@tangle-network/agent-eval 0.23.0 → 0.24.0

package/dist/chunk-ZN274SWR.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/action-policy.ts","../src/run-evidence.ts","../src/propose-review.ts","../src/propose-review-control.ts"],"sourcesContent":["import type { FeedbackLabel, ProposedSideEffect } from './feedback-trajectory'\n\nexport interface ActionExecutionPolicy {\n  allowedTypes?: string[]\n  blockedTypes?: string[]\n  alwaysRequireApprovalTypes?: string[]\n  autoApproveTypes?: string[]\n  requireApprovalForExternalSideEffects?: boolean\n  requireApprovalAboveCostUsd?: number\n  maxActionCostUsd?: number\n  remainingBudgetUsd?: number\n  expectedOutcomeRequired?: boolean\n  killCriteriaRequired?: boolean\n}\n\nexport interface ActionPolicyDecision {\n  allowed: boolean\n  blocked: boolean\n  requiresApproval: boolean\n  reasons: string[]\n  label?: FeedbackLabel\n}\n\nexport function evaluateActionPolicy(\n  action: ProposedSideEffect,\n  policy: ActionExecutionPolicy = {},\n  options: { createdAt?: string } = {},\n): ActionPolicyDecision {\n  const reasons: string[] = []\n  let blocked = false\n  let requiresApproval = Boolean(action.requiresApproval)\n\n  if (policy.allowedTypes?.length && !policy.allowedTypes.includes(action.type)) {\n    blocked = true\n    reasons.push(`action type \"${action.type}\" is not allowed`)\n  }\n  if (policy.blockedTypes?.includes(action.type)) {\n    blocked = true\n    reasons.push(`action type \"${action.type}\" is blocked`)\n  }\n  if (policy.alwaysRequireApprovalTypes?.includes(action.type)) {\n    requiresApproval = true\n    reasons.push(`action type \"${action.type}\" requires approval`)\n  }\n  if (policy.requireApprovalForExternalSideEffects && action.externalSideEffect) {\n    requiresApproval = true\n    reasons.push('external side effect requires approval')\n  }\n  if (\n    policy.requireApprovalAboveCostUsd !== undefined &&\n    (action.costUsd ?? 0) > policy.requireApprovalAboveCostUsd\n  ) {\n    requiresApproval = true\n    reasons.push(\n      `cost ${action.costUsd} exceeds approval threshold ${policy.requireApprovalAboveCostUsd}`,\n    )\n  }\n  if (policy.maxActionCostUsd !== undefined && (action.costUsd ?? 0) > policy.maxActionCostUsd) {\n    blocked = true\n    reasons.push(`cost ${action.costUsd} exceeds max action cost ${policy.maxActionCostUsd}`)\n  }\n  if (\n    policy.remainingBudgetUsd !== undefined &&\n    (action.costUsd ?? 0) > policy.remainingBudgetUsd\n  ) {\n    blocked = true\n    reasons.push(`cost ${action.costUsd} exceeds remaining budget ${policy.remainingBudgetUsd}`)\n  }\n  if (policy.expectedOutcomeRequired && !action.metadata?.expectedOutcome) {\n    blocked = true\n    reasons.push('expected outcome is required')\n  }\n  if (policy.killCriteriaRequired && !action.metadata?.killCriteria) {\n    blocked = true\n    reasons.push('kill criteria are required')\n  }\n  if (policy.autoApproveTypes?.includes(action.type) && requiresApproval) {\n    reasons.push(\n      `action type \"${action.type}\" is auto-approved only when no approval policy applies`,\n    )\n  }\n\n  if (!reasons.length) reasons.push(requiresApproval ? 'approval required' : 'action allowed')\n\n  const label =\n    blocked || requiresApproval\n      ? {\n          source: 'policy' as const,\n          kind: blocked ? ('policy_block' as const) : ('comment' as const),\n          value: { actionType: action.type, blocked, requiresApproval },\n          reason: reasons.join('; '),\n          severity: blocked ? ('critical' as const) : ('warning' as const),\n          createdAt: options.createdAt ?? new Date().toISOString(),\n          metadata: { action, policy },\n        }\n      : undefined\n\n  return {\n    allowed: !blocked,\n    blocked,\n    requiresApproval: !blocked && requiresApproval,\n    reasons,\n    label,\n  }\n}\n","import type { ControlEvalResult, ControlRunResult } from './control-runtime'\nimport {\n  type RunRecord,\n  type RunSplitTag,\n  type RunTokenUsage,\n  validateRunRecord,\n} from './run-record'\nimport type { FailureClass } from './trace/schema'\n\nexport interface RunEvidenceMetadata {\n  experimentId: string\n  candidateId: string\n  seed: number\n  model: string\n  promptHash: string\n  configHash: string\n  commitSha: string\n  splitTag: RunSplitTag\n  tokenUsage: RunTokenUsage\n  queueMs?: number\n  judgeMetadata?: RunRecord['judgeMetadata']\n  raw?: Record<string, number>\n}\n\nexport interface ControlRunToRunRecordOptions extends RunEvidenceMetadata {\n  runId?: string\n  score?: number\n  failureMode?: string\n}\n\n/**\n * Project a completed control-loop run into the strict RunRecord shape used by\n * release gates, optimizer tables, and research reports.\n *\n * The control loop owns live execution evidence. The caller still supplies the\n * experimental cell metadata because prompt/config hashes, split assignment,\n * model snapshot, and commit SHA are product/harness concerns.\n */\nexport function controlRunToRunRecord<\n  TState,\n  TAction,\n  TActionResult,\n  TEval extends ControlEvalResult = ControlEvalResult,\n>(\n  run: ControlRunResult<TState, TAction, TActionResult, TEval>,\n  options: ControlRunToRunRecordOptions,\n): RunRecord {\n  const score = clampScore(\n    options.score ?? run.score ?? scoreFromEvals(run.finalEvals) ?? (run.pass ? 1 : 0),\n  )\n  const outcome =\n    options.splitTag === 'holdout'\n      ? { holdoutScore: score, raw: normalizeRawMetrics(options.raw, run, score) }\n      : { searchScore: score, raw: normalizeRawMetrics(options.raw, run, score) }\n\n  return validateRunRecord({\n    runId:\n      options.runId ??\n      run.runId ??\n      `control:${options.experimentId}:${options.candidateId}:${options.seed}:${options.splitTag}`,\n    experimentId: options.experimentId,\n    candidateId: options.candidateId,\n    seed: options.seed,\n    model: options.model,\n    promptHash: options.promptHash,\n    configHash: options.configHash,\n    commitSha: options.commitSha,\n    wallMs: run.wallMs,\n    ...(options.queueMs !== undefined ? { queueMs: options.queueMs } : {}),\n    costUsd: run.spentCostUsd,\n    tokenUsage: options.tokenUsage,\n    ...(options.judgeMetadata ? { judgeMetadata: options.judgeMetadata } : {}),\n    outcome,\n    failureMode: options.failureMode ?? failureModeFromRun(run),\n    splitTag: options.splitTag,\n  })\n}\n\nexport function scoreFromEvals(evals: readonly ControlEvalResult[]): number | undefined {\n  const scores = evals\n    .map((e) => e.score)\n    .filter((score): score is number => typeof score === 'number' && Number.isFinite(score))\n  if (scores.length === 0) return undefined\n  return clampScore(scores.reduce((sum, score) => sum + score, 0) / scores.length)\n}\n\nfunction normalizeRawMetrics<TState, TAction, TActionResult, TEval extends ControlEvalResult>(\n  raw: Record<string, number> | undefined,\n  run: ControlRunResult<TState, TAction, TActionResult, TEval>,\n  score: number,\n): Record<string, number> {\n  return {\n    ...finiteOnly(raw ?? {}),\n    score,\n    pass: run.pass ? 1 : 0,\n    completed: run.completed ? 1 : 0,\n    steps: run.steps.length,\n    runtimeErrors: run.runtimeErrors.length,\n  }\n}\n\nfunction finiteOnly(values: Record<string, number>): Record<string, number> {\n  const out: Record<string, number> = {}\n  for (const [key, value] of Object.entries(values)) {\n    if (Number.isFinite(value)) out[key] = value\n  }\n  return out\n}\n\nfunction failureModeFromRun<TState, TAction, TActionResult, TEval extends ControlEvalResult>(\n  run: ControlRunResult<TState, TAction, TActionResult, TEval>,\n): FailureClass | undefined {\n  if (run.pass) return undefined\n  return run.failureClass ?? 