev: GateList, list: GateList): Plan { + const from = prev.gate_rev; + if (state.carry_forward === false) return { reason: 'carry-forward is off for this run (--no-carry-forward)', from }; + const p1 = readDecisions(state.out).find(r => r.gate_rev === from && (r.label ?? '').toLowerCase() === 'p1'); + if (!p1 || p1.status === 'pending' || !p1.answered_at) return { reason: `gate_rev ${from} was not answered`, from }; + if (p1.chosen !== 'a' && p1.chosen !== 'b') return { reason: `gate_rev ${from} answer was p1${p1.chosen}, not an approval`, from }; + if (p1.carried_gate_from) return { reason: `gate_rev ${from} was itself carried; the owner answers this one`, from }; + const words = p1.owner_reply ?? (p1.answered_by === 'human' ? p1.reply : undefined); + if (!words?.trim()) return { reason: `no owner words recorded at gate_rev ${from}`, from }; + const amendments = (state.amendments ?? []).filter(a => a.gate_rev === from); + const open = amendments.find(a => a.status !== 'incorporated' && a.status !== 'settled'); + if (open) return { reason: `accepted change ${open.decision} is ${open.status}`, from }; + const journalRows = readJournal(state.out); + const at = journalRows.map(r => r.event === 'gate_written' && r.gate_rev === from).lastIndexOf(true); + let expected: string | undefined = at >= 0 ? journalRows[at]!.source_file_sha256 : undefined; + for (const r of journalRows.slice(at + 1)) if (r.event === 'plan_revised') expected = r.source_file_sha256; + if (!expected) return { reason: 'no plan evidence recorded for the gate or the revise (a run written before carry-forward)', from }; + if (fileSha(planFile(state)) !== expected) return { reason: `the plan file ${path.basename(planFile(state))} changed after the owner's answer or the revise that took it`, from }; + if (driftLine(state, 'PLAN_SOURCE_CHANGED')) return { reason: 'the reviewed plan differs from the plan file', from }; + const urgent = readJsonl(path.join(state.out, 'findings.jsonl')).rows.find(r => (r.urgent === true || r.outside_plan === true) && !r.settled_by); + if (urgent) return { reason: `urgent finding open: ${String(urgent.title ?? urgent.id)}`, from }; + const doIt: GateItem[] = []; const deltas: GateItem[] = []; + for (const item of list.items) { + if (item.id === 'p1' || item.carried) continue; + const option = item.options.find(o => o.key === item.recommended); + if (item.cap?.kind === 'delta' && !option?.changes_plan && item.recommended === 'a') { deltas.push(item); continue; } + if (isDoIt(words) && item.kind === 'auto' && !item.cap && !['scope', 'cost', 'direction'].includes(item.changes ?? '')) { doIt.push(item); continue; } + return { reason: `new item ${item.id}: ${item.title}`, from }; + } + return { source: { id: p1.id, gate_rev: from, words, by: p1.answered_by }, doIt, deltas }; +} + +/** Called by finishGate once gate_rev N+1 is written (status gate_pending): carry, set up the delta check, or say why not. */ +export function evaluateCarry(state: RunState, prev: GateList | undefined, list: GateList): CarryRecord | undefined { + if (!prev || list.gate_rev < 2) return undefined; + try { + const plan = assess(state, prev, list); + if ('reason' in plan) return refuse(state, list.gate_rev, plan.from, plan.reason); + if (plan.deltas.length) return dispatchCheck(state, prev, list, plan); + return carry(state, list, plan.source, plan.doIt, plan.deltas); + } catch (e: any) { + return refuse(state, list.gate_rev, prev.gate_rev, `error: ${e instanceof RunError ? `${e.detail} (${e.code})` : e?.message ?? e}`); + } +} + +function dispatchCheck(state: RunState, prev: GateList, list: GateList, plan: Extract): CarryRecord { + const rev = list.gate_rev; + const cap = state.carry_check_cap_usd ?? DEFAULT_CARRY_CHECK_CAP_USD; + const estimate = attemptEstimate(state, undefined, process.env).usd; + if (estimate > cap) return refuse(state, rev, prev.gate_rev, `delta check estimate ${estimate.toFixed(2)} is over the per-gate cap ${cap.toFixed(2)} (--carry-check-cap)`); + const before = atGateCopy(state.out, prev.gate_rev); + if (!fs.existsSync(before)) return refuse(state, rev, prev.gate_rev, `no copy of the plan the owner answered at gate_rev ${prev.gate_rev}`); + const current = fs.readFileSync(state.active_plan, 'utf8'); + const diff = lineDiff(fs.readFileSync(before, 'utf8'), current).filter(l => l.op !