'unknown'\n}\n\nfunction clampScore(value: number): number {\n  if (!Number.isFinite(value)) return 0\n  return Math.max(0, Math.min(1, value))\n}\n","/**\n * Propose / Verify / Review — the core multi-shot primitive.\n *\n *   shot N:  propose(state, priorReview) → new state\n *            verify(state)               → pass/fail, optional layers\n *            review(state, verification, memory) → observations + next-shot\n *                                                   instruction + shouldContinue\n *            memory.append(entry)\n *\n * Roles are strictly separated:\n *\n *   - The WORKER is whatever the caller wraps in `propose`. It is\n *     stateful — caller owns its resume/session mechanism.\n *   - The VERIFIER grades the state. It produces the ground truth.\n *     The reviewer cannot overturn or downgrade a verification layer.\n *   - The REVIEWER is stateless per call. Its continuity is the\n *     `ReviewMemoryStore` — durable JSONL by default, or any store\n *     implementing the interface. It reads memory + trace summary +\n *     verification and directs the NEXT proposer shot.\n *\n * This shape is load-bearing. The reviewer never grades; the verifier\n * never directs. Two processes, two prompts, two concerns — which is\n * what keeps the loop from confirmation-biasing itself into \"all\n * passed\" when it didn't.\n *\n * Short-circuits and soft-fails are both first-class:\n *   - verify.pass === true  → reviewer LLM call is skipped, memory\n *     records a success entry, loop exits.\n *   - review throws         → the shot still counts; the loop uses the\n *     last-known instruction (or `fallbackInstruction`) for the next\n *     propose call. A transient reviewer failure must NEVER abort a\n *     valid arc.\n *\n * Composable: `propose` itself can be another `runProposeReview` call.\n * That's the dogfooding path — a harness built on this primitive is in\n * turn evaluable by it.\n */\n\nimport { appendFileSync, existsSync, mkdirSync, readFileSync } from 'node:fs'\nimport { dirname } from 'node:path'\nimport { type SpanHandle, TraceEmitter } from './trace/emitter'\nimport type { FailureClass } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\n// ── Types ────────────────────────────────────────────────────────────\n\nexport interface Verification {\n  pass: boolean\n  score?: number\n  failingLayers?: string[]\n  details?: unknown\n}\n\nexport interface Review {\n  observations: string\n  diagnosis: string\n  nextShotInstruction: string\n  shouldContinue: boolean\n  confidence: number\n}\n\nexport interface ReviewMemoryEntry extends Review {\n  shot: number\n  timestamp: number\n  verification: {\n    pass: boolean\n    score?: number\n    failingLayers?: string[]\n  }\n}\n\nexport interface ProposeInput<State> {\n  shot: number\n  goal: string\n  state: State\n  priorReview: Review | null\n  abortSignal: AbortSignal\n  emitter?: TraceEmitter\n}\n\nexport interface ProposeOutput<State, Summary = unknown> {\n  state: State\n  traceSummary?: Summary\n}\n\nexport interface ReviewInput<State, Summary = unknown> {\n  shot: number\n  goal: string\n  state: State\n  verification: Verification\n  traceSummary: Summary | undefined\n  memory: ReviewMemoryEntry[]\n}\n\nexport type ProposeFn<State, Summary = unknown> = (\n  input: ProposeInput<State>,\n) => Promise<ProposeOutput<State, Summary>>\n\nexport type VerifyFn<State> = (state: State) => Promise<Verification>\n\nexport type ReviewFn<State, Summary = unknown> = (\n  input: ReviewInput<State, Summary>,\n) => Promise<Review>\n\nexport interface ReviewMemoryStore {\n  load(): Promise<ReviewMemoryEntry[]>\n  append(entry: ReviewMemoryEntry): Promise<void>\n}\n\nexport interface ProposeReviewConfig<State, Summary = unknown> {\n  goal: string\n  initialState: State\n  propose: ProposeFn<State, Summary>\n  verify: VerifyFn<State>\n  review: ReviewFn<State, Summary>\n  /** Hard shot cap. Default 10. */\n  maxShots?: number\n  /** Wall-clock cap in ms. Default 10 min. */\n  maxWallMs?: number\n  /**\n   * If the reviewer returns confidence ≤ floor on `confidenceFloorWindow`\n   * consecutive shots, terminate early. Default floor 0.3, window 2.\n   * Set window to 0 or floor to <0 to disable.\n   */\n  confidenceFloor?: number\n  confidenceFloorWindow?: number\n  /** Defaults to an in-memory store if omitted. */\n  memory?: ReviewMemoryStore\n  /** If provided, emit a Run + per-shot spans. */\n  store?: TraceStore\n  scenarioId?: string\n  projectId?: string\n  variantId?: string\n  /**\n   * Used when the reviewer soft-fails on shot 1 (no prior instruction to\n   * fall back to). Default is a generic \"inspect failures and fix\".\n   */\n  fallbackInstruction?: string\n}\n\nexport interface ProposeReviewShot<State, Summary = unknown> {\n  shot: number\n  state: State\n  verification: Verification\n  traceSummary: Summary | undefined\n  review: Review\n  reviewAvailable: boolean\n  reviewError?: string\n  durationMs: number\n}\n\nexport interface ProposeReviewReport<State, Summary = unknown> {\n  runId: string | null\n  completed: boolean\n  shots: ProposeReviewShot<State, Summary>[]\n  finalState: State\n  finalVerification: Verification\n  failureClass?: FailureClass\n  wallMs: number\n  score: number\n}\n\n// ── Memory stores ────────────────────────────────────────────────────\n\nexport function inMemoryReviewStore(initial: ReviewMemoryEntry[] = []): ReviewMemoryStore {\n  const entries: ReviewMemoryEntry[] = [...initial]\n  return {\n    async load() {\n      return [...entries]\n    },\n    async append(entry) {\n      entries.push(entry)\n    },\n  }\n}\n\nexport function jsonlReviewStore(path: string): ReviewMemoryStore {\n  return {\n    async load() {\n      if (!existsSync(path)) return []\n      const raw = readFileSync(path, 'utf8')\n      const out: ReviewMemoryEntry[] = []\n      for (const line of raw.split('\\n')) {\n        const trimmed = line.trim()\n        if (!trimmed) continue\n        try {\n          out.push(JSON.parse(trimmed) as ReviewMemoryEntry)\n        } catch {\n          // A corrupt line is an observability problem, not a correctness\n          // one — skip it rather than aborting the loop.\n        }\n      }\n      return out\n    },\n    async append(entry) {\n      mkdirSync(dirname(path), { recursive: true })\n      appendFileSync(path, `${JSON.stringify(entry)}\\n`)\n    },\n  }\n}\n\n// ── Core loop ────────────────────────────────────────────────────────\n\nconst DEFAULT_FALLBACK_INSTRUCTION =\n  'Inspect the verification failures above. Fix the critical issues first, then the major ones. Do not restate the failures — act on them.'\n\nexport async function runProposeReview<State, Summary = unknown>(\n  config: ProposeReviewConfig<State, Summary>,\n): Promise<ProposeReviewReport<State, Summary>> {\n  const maxShots = config.maxShots ?? 10\n  const maxWallMs = config.maxWallMs ?? 10 * 60 * 1000\n  const confidenceFloor = config.confidenceFloor ?? 0.3\n  const confidenceFloorWindow = config.confidenceFloorWindow ?? 2\n  const memory = config.memory ?? inMemoryReviewStore()\n  const fallbackInstruction = config.fallbackInstruction ?? DEFAULT_FALLBACK_INSTRUCTION\n\n  const emitter = config.store ? new TraceEmitter(config.store) : null\n  if (emitter) {\n    await emitter.startRun({\n      scenarioId: config.scenarioId ?? 'propose-review',\n      projectId: config.projectId,\n      variantId: config.variantId,\n      layer: 'meta',\n      tags: {\n        goal: config.goal.slice(0, 120),\n        maxShots: String(maxShots),\n      },\n    })\n  }\n\n  const abort = new AbortController()\n  const wallStart = Date.now()\n  const wallTimer = setTimeout(\n    () => abort.abort(new Error('propose-review wall timeout')),\n    maxWallMs,\n  )\n\n  const shots: ProposeReviewShot<State, Summary>[] = []\n  let state = config.initialState\n  let priorReview: Review | null = null\n  let lastVerification: Verification = { pass: false }\n  let failureClass: FailureClass | undefined\n  let completed = false\n  let lowConfidenceStreak = 0\n\n  try {\n    for (let shot = 1; shot <= maxShots; shot++) {\n      if (abort.signal.aborted) {\n        failureClass = 'timeout'\n        break\n      }\n\n      const shotStart = Date.now()\n      const shotHandle = emitter ? await emitter.span({ kind: 'tool', name: `shot-${shot}` }) : null\n\n      // 1. Propose.\n      let proposeOut: ProposeOutput<State, Summary>\n      try {\n        proposeOut = await config.propose({\n          shot,\n          goal: config.goal,\n          state,\n          priorReview,\n          abortSignal: abort.signal,\n          emitter: emitter ?? undefined,\n        })\n      } catch (err) {\n        await shotHandle?.fail(err instanceof Error ? err : String(err))\n        failureClass = 'unknown'\n        throw err\n      }\n      state = proposeOut.state\n      const traceSummary = proposeOut.traceSummary\n\n      // 2. Verify.\n      let verification: Verification\n      try {\n        verification = await config.verify(state)\n      } catch (err) {\n        await shotHandle?.fail(err instanceof Error ? err : String(err))\n        failureClass = 'unknown'\n        throw err\n      }\n      lastVerification = verification\n\n      // 3. Review — short-circuit on pass; soft-fail on throw.\n      const memorySnapshot = await memory.load()\n      const verificationDigest = {\n        pass: verification.pass,\n        score: verification.score,\n        failingLayers: verification.failingLayers ?? [],\n      }\n\n      let review: Review\n      let reviewAvailable = true\n      let reviewError: string | undefined\n\n      if (verification.pass) {\n        review = {\n          observations: 'verification passed — skipping reviewer LLM call',\n          diagnosis: 'no failures to diagnose',\n          nextShotInstruction: '(done)',\n          shouldContinue: false,\n          confidence: 1,\n        }\n      } else {\n        try {\n          review = await config.review({\n            shot,\n            goal: config.goal,\n            state,\n            verification,\n            traceSummary,\n            memory: memorySnapshot,\n          })\n          review = coerceReview(review)\n        } catch (err) {\n          reviewAvailable = false\n          reviewError = err instanceof Error ? err.message : String(err)\n          const lastInstruction =\n            memorySnapshot.length > 0\n              ? memorySnapshot[memorySnapshot.length - 1]!.nextShotInstruction\n              : fallbackInstruction\n          review = {\n            observations: '(reviewer unavailable — using last-known instruction)',\n            diagnosis: reviewError,\n            nextShotInstruction: lastInstruction,\n            shouldContinue: true,\n            confidence: 0.3,\n          }\n        }\n      }\n\n      const entry: ReviewMemoryEntry = {\n        shot,\n        timestamp: Date.now(),\n        ...review,\n        verification: verificationDigest,\n      }\n      await memory.append(entry)\n\n      const shotRecord: ProposeReviewShot<State, Summary> = {\n        shot,\n        state,\n        verification,\n        traceSummary,\n        review,\n        reviewAvailable,\n        reviewError,\n        durationMs: Date.now() - shotStart,\n      }\n      shots.push(shotRecord)\n\n      await shotHandle?.end({\n        attributes: {\n          verificationPass: verification.pass,\n          verificationScore: verification.score ?? null,\n          reviewShouldContinue: review.shouldContinue,\n          reviewConfidence: review.confidence,\n          reviewAvailable,\n        },\n      } as Partial<SpanHandle['span']>)\n\n      // 4. Exit rules.\n      if (verification.pass) {\n        completed = true\n        break\n      }\n      if (!review.shouldContinue) {\n        break\n      }\n      if (confidenceFloorWindow > 0 && review.confidence <= confidenceFloor) {\n        lowConfidenceStreak += 1\n        if (lowConfidenceStreak >= confidenceFloorWindow) break\n      } else {\n        lowConfidenceStreak = 0\n      }\n\n      priorReview = review\n    }\n\n    if (!completed && !failureClass) {\n      failureClass = shots.length >= maxShots ? 'budget_exceeded' : 'unknown'\n    }\n  } finally {\n    clearTimeout(wallTimer)\n  }\n\n  const score = lastVerification.pass\n    ? 1\n    : typeof lastVerification.score === 'number'\n      ? lastVerification.score\n      : 0\n\n  if (emitter) {\n    await emitter.endRun({\n      pass: completed,\n      score,\n      failureClass,\n      notes: `${shots.length} shot(s); final pass=${lastVerification.pass}`,\n    })\n  }\n\n  return {\n    runId: emitter?.runId ?? null,\n    completed,\n    shots,\n    finalState: state,\n    finalVerification: lastVerification,\n    failureClass,\n    wallMs: Date.now() - wallStart,\n    score,\n  }\n}\n\n// ── Reviewer helper (LLM-backed) ─────────────────────────────────────\n\nexport type LlmJsonCall = (req: { system: string; user: string }) => Promise<unknown>\n\nexport interface LlmReviewerConfig<State, Summary = unknown> {\n  callJson: LlmJsonCall\n  renderState?: (state: State) => string\n  renderTraceSummary?: (summary: Summary | undefined) => string\n  /** Appended to the default system prompt. */\n  systemPromptAddendum?: string\n}\n\nconst REVIEWER_SYSTEM_PROMPT = `You are a senior reviewer directing a multi-shot build loop.\nYou do NOT grade — the verifier already did. Your job is to direct the worker's next shot.\nYou are blind to the worker's inner monologue. You see what it DID, not what it thought.\nReturn STRICT JSON matching the schema. No prose outside the JSON.`\n\nexport function createLlmReviewer<State, Summary = unknown>(\n  cfg: LlmReviewerConfig<State, Summary>,\n): ReviewFn<State, Summary> {\n  const renderState = cfg.renderState ?? ((s: State) => safeJson(s))\n  const renderTraceSummary =\n    cfg.renderTraceSummary ??\n    ((s: Summary | undefined) => (s === undefined ? '(none)' : safeJson(s)))\n  const system = cfg.systemPromptAddendum\n    ? `${REVIEWER_SYSTEM_PROMPT}\\n\\n${cfg.systemPromptAddendum}`\n    : REVIEWER_SYSTEM_PROMPT\n\n  return async (input) => {\n    const memoryBlock =\n      input.memory.length === 0\n        ? '(no prior shots — this is shot 1)'\n        : input.memory\n            .map((m) =>\n              [\n                `shot ${m.shot} — verification.pass=${m.verification.pass}` +\n                  (typeof m.verification.score === 'number'\n                    ? ` score=${m.verification.score.toFixed(2)}`\n                    : '') +\n                  ` confidence=${m.confidence.toFixed(2)} failing=[${(m.verification.failingLayers ?? []).join(',')}]`,\n                `  observations: ${m.observations.slice(0, 400)}`,\n                `  diagnosis: ${m.diagnosis.slice(0, 400)}`,\n                `  instruction given: ${m.nextShotInstruction.slice(0, 400)}`,\n              ].join('\\n'),\n            )\n            .join('\\n\\n')\n\n    const user = [\n      `=== GOAL ===`,\n      input.goal,\n      ``,\n      `=== SHOT NUMBER ===`,\n      String(input.shot),\n      ``,\n      `=== CURRENT STATE ===`,\n      renderState(input.state),\n      ``,\n      `=== TRACE SUMMARY ===`,\n      renderTraceSummary(input.traceSummary),\n      ``,\n      `=== VERIFICATION ===`,\n      summarizeVerification(input.verification),\n      ``,\n      `=== REVIEWER MEMORY (prior shots) ===`,\n      memoryBlock,\n      ``,\n      `=== YOUR TASK ===`,\n      `Return STRICT JSON:`,\n      `{`,\n      `  \"observations\": string (20..2000 chars, first-person worker behavior — quote counts, errors, loops)`,\n      `  \"diagnosis\": string (20..1500 chars, root cause, NOT a restatement of verification)`,\n      `  \"nextShotInstruction\": string (40..3000 chars, concrete directive to the worker)`,\n      `  \"shouldContinue\": boolean (false if verification.pass, or if thrashing, or unachievable)`,\n      `  \"confidence\": number in [0,1]`,\n      `}`,\n    ].join('\\n')\n\n    const raw = await cfg.callJson({ system, user })\n    return coerceReview(raw as Partial<Review>)\n  }\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────\n\nfunction coerceReview(raw: Partial<Review> | null | undefined): Review {\n  if (!raw || typeof raw !