== ' ').map(l => `${l.op} ${l.text}`).join('\n'); + const deltaSha = sha256(diff); + const attempt = `${state.run}-c${rev}`; + const prompt = path.join(state.out, `carry-check-g${rev}-prompt.md`); + const result = path.join(state.out, `carry-check-g${rev}.md`); + try { reserve(path.join(state.out, FILES.spend), attempt, estimate, { cap: state.spend_cap_usd, label: 'carry-check' }); } + catch (e: any) { return refuse(state, rev, prev.gate_rev, `delta check does not fit the run's spend cap: ${e?.message ?? e}`); } + const accepted = (state.amendments ?? []).filter(a => a.gate_rev === prev.gate_rev).map(a => `- ${a.title} -> ${a.text}`).join('\n') || '- (none recorded)'; + fs.writeFileSync(prompt, [BOUNDARY, '', + `You are checking one edit to a reviewed plan before gstack reuses the owner's approval of gate_rev ${prev.gate_rev} for gate_rev ${rev}. The owner accepted these changes:`, accepted, '', + 'The plan was edited after that answer, but the phases that would re-review the edit were at their round cap, so no reviewer has read it. Read the delta and say whether it implements the accepted changes, adds nothing the owner did not accept (scope, cost or direction), and leaves the plan correct. Rate a finding High or Critical only when the edit adds an unaccepted change, contradicts an accepted one, or would make the plan fail a test, break production, or lose data or secrets; anything else is Medium or lower.', '', + `Delta (reviewed plan at gate_rev ${prev.gate_rev} -> now; + added, - removed; SHA-256 ${deltaSha}):`, '```diff', diff, '```', '', + 'The complete current plan follows as context.', '', current, + REVIEWER_TRAILER, RESULT_FORMAT('carry-check', deltaSha, result)].join('\n')); + const check: CarryCheck = { attempt, prompt, result, delta_sha256: deltaSha, items: plan.deltas.map(i => i.id), estimate_usd: estimate, status: 'dispatched', at: now() }; + const family = state.native_model ? `not ${modelFamily(state.native_model)}` : 'differs from the plan reviewers where available'; + return record(state, { gate_rev: rev, from_gate_rev: prev.gate_rev, verdict: 'check_pending', words: plan.source.words, under_do_it: plan.doIt.map(i => i.id), check, + reason: `only an unreviewed post-cap delta is new (${check.items.join(',')}); one reviewer reads it first`, + lines: [`CARRY_CHECK: attempt=${attempt} items=${check.items.join(',')} prompt=${prompt} result=${result} model_family=${family} estimate=${estimate.toFixed(2)} cap=${cap.toFixed(2)}`, + `NEXT: dispatch one reviewer on the prompt, then gstack-autoplan check --out ${state.out} --attempt ${attempt} --result ${result} --model (or --unavailable "" to relay the gate as today)`] }); +} + +function carry(state: RunState, list: GateList, source: Source, doIt: GateItem[], deltas: GateItem[], check?: CarryCheck): CarryRecord { + const rev = list.gate_rev; + const p1 = list.items.find(i => i.id === 'p1')!; + const reply = list.items.map(i => `${i.id}${i.id === 'p1' ? p1.recommended : i.carried?.chosen ?? i.recommended}`).join(' '); + const reason = [doIt.length || deltas.length ? '' : 'no new choice', doIt.length ? `"do it" covers ${doIt.map(i => i.id).join(',')}` : '', deltas.length ? `delta check passed on ${deltas.map(i => i.id).join(',')}` : '', + 'plan unchanged since the accepted changes'].filter(Boolean).join('; '); + const answered = answerGate(state, rev, reply, 'agent:gstack-autoplan', { carry: { from: source.id, gate_rev: source.gate_rev, words: source.words, by: source.by, reason } }); + const marks = new Map>([...doIt.map(i => [decisionId(list.run, rev, i.id), { carried_under: 'do it' }] as const), ...deltas.map(i => [decisionId(list.run, rev, i.id), { carry_check: check?.attempt ?? '' }] as const)]); + if (marks.size) writeDecisions(state.out, readDecisions(state.out).map(r => (marks.has(r.id) ? { ...r, ...marks.get(r.id) } : r))); + const lines = [`CARRIED: gate_rev=${rev} <- gate_rev=${source.gate_rev} words=${JSON.stringify(source.words)} by=${source.by ?? 'unknown'} reason=${reason}`, + ...doIt.map(i => `CARRIED_UNDER_DO_IT: ${i.id} ${i.title} -> ${i.options.find(o => o.key === i.recommended)?.text ?? i.recommended} (the owner has not seen this item)`), + ...answered.lines]; + return record(state, { gate_rev: rev, from_gate_rev: source.gate_rev, verdict: 'carried', reason, words: source.words, under_do_it: doIt.map(i => i.id), ...(check ? { check } : {}), lines }); +} + +export interface CheckOptions { out: string; attempt?: string; result?: string; model?: string; unavailable?: string; sameFamily?: string } + +/** `gstack-autoplan check`: bind the delta reviewer's result (or its absence) and carry, or leave the gate for the owner. */ +export function submitCarryCheck(o: CheckOptions): { state: RunState; lines: string[] } { + const state = readState(path.resolve(o.out)); + const c = state.carry; + if (!c || c.verdict !== 'check_pending' || !c.check || state.status !== 'gate_pending' || c.gate_rev !== state.gate_rev) throw new RunError('PHASE_NOT_AWAITING', `no delta check is pending (run is ${state.status}, gate_rev=${state.gate_rev ?? 'none'})`, 3); + const check = c.check; + if (o.attempt !== check.attempt) throw new RunError('ATTEMPT_MISMATCH', `pass --attempt ${check.attempt}`, 3); + const gateFile = path.join(state.out, 'gate.json'); + const list = JSON.parse(fs.readFileSync(gateFile, 'utf8')) as GateList; + const spend = path.join(state.out, FILES.spend); + let out: CarryRecord; + if (o.unavailable !== undefined) { + release(spend, check.attempt); + list.carry_check = { attempt: check.attempt, model: 'none', verdict: 'unavailable', max_severity: 'none', findings: 0, at: now() }; + fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n'); + out = refuse(state, c.gate_rev, c.from_gate_rev, `delta check unavailable: ${o.unavailable}`); + out.check = { ...check, status: 'unavailable' }; + } else { + if (!o.result || !o.model) throw new RunError('RESULT_RECEIPT_MISSING', 'check needs --result and --model , or --unavailable ""', 3); + if (!fs.existsSync(o.result)) throw new RunError('RESULT_RECEIPT_MISSING', `result file not found: ${o.result}`); + const family = modelFamily(o.model); + const author = state.native_model ? modelFamily(state.native_model) : undefined; + if (!o.sameFamily && (family === 'unknown' || family === author)) throw new RunError('MODEL_FAMILY_CONFLICT', `${o.model} is ${family === 'unknown' ? 'of no recognized family' : `the plan reviewers' family (${family})`}; use another family, or pass --same-family ""`, 3); + const text = fs.readFileSync(o.result, 'utf8'); + const parsed = parseReviewResult(text, { phase: 'carry-check', sha256: check.delta_sha256 }); + if (!parsed.receipt.ok) throw new RunError('RESULT_RECEIPT_MISSING', parsed.errors[0]!); + if (!parsed.fenceFound || parsed.errors.length) throw new RunError('RESULT_FINDINGS_MISSING', parsed.errors.join('; ')); + fs.writeFileSync(path.join(state.out, `carry-check-g${c.gate_rev}.bound.md`), text, process.platform === 'win32' ? {} : { mode: 0o444 }); + settle(spend, check.attempt, 'unknown'); + const order = ['Informational', 'Low', 'Medium', 'High', 'Critical']; + const max = parsed.findings.map(f => String(f.severity)).reduce((a, b) => (order.indexOf(b) > order.indexOf(a) ? b : a), 'none'); + const blocking = parsed.findings.find(f => HIGH.includes(String(f.severity))); + const done: CarryCheck = { ...check, status: blocking ? 'failed' : 'passed', model: o.model, max_severity: max, findings: parsed.findings.length, at: now() }; + list.carry_check = { attempt: check.attempt, model: o.model + (o.sameFamily ? ` (same family: ${o.sameFamily})` : ''), verdict: blocking ? 'failed' : 'passed', max_severity: max, findings: parsed.findings.length, at: done.at }; + if (blocking) for (const item of list.items) if (check.items.includes(item.id)) (item.flags ??