== 'object') {\n    throw new Error('reviewer returned non-object')\n  }\n  const observations = typeof raw.observations === 'string' ? raw.observations : ''\n  const diagnosis = typeof raw.diagnosis === 'string' ? raw.diagnosis : ''\n  const nextShotInstruction =\n    typeof raw.nextShotInstruction === 'string' ? raw.nextShotInstruction : ''\n  if (!observations || !diagnosis || !nextShotInstruction) {\n    throw new Error('reviewer missing required string fields')\n  }\n  if (typeof raw.shouldContinue !== 'boolean') {\n    throw new Error('reviewer missing shouldContinue boolean')\n  }\n  const confidenceRaw = Number(raw.confidence)\n  if (!Number.isFinite(confidenceRaw)) {\n    throw new Error('reviewer confidence not finite')\n  }\n  return {\n    observations,\n    diagnosis,\n    nextShotInstruction,\n    shouldContinue: raw.shouldContinue,\n    confidence: Math.max(0, Math.min(1, confidenceRaw)),\n  }\n}\n\nfunction summarizeVerification(v: Verification): string {\n  const header =\n    `pass=${v.pass}` +\n    (typeof v.score === 'number' ? ` score=${v.score.toFixed(3)}` : '') +\n    (v.failingLayers && v.failingLayers.length > 0\n      ? ` failing=[${v.failingLayers.join(', ')}]`\n      : '')\n  const details = v.details === undefined ? '' : `\\n${safeJson(v.details).slice(0, 1500)}`\n  return header + details\n}\n\nfunction safeJson(x: unknown): string {\n  try {\n    return JSON.stringify(x, null, 2)\n  } catch {\n    return String(x)\n  }\n}\n","import {\n  type ControlRunResult,\n  type ControlRuntimeConfig,\n  objectiveEval,\n  runAgentControlLoop,\n} from './control-runtime'\nimport {\n  inMemoryReviewStore,\n  type ProposeFn,\n  type ProposeOutput,\n  type Review,\n  type ReviewFn,\n  type ReviewMemoryEntry,\n  type ReviewMemoryStore,\n  type Verification,\n  type VerifyFn,\n} from './propose-review'\nimport type { FailureClass } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface ProposeReviewControlState<State, Summary = unknown> {\n  shot: number\n  state: State\n  priorReview: Review | null\n  verification: Verification\n  traceSummary?: Summary\n  memory: ReviewMemoryEntry[]\n  completed: boolean\n  reviewAvailable: boolean\n  reviewError?: string\n}\n\nexport interface ProposeReviewControlAction {\n  type: 'propose-review-shot'\n  shot: number\n}\n\nexport interface ProposeReviewControlResult<State, Summary = unknown> {\n  state: State\n  verification: Verification\n  traceSummary?: Summary\n  review: Review | null\n  reviewAvailable: boolean\n  reviewError?: string\n}\n\nexport interface ProposeReviewControlConfig<State, Summary = unknown> {\n  goal: string\n  initialState: State\n  propose: ProposeFn<State, Summary>\n  verify: VerifyFn<State>\n  review: ReviewFn<State, Summary>\n  maxShots?: number\n  maxWallMs?: number\n  memory?: ReviewMemoryStore\n  store?: TraceStore\n  scenarioId?: string\n  projectId?: string\n  variantId?: string\n  fallbackInstruction?: string\n  confidenceFloor?: number\n  confidenceFloorWindow?: number\n  failureClassFromVerification?: (verification: Verification) => FailureClass | undefined\n  actionFailure?: ControlRuntimeConfig<\n    ProposeReviewControlState<State, Summary>,\n    ProposeReviewControlAction,\n    ProposeReviewControlResult<State, Summary>\n  >['actionFailure']\n}\n\nconst DEFAULT_FALLBACK_INSTRUCTION =\n  'Inspect the verification failures above. Fix the critical issues first, then the major ones. Do not restate the failures — act on them.'\n\nexport async function runProposeReviewAsControlLoop<State, Summary = unknown>(\n  config: ProposeReviewControlConfig<State, Summary>,\n): Promise<\n  ControlRunResult<\n    ProposeReviewControlState<State, Summary>,\n    ProposeReviewControlAction,\n    ProposeReviewControlResult<State, Summary>\n  >\n> {\n  const maxShots = config.maxShots ?? 10\n  const confidenceFloor = config.confidenceFloor ?? 0.3\n  const confidenceFloorWindow = config.confidenceFloorWindow ?? 2\n  const memory = config.memory ?? inMemoryReviewStore()\n  const fallbackInstruction = config.fallbackInstruction ?? DEFAULT_FALLBACK_INSTRUCTION\n  const failureClassFromVerification =\n    config.failureClassFromVerification ?? controlFailureClassFromVerification\n  let lowConfidenceStreak = 0\n\n  let current: ProposeReviewControlState<State, Summary> = {\n    shot: 0,\n    state: config.initialState,\n    priorReview: null,\n    verification: { pass: false },\n    memory: await memory.load(),\n    completed: false,\n    reviewAvailable: false,\n  }\n\n  return runAgentControlLoop({\n    intent: config.goal,\n    budget: { maxSteps: maxShots, maxWallMs: config.maxWallMs },\n    store: config.store,\n    scenarioId: config.scenarioId ?? 'propose-review-control',\n    projectId: config.projectId,\n    variantId: config.variantId,\n    actionFailure: config.actionFailure ?? 'stop',\n    observe: () => current,\n    validate: ({ state }) => [\n      objectiveEval({\n        id: 'verification',\n        passed: state.verification.pass,\n        score: state.verification.score,\n        severity: 'critical',\n        detail: state.verification.pass\n          ? 'verification passed'\n          : `verification failed${state.verification.failingLayers?.length ? `: ${state.verification.failingLayers.join(', ')}` : ''}`,\n      }),\n    ],\n    shouldStop: ({ state }) => {\n      if (state.verification.pass) {\n        return {\n          stop: true,\n          pass: true,\n          reason: 'verification passed',\n          score: state.verification.score,\n        }\n      }\n      if (state.completed) {\n        return {\n          stop: true,\n          pass: false,\n          reason: 'reviewer stopped continuation',\n          score: state.verification.score,\n          failureClass: failureClassFromVerification(state.verification),\n        }\n      }\n      return {\n        stop: false,\n        pass: false,\n        reason: 'verification still failing',\n        score: state.verification.score,\n      }\n    },\n    decide: ({ state }) => ({\n      type: 'continue',\n      action: { type: 'propose-review-shot', shot: state.shot + 1 },\n      reason: state.priorReview?.nextShotInstruction ?? fallbackInstruction,\n    }),\n    act: async (action, ctx) => {\n      const shot = action.shot\n      const proposeOut: ProposeOutput<State, Summary> = await config.propose({\n        shot,\n        goal: config.goal,\n        state: current.state,\n        priorReview: current.priorReview,\n        abortSignal: ctx.abortSignal,\n        emitter: ctx.emitter,\n      })\n\n      const nextState = proposeOut.state\n      const verification = await config.verify(nextState)\n      let review: Review | null = null\n      let reviewAvailable = false\n      let reviewError: string | undefined\n      let shouldContinue = !verification.pass\n\n      if (!verification.pass) {\n        try {\n          review = await config.review({\n            shot,\n            goal: config.goal,\n            state: nextState,\n            verification,\n            traceSummary: proposeOut.traceSummary,\n            memory: await memory.load(),\n          })\n          reviewAvailable = true\n          shouldContinue = review.shouldContinue\n          lowConfidenceStreak = review.confidence <= confidenceFloor ? lowConfidenceStreak + 1 : 0\n          if (confidenceFloorWindow > 0 && lowConfidenceStreak >= confidenceFloorWindow)\n            shouldContinue = false\n        } catch (err) {\n          reviewError = err instanceof Error ? err.message : String(err)\n          review = current.priorReview ?? {\n            observations: 'Reviewer unavailable.',\n            diagnosis: reviewError,\n            nextShotInstruction: fallbackInstruction,\n            shouldContinue: true,\n            confidence: 0,\n          }\n          shouldContinue = true\n        }\n      } else {\n        review = {\n          observations: 'Verification passed.',\n          diagnosis: 'No further revision needed.',\n          nextShotInstruction: '',\n          shouldContinue: false,\n          confidence: 1,\n        }\n      }\n\n      const entry: ReviewMemoryEntry = {\n        ...(review ?? {\n          observations: 'No review.',\n          diagnosis: '',\n          nextShotInstruction: fallbackInstruction,\n          shouldContinue,\n          confidence: 0,\n        }),\n        shot,\n        timestamp: Date.now(),\n        verification: {\n          pass: verification.pass,\n          score: verification.score,\n          failingLayers: verification.failingLayers,\n        },\n      }\n      await memory.append(entry)\n\n      current = {\n        shot,\n        state: nextState,\n        priorReview: review,\n        verification,\n        traceSummary: proposeOut.traceSummary,\n        memory: await memory.load(),\n        completed: verification.pass || !shouldContinue,\n        reviewAvailable,\n        reviewError,\n      }\n\n      return {\n        state: nextState,\n        verification,\n        traceSummary: proposeOut.traceSummary,\n        review,\n        reviewAvailable,\n        reviewError,\n      }\n    },\n  })\n}\n\nexport function controlFailureClassFromVerification(\n  verification: Verification,\n): FailureClass | undefined {\n  if (verification.pass) return undefined\n  return verification.failingLayers?.length ? 'instruction_following' : 'unknown'\n}\n"],"mappings":";;;;;;;;;;;;AAuBO,SAAS,qBACd,QACA,SAAgC,CAAC,GACjC,UAAkC,CAAC,GACb;AACtB,QAAM,UAAoB,CAAC;AAC3B,MAAI,UAAU;AACd,MAAI,mBAAmB,QAAQ,OAAO,gBAAgB;AAEtD,MAAI,OAAO,cAAc,UAAU,CAAC,OAAO,aAAa,SAAS,OAAO,IAAI,GAAG;AAC7E,cAAU;AACV,YAAQ,KAAK,gBAAgB,OAAO,IAAI,kBAAkB;AAAA,EAC5D;AACA,MAAI,OAAO,cAAc,SAAS,OAAO,IAAI,GAAG;AAC9C,cAAU;AACV,YAAQ,KAAK,gBAAgB,OAAO,IAAI,cAAc;AAAA,EACxD;AACA,MAAI,OAAO,4BAA4B,SAAS,OAAO,IAAI,GAAG;AAC5D,uBAAmB;AACnB,YAAQ,KAAK,gBAAgB,OAAO,IAAI,qBAAqB;AAAA,EAC/D;AACA,MAAI,OAAO,yCAAyC,OAAO,oBAAoB;AAC7E,uBAAmB;AACnB,YAAQ,KAAK,wCAAwC;AAAA,EACvD;AACA,MACE,OAAO,gCAAgC,WACtC,OAAO,WAAW,KAAK,OAAO,6BAC/B;AACA,uBAAmB;AACnB,YAAQ;AAAA,MACN,QAAQ,OAAO,OAAO,+BAA+B,OAAO,2BAA2B;AAAA,IACzF;AAAA,EACF;AACA,MAAI,OAAO,qBAAqB,WAAc,OAAO,WAAW,KAAK,OAAO,kBAAkB;AAC5F,cAAU;AACV,YAAQ,KAAK,QAAQ,OAAO,OAAO,4BAA4B,OAAO,gBAAgB,EAAE;AAAA,EAC1F;AACA,MACE,OAAO,uBAAuB,WAC7B,OAAO,WAAW,KAAK,OAAO,oBAC/B;AACA,cAAU;AACV,YAAQ,KAAK,QAAQ,OAAO,OAAO,6BAA6B,OAAO,kBAAkB,EAAE;AAAA,EAC7F;AACA,MAAI,OAAO,2BAA2B,CAAC,OAAO,UAAU,iBAAiB;AACvE,cAAU;AACV,YAAQ,KAAK,8BAA8B;AAAA,EAC7C;AACA,MAAI,OAAO,wBAAwB,CAAC,OAAO,UAAU,cAAc;AACjE,cAAU;AACV,YAAQ,KAAK,4BAA4B;AAAA,EAC3C;AACA,MAAI,OAAO,kBAAkB,SAAS,OAAO,IAAI,KAAK,kBAAkB;AACtE,YAAQ;AAAA,MACN,gBAAgB,OAAO,IAAI;AAAA,IAC7B;AAAA,EACF;AAEA,MAAI,CAAC,QAAQ,OAAQ,SAAQ,KAAK,mBAAmB,sBAAsB,gBAAgB;AAE3F,QAAM,QACJ,WAAW,mBACP;AAAA,IACE,QAAQ;AAAA,IACR,MAAM,UAAW,iBAA4B;AAAA,IAC7C,OAAO,EAAE,YAAY,OAAO,MAAM,SAAS,iBAAiB;AAAA,IAC5D,QAAQ,QAAQ,KAAK,IAAI;AAAA,IACzB,UAAU,UAAW,aAAwB;AAAA,IAC7C,WAAW,QAAQ,cAAa,oBAAI,KAAK,GAAE,YAAY;AAAA,IACvD,UAAU,EAAE,QAAQ,OAAO;AAAA,EAC7B,IACA;AAEN,SAAO;AAAA,IACL,SAAS,CAAC;AAAA,IACV;AAAA,IACA,kBAAkB,CAAC,WAAW;AAAA,IAC9B;AAAA,IACA;AAAA,EACF;AACF;;;AClEO,SAAS,sBAMd,KACA,SACW;AACX,QAAM,QAAQ;AAAA,IACZ,QAAQ,SAAS,IAAI,SAAS,eAAe,IAAI,UAAU,MAAM,IAAI,OAAO,IAAI;AAAA,EAClF;AACA,QAAM,UACJ,QAAQ,aAAa,YACjB,EAAE,cAAc,OAAO,KAAK,oBAAoB,QAAQ,KAAK,KAAK,KAAK,EAAE,IACzE,EAAE,aAAa,OAAO,KAAK,oBAAoB,QAAQ,KAAK,KAAK,KAAK,EAAE;AAE9E,SAAO,kBAAkB;AAAA,IACvB,OACE,QAAQ,SACR,IAAI,SACJ,WAAW,QAAQ,YAAY,IAAI,QAAQ,WAAW,IAAI,QAAQ,IAAI,IAAI,QAAQ,QAAQ;AAAA,IAC5F,cAAc,QAAQ;AAAA,IACtB,aAAa,QAAQ;AAAA,IACrB,MAAM,QAAQ;AAAA,IACd,OAAO,QAAQ;AAAA,IACf,YAAY,QAAQ;AAAA,IACpB,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,IACnB,QAAQ,IAAI;AAAA,IACZ,GAAI,QAAQ,YAAY,SAAY,EAAE,SAAS,QAAQ,QAAQ,IAAI,CAAC;AAAA,IACpE,SAAS,IAAI;AAAA,IACb,YAAY,QAAQ;AAAA,IACpB,GAAI,QAAQ,gBAAgB,EAAE,eAAe,QAAQ,cAAc,IAAI,CAAC;AAAA,IACxE;AAAA,IACA,aAAa,QAAQ,eAAe,mBAAmB,GAAG;AAAA,IAC1D,UAAU,QAAQ;AAAA,EACpB,CAAC;AACH;AAEO,SAAS,eAAe,OAAyD;AACtF,QAAM,SAAS,MACZ,IAAI,CAAC,MAAM,EAAE,KAAK,EAClB,OAAO,CAAC,UAA2B,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,CAAC;AACzF,MAAI,OAAO,WAAW,EAAG,QAAO;AAChC,SAAO,WAAW,OAAO,OAAO,CAAC,KAAK,UAAU,MAAM,OAAO,CAAC,IAAI,OAAO,MAAM;AACjF;AAEA,SAAS,oBACP,KACA,KACA,OACwB;AACxB,SAAO;AAAA,IACL,GAAG,WAAW,OAAO,CAAC,CAAC;AAAA,IACvB;AAAA,IACA,MAAM,IAAI,OAAO,IAAI;AAAA,IACrB,WAAW,IAAI,YAAY,IAAI;AAAA,IAC/B,OAAO,IAAI,MAAM;AAAA,IACjB,eAAe,IAAI,cAAc;AAAA,EACnC;AACF;AAEA,SAAS,WAAW,QAAwD;AAC1E,QAAM,MAA8B,CAAC;AACrC,aAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,MAAM,GAAG;AACjD,QAAI,OAAO,SAAS,KAAK,EAAG,KAAI,GAAG,IAAI;AAAA,EACzC;AACA,SAAO;AACT;AAEA,SAAS,mBACP,KAC0B;AAC1B,MAAI,IAAI,KAAM,QAAO;AACrB,SAAO,IAAI,gBAAgB;AAC7B;AAEA,SAAS,WAAW,OAAuB;AACzC,MAAI,CAAC,OAAO,SAAS,KAAK,EAAG,QAAO;AACpC,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAK,CAAC;AACvC;;;ACjFA,SAAS,gBAAgB,YAAY,WAAW,oBAAoB;AACpE,SAAS,eAAe;AA6HjB,SAAS,oBAAoB,UAA+B,CAAC,GAAsB;AACxF,QAAM,UAA+B,CAAC,GAAG,OAAO;AAChD,SAAO;AAAA,IACL,MAAM,OAAO;AACX,aAAO,CAAC,GAAG,OAAO;AAAA,IACpB;AAAA,IACA,MAAM,OAAO,OAAO;AAClB,cAAQ,KAAK,KAAK;AAAA,IACpB;AAAA,EACF;AACF;AAEO,SAAS,iBAAiB,MAAiC;AAChE,SAAO;AAAA,IACL,MAAM,OAAO;AACX,UAAI,CAAC,WAAW,IAAI,EAAG,QAAO,CAAC;AAC/B,YAAM,MAAM,aAAa,MAAM,MAAM;AACrC,YAAM,MAA2B,CAAC;AAClC,iBAAW,QAAQ,IAAI,MAAM,IAAI,GAAG;AAClC,cAAM,UAAU,KAAK,KAAK;AAC1B,YAAI,CAAC,QAAS;AACd,YAAI;AACF,cAAI,KAAK,KAAK,MAAM,OAAO,CAAsB;AAAA,QACnD,QAAQ;AAAA,QAGR;AAAA,MACF;AACA,aAAO;AAAA,IACT;AAAA,IACA,MAAM,OAAO,OAAO;AAClB,gBAAU,QAAQ,IAAI,GAAG,EAAE,WAAW,KAAK,CAAC;AAC5C,qBAAe,MAAM,GAAG,KAAK,UAAU,KAAK,CAAC;AAAA,CAAI;AAAA,IACnD;AAAA,EACF;AACF;AAIA,IAAM,+BACJ;AAEF,eAAsB,iBACpB,QAC8C;AAC9C,QAAM,WAAW,OAAO,YAAY;AACpC,QAAM,YAAY,OAAO,aAAa,KAAK,KAAK;AAChD,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,wBAAwB,OAAO,yBAAyB;AAC9D,QAAM,SAAS,OAAO,UAAU,oBAAoB;AACpD,QAAM,sBAAsB,OAAO,uBAAuB;AAE1D,QAAM,UAAU,OAAO,QAAQ,IAAI,aAAa,OAAO,KAAK,IAAI;AAChE,MAAI,SAAS;AACX,UAAM,QAAQ,SAAS;AAAA,MACrB,YAAY,OAAO,cAAc;AAAA,MACjC,WAAW,OAAO;AAAA,MAClB,WAAW,OAAO;AAAA,MAClB,OAAO;AAAA,MACP,MAAM;AAAA,QACJ,MAAM,OAAO,KAAK,MAAM,GAAG,GAAG;AAAA,QAC9B,UAAU,OAAO,QAAQ;AAAA,MAC3B;AAAA,IACF,CAAC;AAAA,EACH;AAEA,QAAM,QAAQ,IAAI,gBAAgB;AAClC,QAAM,YAAY,KAAK,IAAI;AAC3B,QAAM,YAAY;AAAA,IAChB,MAAM,MAAM,MAAM,IAAI,MAAM,6BAA6B,CAAC;AAAA,IAC1D;AAAA,EACF;AAEA,QAAM,QAA6C,CAAC;AACpD,MAAI,QAAQ,OAAO;AACnB,MAAI,cAA6B;AACjC,MAAI,mBAAiC,EAAE,MAAM,MAAM;AACnD,MAAI;AACJ,MAAI,YAAY;AAChB,MAAI,sBAAsB;AAE1B,MAAI;AACF,aAAS,OAAO,GAAG,QAAQ,UAAU,QAAQ;AAC3C,UAAI,MAAM,OAAO,SAAS;AACxB,uBAAe;AACf;AAAA,MACF;AAEA,YAAM,YAAY,KAAK,IAAI;AAC3B,YAAM,aAAa,UAAU,MAAM,QAAQ,KAAK,EAAE,MAAM,QAAQ,MAAM,QAAQ,IAAI,GAAG,CAAC,IAAI;AAG1F,UAAI;AACJ,UAAI;AACF,qBAAa,MAAM,OAAO,QAAQ;AAAA,UAChC;AAAA,UACA,MAAM,OAAO;AAAA,UACb;AAAA,UACA;AAAA,UACA,aAAa,MAAM;AAAA,UACnB,SAAS,WAAW;AAAA,QACtB,CAAC;AAAA,MACH,SAAS,KAAK;AACZ,cAAM,YAAY,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;AAC/D,uBAAe;AACf,cAAM;AAAA,MACR;AACA,cAAQ,WAAW;AACnB,YAAM,eAAe,WAAW;AAGhC,UAAI;AACJ,UAAI;AACF,uBAAe,MAAM,OAAO,OAAO,KAAK;AAAA,MAC1C,SAAS,KAAK;AACZ,cAAM,YAAY,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;AAC/D,uBAAe;AACf,cAAM;AAAA,MACR;AACA,yBAAmB;AAGnB,YAAM,iBAAiB,MAAM,OAAO,KAAK;AACzC,YAAM,qBAAqB;AAAA,QACzB,MAAM,aAAa;AAAA,QACnB,OAAO,aAAa;AAAA,QACpB,eAAe,aAAa,iBAAiB,CAAC;AAAA,MAChD;AAEA,UAAI;AACJ,UAAI,kBAAkB;AACtB,UAAI;AAEJ,UAAI,aAAa,MAAM;AACrB,iBAAS;AAAA,UACP,cAAc;AAAA,UACd,WAAW;AAAA,UACX,qBAAqB;AAAA,UACrB,gBAAgB;AAAA,UAChB,YAAY;AAAA,QACd;AAAA,MACF,OAAO;AACL,YAAI;AACF,mBAAS,MAAM,OAAO,OAAO;AAAA,YAC3B;AAAA,YACA,MAAM,OAAO;AAAA,YACb;AAAA,YACA;AAAA,YACA;AAAA,YACA,QAAQ;AAAA,UACV,CAAC;AACD,mBAAS,aAAa,MAAM;AAAA,QAC9B,SAAS,KAAK;AACZ,4BAAkB;AAClB,wBAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC7D,gBAAM,kBACJ,eAAe,SAAS,IACpB,eAAe,eAAe,SAAS,CAAC,EAAG,sBAC3C;AACN,mBAAS;AAAA,YACP,cAAc;AAAA,YACd,WAAW;AAAA,YACX,qBAAqB;AAAA,YACrB,gBAAgB;AAAA,YAChB,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAEA,YAAM,QAA2B;AAAA,QAC/B;AAAA,QACA,WAAW,KAAK,IAAI;AAAA,QACpB,GAAG;AAAA,QACH,cAAc;AAAA,MAChB;AACA,YAAM,OAAO,OAAO,KAAK;AAEzB,YAAM,aAAgD;AAAA,QACpD;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,YAAY,KAAK,IAAI,IAAI;AAAA,MAC3B;AACA,YAAM,KAAK,UAAU;AAErB,YAAM,YAAY,IAAI;AAAA,QACpB,YAAY;AAAA,UACV,kBAAkB,aAAa;AAAA,UAC/B,mBAAmB,aAAa,SAAS;AAAA,UACzC,sBAAsB,OAAO;AAAA,UAC7B,kBAAkB,OAAO;AAAA,UACzB;AAAA,QACF;AAAA,MACF,CAAgC;AAGhC,UAAI,aAAa,MAAM;AACrB,oBAAY;AACZ;AAAA,MACF;AACA,UAAI,CAAC,OAAO,gBAAgB;AAC1B;AAAA,MACF;AACA,UAAI,wBAAwB,KAAK,OAAO,cAAc,iBAAiB;AACrE,+BAAuB;AACvB,YAAI,uBAAuB,sBAAuB;AAAA,MACpD,OAAO;AACL,8BAAsB;AAAA,MACxB;AAEA,oBAAc;AAAA,IAChB;AAEA,QAAI,CAAC,aAAa,CAAC,cAAc;AAC/B,qBAAe,MAAM,UAAU,WAAW,oBAAoB;AAAA,IAChE;AAAA,EACF,UAAE;AACA,iBAAa,SAAS;AAAA,EACxB;AAEA,QAAM,QAAQ,iBAAiB,OAC3B,IACA,OAAO,iBAAiB,UAAU,WAChC,iBAAiB,QACjB;AAEN,MAAI,SAAS;AACX,UAAM,QAAQ,OAAO;AAAA,MACnB,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA,OAAO,GAAG,MAAM,MAAM,wBAAwB,iBAAiB,IAAI;AAAA,IACrE,CAAC;AAAA,EACH;AAEA,SAAO;AAAA,IACL,OAAO,SAAS,SAAS;AAAA,IACzB;AAAA,IACA;AAAA,IACA,YAAY;AAAA,IACZ,mBAAmB;AAAA,IACnB;AAAA,IACA,QAAQ,KAAK,IAAI,IAAI;AAAA,IACrB;AAAA,EACF;AACF;AAcA,IAAM,yBAAyB;AAAA;AAAA;AAAA;AAKxB,SAAS,kBACd,KAC0B;AAC1B,QAAM,cAAc,IAAI,gBAAgB,CAAC,MAAa,SAAS,CAAC;AAChE,QAAM,qBACJ,IAAI,uBACH,CAAC,MAA4B,MAAM,SAAY,WAAW,SAAS,CAAC;AACvE,QAAM,SAAS,IAAI,uBACf,GAAG,sBAAsB;AAAA;AAAA,EAAO,IAAI,oBAAoB,KACxD;AAEJ,SAAO,OAAO,UAAU;AACtB,UAAM,cACJ,MAAM,OAAO,WAAW,IACpB,2CACA,MAAM,OACH;AAAA,MAAI,CAAC,MACJ;AAAA,QACE,QAAQ,EAAE,IAAI,6BAAwB,EAAE,aAAa,IAAI,MACtD,OAAO,EAAE,aAAa,UAAU,WAC7B,UAAU,EAAE,aAAa,MAAM,QAAQ,CAAC,CAAC,KACzC,MACJ,eAAe,EAAE,WAAW,QAAQ,CAAC,CAAC,cAAc,EAAE,aAAa,iBAAiB,CAAC,GAAG,KAAK,GAAG,CAAC;AAAA,QACnG,mBAAmB,EAAE,aAAa,MAAM,GAAG,GAAG,CAAC;AAAA,QAC/C,gBAAgB,EAAE,UAAU,MAAM,GAAG,GAAG,CAAC;AAAA,QACzC,wBAAwB,EAAE,oBAAoB,MAAM,GAAG,GAAG,CAAC;AAAA,MAC7D,EAAE,KAAK,IAAI;AAAA,IACb,EACC,KAAK,MAAM;AAEpB,UAAM,OAAO;AAAA,MACX;AAAA,MACA,MAAM;AAAA,MACN;AAAA,MACA;AAAA,MACA,OAAO,MAAM,IAAI;AAAA,MACjB;AAAA,MACA;AAAA,MACA,YAAY,MAAM,KAAK;AAAA,MACvB;AAAA,MACA;AAAA,MACA,mBAAmB,MAAM,YAAY;AAAA,MACrC;AAAA,MACA;AAAA,MACA,sBAAsB,MAAM,YAAY;AAAA,MACxC;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF,EAAE,KAAK,IAAI;AAEX,UAAM,MAAM,MAAM,IAAI,SAAS,EAAE,QAAQ,KAAK,CAAC;AAC/C,WAAO,aAAa,GAAsB;AAAA,EAC5C;AACF;AAIA,SAAS,aAAa,KAAiD;AACrE,MAAI,CAAC,OAAO,OAAO,QAAQ,UAAU;AACnC,UAAM,IAAI,MAAM,8BAA8B;AAAA,EAChD;AACA,QAAM,eAAe,OAAO,IAAI,iBAAiB,WAAW,IAAI,eAAe;AAC/E,QAAM,YAAY,OAAO,IAAI,cAAc,WAAW,IAAI,YAAY;AACtE,QAAM,sBACJ,OAAO,IAAI,wBAAwB,WAAW,IAAI,sBAAsB;AAC1E,MAAI,CAAC,gBAAgB,CAAC,aAAa,CAAC,qBAAqB;AACvD,UAAM,IAAI,MAAM,yCAAyC;AAAA,EAC3D;AACA,MAAI,OAAO,IAAI,mBAAmB,WAAW;AAC3C,UAAM,IAAI,MAAM,yCAAyC;AAAA,EAC3D;AACA,QAAM,gBAAgB,OAAO,IAAI,UAAU;AAC3C,MAAI,CAAC,OAAO,SAAS,aAAa,GAAG;AACnC,UAAM,IAAI,MAAM,gCAAgC;AAAA,EAClD;AACA,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,gBAAgB,IAAI;AAAA,IACpB,YAAY,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,aAAa,CAAC;AAAA,EACpD;AACF;AAEA,SAAS,sBAAsB,GAAyB;AACtD,QAAM,SACJ,QAAQ,EAAE,IAAI,MACb,OAAO,EAAE,UAAU,WAAW,UAAU,EAAE,MAAM,QAAQ,CAAC,CAAC,KAAK,OAC/D,EAAE,iBAAiB,EAAE,cAAc,SAAS,IACzC,aAAa,EAAE,cAAc,KAAK,IAAI,CAAC,MACvC;AACN,QAAM,UAAU,EAAE,YAAY,SAAY,KAAK;AAAA,EAAK,SAAS,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;AACtF,SAAO,SAAS;AAClB;AAEA,SAAS,SAAS,GAAoB;AACpC,MAAI;AACF,WAAO,KAAK,UAAU,GAAG,MAAM,CAAC;AAAA,EAClC,QAAQ;AACN,WAAO,OAAO,CAAC;AAAA,EACjB;AACF;;;ACzdA,IAAMA,gCACJ;AAEF,eAAsB,8BACpB,QAOA;AACA,QAAM,WAAW,OAAO,YAAY;AACpC,QAAM,kBAAkB,OAAO,mBAAmB;AAClD,QAAM,wBAAwB,OAAO,yBAAyB;AAC9D,QAAM,SAAS,OAAO,UAAU,oBAAoB;AACpD,QAAM,sBAAsB,OAAO,uBAAuBA;AAC1D,QAAM,+BACJ,OAAO,gCAAgC;AACzC,MAAI,sBAAsB;AAE1B,MAAI,UAAqD;AAAA,IACvD,MAAM;AAAA,IACN,OAAO,OAAO;AAAA,IACd,aAAa;AAAA,IACb,cAAc,EAAE,MAAM,MAAM;AAAA,IAC5B,QAAQ,MAAM,OAAO,KAAK;AAAA,IAC1B,WAAW;AAAA,IACX,iBAAiB;AAAA,EACnB;AAEA,SAAO,oBAAoB;AAAA,IACzB,QAAQ,OAAO;AAAA,IACf,QAAQ,EAAE,UAAU,UAAU,WAAW,OAAO,UAAU;AAAA,IAC1D,OAAO,OAAO;AAAA,IACd,YAAY,OAAO,cAAc;AAAA,IACjC,WAAW,OAAO;AAAA,IAClB,WAAW,OAAO;AAAA,IAClB,eAAe,OAAO,iBAAiB;AAAA,IACvC,SAAS,MAAM;AAAA,IACf,UAAU,CAAC,EAAE,MAAM,MAAM;AAAA,MACvB,cAAc;AAAA,QACZ,IAAI;AAAA,QACJ,QAAQ,MAAM,aAAa;AAAA,QAC3B,OAAO,MAAM,aAAa;AAAA,QAC1B,UAAU;AAAA,QACV,QAAQ,MAAM,aAAa,OACvB,wBACA,sBAAsB,MAAM,aAAa,eAAe,SAAS,KAAK,MAAM,aAAa,cAAc,KAAK,IAAI,CAAC,KAAK,EAAE;AAAA,MAC9H,CAAC;AAAA,IACH;AAAA,IACA,YAAY,CAAC,EAAE,MAAM,MAAM;AACzB,UAAI,MAAM,aAAa,MAAM;AAC3B,eAAO;AAAA,UACL,MAAM;AAAA,UACN,MAAM;AAAA,UACN,QAAQ;AAAA,UACR,OAAO,MAAM,aAAa;AAAA,QAC5B;AAAA,MACF;AACA,UAAI,MAAM,WAAW;AACnB,eAAO;AAAA,UACL,MAAM;AAAA,UACN,MAAM;AAAA,UACN,QAAQ;AAAA,UACR,OAAO,MAAM,aAAa;AAAA,UAC1B,cAAc,6BAA6B,MAAM,YAAY;AAAA,QAC/D;AAAA,MACF;AACA,aAAO;AAAA,QACL,MAAM;AAAA,QACN,MAAM;AAAA,QACN,QAAQ;AAAA,QACR,OAAO,MAAM,aAAa;AAAA,MAC5B;AAAA,IACF;AAAA,IACA,QAAQ,CAAC,EAAE,MAAM,OAAO;AAAA,MACtB,MAAM;AAAA,MACN,QAAQ,EAAE,MAAM,uBAAuB,MAAM,MAAM,OAAO,EAAE;AAAA,MAC5D,QAAQ,MAAM,aAAa,uBAAuB;AAAA,IACpD;AAAA,IACA,KAAK,OAAO,QAAQ,QAAQ;AAC1B,YAAM,OAAO,OAAO;AACpB,YAAM,aAA4C,MAAM,OAAO,QAAQ;AAAA,QACrE;AAAA,QACA,MAAM,OAAO;AAAA,QACb,OAAO,QAAQ;AAAA,QACf,aAAa,QAAQ;AAAA,QACrB,aAAa,IAAI;AAAA,QACjB,SAAS,IAAI;AAAA,MACf,CAAC;AAED,YAAM,YAAY,WAAW;AAC7B,YAAM,eAAe,MAAM,OAAO,OAAO,SAAS;AAClD,UAAI,SAAwB;AAC5B,UAAI,kBAAkB;AACtB,UAAI;AACJ,UAAI,iBAAiB,CAAC,aAAa;AAEnC,UAAI,CAAC,aAAa,MAAM;AACtB,YAAI;AACF,mBAAS,MAAM,OAAO,OAAO;AAAA,YAC3B;AAAA,YACA,MAAM,OAAO;AAAA,YACb,OAAO;AAAA,YACP;AAAA,YACA,cAAc,WAAW;AAAA,YACzB,QAAQ,MAAM,OAAO,KAAK;AAAA,UAC5B,CAAC;AACD,4BAAkB;AAClB,2BAAiB,OAAO;AACxB,gCAAsB,OAAO,cAAc,kBAAkB,sBAAsB,IAAI;AACvF,cAAI,wBAAwB,KAAK,uBAAuB;AACtD,6BAAiB;AAAA,QACrB,SAAS,KAAK;AACZ,wBAAc,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC7D,mBAAS,QAAQ,eAAe;AAAA,YAC9B,cAAc;AAAA,YACd,WAAW;AAAA,YACX,qBAAqB;AAAA,YACrB,gBAAgB;AAAA,YAChB,YAAY;AAAA,UACd;AACA,2BAAiB;AAAA,QACnB;AAAA,MACF,OAAO;AACL,iBAAS;AAAA,UACP,cAAc;AAAA,UACd,WAAW;AAAA,UACX,qBAAqB;AAAA,UACrB,gBAAgB;AAAA,UAChB,YAAY;AAAA,QACd;AAAA,MACF;AAEA,YAAM,QAA2B;AAAA,QAC/B,GAAI,UAAU;AAAA,UACZ,cAAc;AAAA,UACd,WAAW;AAAA,UACX,qBAAqB;AAAA,UACrB;AAAA,UACA,YAAY;AAAA,QACd;AAAA,QACA;AAAA,QACA,WAAW,KAAK,IAAI;AAAA,QACpB,cAAc;AAAA,UACZ,MAAM,aAAa;AAAA,UACnB,OAAO,aAAa;AAAA,UACpB,eAAe,aAAa;AAAA,QAC9B;AAAA,MACF;AACA,YAAM,OAAO,OAAO,KAAK;AAEzB,gBAAU;AAAA,QACR;AAAA,QACA,OAAO;AAAA,QACP,aAAa;AAAA,QACb;AAAA,QACA,cAAc,WAAW;AAAA,QACzB,QAAQ,MAAM,OAAO,KAAK;AAAA,QAC1B,WAAW,aAAa,QAAQ,CAAC;AAAA,QACjC;AAAA,QACA;AAAA,MACF;AAEA,aAAO;AAAA,QACL,OAAO;AAAA,QACP;AAAA,QACA,cAAc,WAAW;AAAA,QACzB;AAAA,QACA;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAAA,EACF,CAAC;AACH;AAEO,SAAS,oCACd,cAC0B;AAC1B,MAAI,aAAa,KAAM,QAAO;AAC9B,SAAO,aAAa,eAAe,SAAS,0BAA0B;AACxE;","names":["DEFAULT_FALLBACK_INSTRUCTION"]}