= []).push(`carry check ${o.model}: ${blocking.severity} ${blocking.title}`); + fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n'); + if (blocking) { + out = refuse(state, c.gate_rev, c.from_gate_rev, `delta check found ${blocking.severity}: ${blocking.title}`); + out.check = done; + } else { + const plan = assess(state, { ...list, gate_rev: c.from_gate_rev! } as GateList, list); + if ('reason' in plan) { out = refuse(state, c.gate_rev, c.from_gate_rev, plan.reason); out.check = done; } + else { + out = carry(state, list, plan.source, plan.doIt, plan.deltas, done); + out.lines.unshift(`CARRY_CHECK: passed attempt=${check.attempt} model=${o.model} findings=${done.findings} highest=${max}`); + } + } + } + writeState(state); + return { state, lines: out.lines }; +} diff --git a/lib/autoplan-gate.ts b/lib/autoplan-gate.ts index 98b381f4f..bfb0e28ef 100644 --- a/lib/autoplan-gate.ts +++ b/lib/autoplan-gate.ts @@ -24,7 +24,8 @@ import { consensusRecord, reconcile, remedyKey, renderConsensus, SETTLEABLE_CHAN import type { Phase } from './autoplan-prompts'; import { FILES, RunError, journal as journalRow, now, readDecisions, voiceRows, voiceTerminal, writeDecisions, type Amendment, type RunState, type VoiceState } from './autoplan-state'; import { applyCapChoices, pendingDeltas, releaseUndispatched, roundCap, roundOf } from './autoplan-budget'; -import { capyRelayLine } from './captain-brief'; +import { evaluateCarry, gateEvidence } from './autoplan-carry'; +import { buildEli5, capyRelayLine } from './captain-brief'; import { previousReview, reviewRows, settledIds } from './autoplan-confirm'; import { phaseClose } from './autoplan-timing'; import { ANSWERED_BY_RE, applyReply, decisionId, decisionRows, keepDirection, parseReply, pendingHint, recHash, validateGateList, type GateChoice, type GateItem, type GateList, type GateOption, type ParsedReply } from './gate-list'; @@ -237,7 +238,8 @@ export function buildGate(state: RunState, gateRev: number): GateList { } items.push(...capItems(state, { d: () => `d${++d}`, uc: () => `uc${++uc}` })); items.push(planApproval(state, items)); - const list: GateList = { schema_version: 1, run: state.run, gate_rev: gateRev, items, title: '/autoplan final approval gate', rec_rev: 0 }; + const list: GateList = { schema_version: 1, run: state.run, gate_rev: gateRev, items, title: '/autoplan final approval gate', rec_rev: 0, + ...(fs.existsSync(state.active_plan) ? { eli5: buildEli5(fs.readFileSync(state.active_plan, 'utf8'), state.active_plan) } : {}) }; const problems = validateGateList(list); if (problems.length) throw new Error(`gate list invalid: ${problems.join('; ')}`); return list; @@ -285,9 +287,29 @@ export function recommendGate(state: RunState, gateRev: number, itemId: string, ...[capyRelayLine(state.out)].filter((l): l is string => !!l)]; } -/** Every phase closed: write the gate list, decisions, tasks and the exported files; the run ends gate_pending. Unchanged answered proposals carry forward. */ +/** The gate list the owner last answered: gate.json, or its archive (`gate.r[.v].json`) once a reopen moved it. */ +function previousGate(out: string, rev: number): GateList | undefined { + if (rev < 1) return undefined; + const re = new RegExp(`^gate\\.r${rev}(?:\\.v(\\d+))?\\.json$`); + const archived = fs.readdirSync(out).map(f => [f, re.exec(f)] as const).filter(([, m]) => m).sort((a, b) => Number(b[1]![1] ?? 1) - Number(a[1]![1] ?? 1)).map(([f]) => f); + for (const name of ['gate.json', ...archived]) { + const file = path.join(out, name); + if (!fs.existsSync(file)) continue; + const list = JSON.parse(fs.readFileSync(file, 'utf8')) as GateList; + if (list.gate_rev === rev) return list; + } + return undefined; +} + +/** + * Every phase closed: write the gate list, decisions, tasks and the exported files; the run ends gate_pending. Unchanged + * answered proposals carry forward, and a later revision that asks nothing new carries the owner's last answer + * (lib/autoplan-carry.ts; its lines are on `carryLines(state)`). + */ export function finishGate(state: RunState): GateList { const gateRev = (state.gate_rev ?? 0) + 1; + const gateFile = path.join(state.out, 'gate.json'); + const prev = previousGate(state.out, state.gate_rev ?? 0); const list = buildGate(state, gateRev); const kept = readDecisions(state.out).filter(r => r.gate_rev !== gateRev); const answered = new Map(kept.filter(r => r.origin && r.chosen != null && r.status !== 'pending').map(r => [r.origin!, r])); @@ -295,7 +317,7 @@ export function finishGate(state: RunState): GateList { const prev = item.origin ? answered.get(item.origin) : undefined; if (prev && item.options.some(o => o.key === prev.chosen)) item.carried = { from: prev.carried_from ?? prev.id, chosen: prev.chosen!, gate_rev: prev.gate_rev }; } - fs.writeFileSync(path.join(state.out, 'gate.json'), JSON.stringify(list, null, 2) + '\n'); + fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n'); const rows = decisionRows(list).map(row => { const item = list.items.find(i => decisionId(list.run, gateRev, i.id) === row.id)!; if (!item.carried) return row; @@ -308,7 +330,8 @@ export function finishGate(state: RunState): GateList { state.status = 'gate_pending'; state.gate = { rev: gateRev, items: list.items.length, written_at: now() }; exportRun(state); - journalRow(state.out, { event: 'gate_written', gate_rev: gateRev, items: list.items.length, carried: list.items.filter(i => i.carried).map(i => i.id) }); + journalRow(state.out, { event: 'gate_written', gate_rev: gateRev, items: list.items.length, carried: list.items.filter(i => i.carried).map(i => i.id), ...gateEvidence(state, gateRev) }); + evaluateCarry(state, prev, list); return list; } @@ -321,6 +344,8 @@ export interface AnswerOptions { recHash?: string; /** The owner's verbatim words; required for every `--by agent:` answer. */ ownerReply?: string; + /** A whole-gate carry (lib/autoplan-carry.ts): the earlier p1 decision and the owner's words it reuses. Replaces ownerReply and its re-parse. */ + carry?: { from: string; gate_rev: number; words: string; by?: string; reason: string }; } /** How an accepted plan change is discharged: a packaging/ordering change settles without a plan edit. */ @@ -361,7 +386,7 @@ export function answerGate(state: RunState, gateRev: number, reply: string, answ if (state.status !== 'gate_pending' || !fs.existsSync(gateFile)) throw new RunError('PHASE_NOT_AWAITING', `run is ${state.status}, not gate_pending`, 3); if (answeredBy !== undefined && !ANSWERED_BY_RE.test(answeredBy)) throw new RunError('ANSWER_BY_REQUIRED', `--by ${JSON.stringify(answeredBy)} is neither human nor agent:`, 3); if (answeredBy === undefined && state.session_kind === 'unattended') throw new RunError('ANSWER_BY_REQUIRED', 'an unattended run records who answered; pass --by human or --by agent:', 3); - if (answeredBy?.startsWith('agent:') && o.ownerReply === undefined) throw new RunError('OWNER_REPLY_REQUIRED', `${answeredBy} relayed or decided this reply; pass the owner's words verbatim with --owner-reply "" (or --owner-reply - / --owner-reply-file )`, 3); + if (answeredBy?.startsWith('agent:') && o.ownerReply === undefined && !o.carry) throw new RunError('OWNER_REPLY_REQUIRED', `${answeredBy} relayed or decided this reply; pass the owner's words verbatim with --owner-reply "" (or --owner-reply - / --owner-reply-file )`, 3); const list = JSON.parse(fs.readFileSync(gateFile, 'utf8')) as GateList; const rows = readDecisions(state.out); const prior = Object.fromEntries(rows.filter(r => r.gate_rev === list.gate_rev && r.chosen != null && r.status !== 'pending' && r.label).map(r => [r.label!.toLowerCase(), r.chosen!])); @@ -372,8 +397,9 @@ export function answerGate(state: RunState, gateRev: number, reply: string, answ if (o.recHash !== undefined && o.recHash !== currentHash) throw new RunError('GATE_REC_STALE', `--rec-hash ${o.recHash} was rendered before the recommendations changed; current rec_hash=${currentHash}`, 3); if (parsed.mode === 'rec' && list.rec_rev === undefined && o.recHash === undefined) throw new RunError('GATE_REC_STALE', `this gate was written before \`rec\`/\`all\` took plan changes (gate.json has no rec_rev), so the owner saw a different meaning; render it again (gstack-gate render ${gateFile} --all), relay it, and answer with --rec-hash ${currentHash}, or reply with explicit