package/dist/cli.js

@@ -5,9 +5,10 @@ import {
   runRpcBatch,
   runRpcOnce,
   startServer
-} from "./chunk-6KQG5HAH.js";
-import "./chunk-KAO3Q65R.js";
-import "./chunk-SQQLHODJ.js";
+} from "./chunk-SY6WAAAD.js";
+import "./chunk-4S4BM3QQ.js";
+import "./chunk-PC4UYEBM.js";
+import "./chunk-NG236HPC.js";
 import "./chunk-PZ5AY32C.js";
 
 // src/cli.ts
@@ -78,19 +79,22 @@ async function main() {
     case "openapi": {
       const out = flags.out ?? "openapi.json";
       const spec = buildOpenApi(handleVersion().version);
-      writeFileSync(out, JSON.stringify(spec, null, 2) + "\n", "utf-8");
+      writeFileSync(out, `${JSON.stringify(spec, null, 2)}
+`, "utf-8");
       console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`);
       return 0;
     }
     case "version": {
-      process.stdout.write(JSON.stringify(handleVersion(), null, 2) + "\n");
+      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}
+`);
       return 0;
     }
     case "help":
     case "--help":
     case "-h":
     case "":
-      process.stdout.write(HELP + "\n");
+      process.stdout.write(`${HELP}
+`);
       return 0;
     default:
       process.stderr.write(`unknown command: ${command}

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/cli.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\n\nimport { buildOpenApi } from './wire/openapi'\nimport { handleVersion } from './wire/handlers'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServer } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]\n    if (tok.startsWith('--')) {\n      const key = tok.slice(2)\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval — wire-protocol entry point.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Method ∈ {judge, listRubrics, version}.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  version\n        Print server + wire-protocol version JSON.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n\n  switch (command) {\n    case 'serve': {\n      const port = Number(flags.port ?? 5005)\n      const host = flags.host ?? '127.0.0.1'\n      const server = startServer({ port, host })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      return await runRpcOnce(method)\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      return await runRpcBatch(method)\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, JSON.stringify(spec, null, 2) + '\\n', 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'version': {\n      process.stdout.write(JSON.stringify(handleVersion(), null, 2) + '\\n')\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(HELP + '\\n')\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;AAaA,SAAS,qBAAqB;AAa9B,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAiBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AAEtE,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,OAAO,MAAM,QAAQ,IAAI;AACtC,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,SAAS,YAAY,EAAE,MAAM,KAAK,CAAC;AAEzC,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,WAAW,MAAM;AAAA,IAChC;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,YAAY,MAAM;AAAA,IACjC;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,KAAK,UAAU,MAAM,MAAM,CAAC,IAAI,MAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,IAAI,IAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,OAAO,IAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}
+{"version":3,"sources":["../src/cli.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServer } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const key = tok.slice(2)\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval — wire-protocol entry point.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Method ∈ {judge, listRubrics, version}.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  version\n        Print server + wire-protocol version JSON.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n\n  switch (command) {\n    case 'serve': {\n      const port = Number(flags.port ?? 5005)\n      const host = flags.host ?? '127.0.0.1'\n      const server = startServer({ port, host })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      return await runRpcOnce(method)\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      return await runRpcBatch(method)\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;AAY9B,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAiBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AAEtE,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,OAAO,MAAM,QAAQ,IAAI;AACtC,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,SAAS,YAAY,EAAE,MAAM,KAAK,CAAC;AAEzC,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,WAAW,MAAM;AAAA,IAChC;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,aAAO,MAAM,YAAY,MAAM;AAAA,IACjC;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}

package/dist/{control-DvkH87qJ.d.ts → control-CBShYYA6.d.ts}

@@ -1,7 +1,31 @@
-import { G as ControlEvalResult, N as ControlRunResult, b as FeedbackLabel, p as ProposedSideEffect, O as ControlRuntimeConfig } from './feedback-trajectory-c43WGtTX.js';
-import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from './run-record-DNiOMBrZ.js';
-import { T as TraceStore, F as FailureClass } from './store-u47QaJ9G.js';
-import { T as TraceEmitter } from './emitter-B2XqDKFU.js';
+import { b as FeedbackLabel, p as ProposedSideEffect } from './feedback-trajectory-DfFdrraJ.js';
+import { C as ControlEvalResult, a as ControlRunResult, h as ControlRuntimeConfig } from './control-runtime-BuJHoLg0.js';
+import { T as TraceEmitter } from './emitter-DP_cSSiw.js';
+import { T as TraceStore, F as FailureClass } from './store-Db2Bv8Cf.js';
+import { a as RunSplitTag, b as RunTokenUsage, R as RunRecord } from './run-record-CqzahIbx.js';
+
+interface ActionExecutionPolicy {
+    allowedTypes?: string[];
+    blockedTypes?: string[];
+    alwaysRequireApprovalTypes?: string[];
+    autoApproveTypes?: string[];
+    requireApprovalForExternalSideEffects?: boolean;
+    requireApprovalAboveCostUsd?: number;
+    maxActionCostUsd?: number;
+    remainingBudgetUsd?: number;
+    expectedOutcomeRequired?: boolean;
+    killCriteriaRequired?: boolean;
+}
+interface ActionPolicyDecision {
+    allowed: boolean;
+    blocked: boolean;
+    requiresApproval: boolean;
+    reasons: string[];
+    label?: FeedbackLabel;
+}
+declare function evaluateActionPolicy(action: ProposedSideEffect, policy?: ActionExecutionPolicy, options?: {
+    createdAt?: string;
+}): ActionPolicyDecision;
 
 interface RunEvidenceMetadata {
     experimentId: string;
@@ -33,29 +57,6 @@ interface ControlRunToRunRecordOptions extends RunEvidenceMetadata {
 declare function controlRunToRunRecord<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult>(run: ControlRunResult<TState, TAction, TActionResult, TEval>, options: ControlRunToRunRecordOptions): RunRecord;
 declare function scoreFromEvals(evals: readonly ControlEvalResult[]): number | undefined;
 
-interface ActionExecutionPolicy {
-    allowedTypes?: string[];
-    blockedTypes?: string[];
-    alwaysRequireApprovalTypes?: string[];
-    autoApproveTypes?: string[];
-    requireApprovalForExternalSideEffects?: boolean;
-    requireApprovalAboveCostUsd?: number;
-    maxActionCostUsd?: number;
-    remainingBudgetUsd?: number;
-    expectedOutcomeRequired?: boolean;
-    killCriteriaRequired?: boolean;
-}
-interface ActionPolicyDecision {
-    allowed: boolean;
-    blocked: boolean;
-    requiresApproval: boolean;
-    reasons: string[];
-    label?: FeedbackLabel;
-}
-declare function evaluateActionPolicy(action: ProposedSideEffect, policy?: ActionExecutionPolicy, options?: {
-    createdAt?: string;
-}): ActionPolicyDecision;
-
 /**
  * Propose / Verify / Review — the core multi-shot primitive.
  *
@@ -196,12 +197,10 @@ interface ProposeReviewReport<State, Summary = unknown> {
 declare function inMemoryReviewStore(initial?: ReviewMemoryEntry[]): ReviewMemoryStore;
 declare function jsonlReviewStore(path: string): ReviewMemoryStore;
 declare function runProposeReview<State, Summary = unknown>(config: ProposeReviewConfig<State, Summary>): Promise<ProposeReviewReport<State, Summary>>;
-interface LlmJsonCall {
-    (req: {
-        system: string;
-        user: string;
-    }): Promise<unknown>;
-}
+type LlmJsonCall = (req: {
+    system: string;
+    user: string;
+}) => Promise<unknown>;
 interface LlmReviewerConfig<State, Summary = unknown> {
     callJson: LlmJsonCall;
     renderState?: (state: State) => string;

package/dist/control-runtime-BuJHoLg0.d.ts

@@ -0,0 +1,180 @@
+import { T as TraceEmitter } from './emitter-DP_cSSiw.js';
+import { F as FailureClass, T as TraceStore } from './store-Db2Bv8Cf.js';
+
+/**
+ * Policy-based agent control runtime.
+ *
+ * This is the minimal reusable loop behind driver-agent patterns:
+ *
+ *   observe state -> validate -> decide next action -> act -> observe -> ...
+ *
+ * It deliberately does not model named "topologies". Direct execution,
+ * critic/revise, driver intervention, specialist calls, and human escalation
+ * are all just actions chosen by the control policy.
+ */
+
+type ControlSeverity = 'info' | 'warning' | 'error' | 'critical';
+type ControlActionFailureMode = 'continue' | 'stop';
+interface ControlEvalResult {
+    /** Stable validator or judge id. */
+    id: string;
+    /** Whether this check passed. */
+    passed: boolean;
+    /** Optional normalized score. 1 = best, 0 = worst. */
+    score?: number;
+    /** Objective validators should usually be "error" or "critical" when failed. */
+    severity?: ControlSeverity;
+    /** Human-readable result. */
+    detail?: string;
+    /** Small evidence string or pointer. Avoid large payloads. */
+    evidence?: string;
+    /** True when the result came from deterministic state, not LLM judgment. */
+    objective?: boolean;
+    /** Structured details for downstream control policies and reports. */
+    metadata?: Record<string, unknown>;
+}
+interface ControlBudget {
+    maxSteps: number;
+    maxWallMs?: number;
+    maxCostUsd?: number;
+}
+interface ControlStopPolicies<TState, TAction> {
+    /**
+     * Stop after N consecutive steps with no state fingerprint change and
+     * less than `minScoreDelta` score movement. Disabled when omitted.
+     */
+    maxNoProgressSteps?: number;
+    /**
+     * Stop after the same action fingerprint is selected N consecutive
+     * times. Disabled when omitted.
+     */
+    maxRepeatedActions?: number;
+    /** Minimum score movement that counts as progress. Default 0.001. */
+    minScoreDelta?: number;
+    /** Override the default JSON/string fingerprint for state comparisons. */
+    stateFingerprint?: (state: TState) => string;
+    /** Override the default JSON/string fingerprint for repeated-action checks. */
+    actionFingerprint?: (action: TAction) => string;
+}
+interface ControlContext<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
+    intent: string;
+    state: TState;
+    evals: TEval[];
+    history: ControlStep<TState, TAction, TActionResult, TEval>[];
+    budget: ControlBudget;
+    stepIndex: number;
+    wallMs: number;
+    spentCostUsd: number;
+    remainingCostUsd?: number;
+    abortSignal: AbortSignal;
+    emitter?: TraceEmitter;
+}
+type ControlDecision<TAction> = {
+    type: 'continue';
+    action: TAction;
+    reason?: string;
+} | {
+    type: 'stop';
+    reason: string;
+    pass?: boolean;
+    score?: number;
+};
+interface StopDecision {
+    stop: boolean;
+    pass: boolean;
+    reason: string;
+    score?: number;
+    failureClass?: FailureClass;
+}
+interface ControlActionOutcome<TActionResult> {
+    ok: boolean;
+    result?: TActionResult;
+    error?: string;
+    costUsd?: number;
+    durationMs: number;
+}
+interface ControlRuntimeError {
+    phase: 'observe' | 'validate' | 'decide' | 'act' | 'stop-policy' | 'on-step' | 'trace';
+    stepIndex: number;
+    message: string;
+}
+interface ControlStep<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
+    index: number;
+    decision: ControlDecision<TAction>;
+    beforeState: TState;
+    afterState: TState;
+    evalsBefore: TEval[];
+    evalsAfter: TEval[];
+    actionOutcome?: ControlActionOutcome<TActionResult>;
+    startedAt: string;
+    endedAt: string;
+}
+interface ControlRunResult<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
+    intent: string;
+    pass: boolean;
+    completed: boolean;
+    reason: string;
+    score?: number;
+    steps: ControlStep<TState, TAction, TActionResult, TEval>[];
+    finalState: TState | undefined;
+    finalEvals: TEval[];
+    wallMs: number;
+    spentCostUsd: number;
+    runId: string | null;
+    failureClass?: FailureClass;
+    runtimeErrors: ControlRuntimeError[];
+    stoppedBy: 'policy' | 'stop-policy' | 'budget' | 'abort' | 'runtime-error';
+}
+interface ControlRuntimeConfig<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult> {
+    intent: string;
+    budget?: Partial<ControlBudget>;
+    signal?: AbortSignal;
+    /** Defaults to `continue`: action failures are recorded, then the policy gets another chance. */
+    actionFailure?: ControlActionFailureMode;
+    /**
+     * Extract cost from an action result. Used for `maxCostUsd` budget
+     * enforcement and trace budget ledger emission.
+     */
+    getActionCostUsd?: (ctx: {
+        action: TAction;
+        result: TActionResult;
+        state: TState;
+        evals: TEval[];
+        history: ControlStep<TState, TAction, TActionResult, TEval>[];
+    }) => number | undefined;
+    /** Read typed task/product state. Prefer structured state over transcript-only context. */
+    observe: (ctx: {
+        history: ControlStep<TState, TAction, TActionResult, TEval>[];
+        abortSignal: AbortSignal;
+    }) => Promise<TState> | TState;
+    /** Objective validators first, subjective judges only where objective state is insufficient. */
+    validate: (ctx: {
+        intent: string;
+        state: TState;
+        history: ControlStep<TState, TAction, TActionResult, TEval>[];
+        abortSignal: AbortSignal;
+    }) => Promise<TEval[]> | TEval[];
+    /** Choose the next control action. Can call a worker, ask user, run critic, inspect state, or stop. */
+    decide: (ctx: ControlContext<TState, TAction, TActionResult, TEval>) => Promise<ControlDecision<TAction>> | ControlDecision<TAction>;
+    /** Execute the action selected by the policy. */
+    act: (action: TAction, ctx: ControlContext<TState, TAction, TActionResult, TEval>) => Promise<TActionResult> | TActionResult;
+    /** Final stopping policy. Called before decide and after each action. */
+    shouldStop?: (ctx: ControlContext<TState, TAction, TActionResult, TEval>) => Promise<StopDecision> | StopDecision;
+    /** Optional hook for tracing or live progress updates. */
+    onStep?: (step: ControlStep<TState, TAction, TActionResult, TEval>) => Promise<void> | void;
+    /** Optional generic stuck-loop policies. Custom `shouldStop` still runs first. */
+    stopPolicies?: ControlStopPolicies<TState, TAction>;
+    /** Optional trace sink. Emits one run plus one span per control step. */
+    store?: TraceStore;
+    scenarioId?: string;
+    projectId?: string;
+    variantId?: string;
+}
+declare function runAgentControlLoop<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult>(config: ControlRuntimeConfig<TState, TAction, TActionResult, TEval>): Promise<ControlRunResult<TState, TAction, TActionResult, TEval>>;
+declare function stopOnNoProgress<TState, TAction>(maxNoProgressSteps: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxNoProgressSteps'>): ControlStopPolicies<TState, TAction>;
+declare function stopOnRepeatedAction<TState, TAction>(maxRepeatedActions: number, options?: Omit<ControlStopPolicies<TState, TAction>, 'maxRepeatedActions'>): ControlStopPolicies<TState, TAction>;
+declare function objectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
+declare function subjectiveEval(input: Omit<ControlEvalResult, 'objective'>): ControlEvalResult;
+declare function allCriticalPassed(evals: ControlEvalResult[]): boolean;
+
+export { type ControlEvalResult as C, type StopDecision as S, type ControlRunResult as a, type ControlStep as b, type ControlActionFailureMode as c, type ControlActionOutcome as d, type ControlBudget as e, type ControlContext as f, type ControlDecision as g, type ControlRuntimeConfig as h, type ControlRuntimeError as i, type ControlSeverity as j, type ControlStopPolicies as k, allCriticalPassed as l, stopOnRepeatedAction as m, subjectiveEval as n, objectiveEval as o, runAgentControlLoop as r, stopOnNoProgress as s };

package/dist/control.d.ts

@@ -1,6 +1,8 @@
-export { H as ControlActionFailureMode, J as ControlActionOutcome, K as ControlBudget, L as ControlContext, M as ControlDecision, G as ControlEvalResult, N as ControlRunResult, O as ControlRuntimeConfig, Q as ControlRuntimeError, E as ControlSeverity, R as ControlStep, S as ControlStopPolicies, T as StopDecision, U as allCriticalPassed, V as objectiveEval, W as runAgentControlLoop, X as stopOnNoProgress, Y as stopOnRepeatedAction, Z as subjectiveEval } from './feedback-trajectory-c43WGtTX.js';
-export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, p as RunEvidenceMetadata, s as controlRunToRunRecord, u as evaluateActionPolicy, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-DvkH87qJ.js';
-import './dataset-B9qvlm_o.js';
-import './emitter-B2XqDKFU.js';
-import './store-u47QaJ9G.js';
-import './run-record-DNiOMBrZ.js';
+export { A as ActionExecutionPolicy, a as ActionPolicyDecision, C as ControlRunToRunRecordOptions, e as ProposeReviewConfig, f as ProposeReviewControlAction, g as ProposeReviewControlConfig, h as ProposeReviewControlResult, i as ProposeReviewControlState, j as ProposeReviewReport, p as RunEvidenceMetadata, s as controlRunToRunRecord, u as evaluateActionPolicy, x as runProposeReview, y as runProposeReviewAsControlLoop, z as scoreFromEvals } from './control-CBShYYA6.js';
+export { c as ControlActionFailureMode, d as ControlActionOutcome, e as ControlBudget, f as ControlContext, g as ControlDecision, C as ControlEvalResult, a as ControlRunResult, h as ControlRuntimeConfig, i as ControlRuntimeError, j as ControlSeverity, b as ControlStep, k as ControlStopPolicies, S as StopDecision, l as allCriticalPassed, o as objectiveEval, r as runAgentControlLoop, s as stopOnNoProgress, m as stopOnRepeatedAction, n as subjectiveEval } from './control-runtime-BuJHoLg0.js';
+import './feedback-trajectory-DfFdrraJ.js';
+import './dataset-CiK_3LDr.js';
+import './errors-BZ9sTdz7.js';
+import './emitter-DP_cSSiw.js';
+import './store-Db2Bv8Cf.js';
+import './run-record-CqzahIbx.js';

package/dist/control.js

@@ -1,18 +1,21 @@
 import {
-  allCriticalPassed,
   controlRunToRunRecord,
   evaluateActionPolicy,
-  objectiveEval,
-  runAgentControlLoop,
   runProposeReview,
   runProposeReviewAsControlLoop,
-  scoreFromEvals,
+  scoreFromEvals
+} from "./chunk-ZN274SWR.js";
+import {
+  allCriticalPassed,
+  objectiveEval,
+  runAgentControlLoop,
   stopOnNoProgress,
   stopOnRepeatedAction,
   subjectiveEval
-} from "./chunk-V5QSWN7L.js";
-import "./chunk-QBW3YBTR.js";
-import "./chunk-5IIQKMD5.js";
+} from "./chunk-LSH4MMOZ.js";
+import "./chunk-NLMNWKVM.js";
+import "./chunk-TVVP3ZZQ.js";
+import "./chunk-NG236HPC.js";
 import "./chunk-PZ5AY32C.js";
 export {
   allCriticalPassed,

package/dist/{dataset-B9qvlm_o.d.ts → dataset-CiK_3LDr.d.ts}

@@ -1,3 +1,5 @@
+import { V as ValidationError } from './errors-BZ9sTdz7.js';
+
 /**
  * Dataset — versioned, sliceable, content-hashed scenario collection.
  *
@@ -64,8 +66,9 @@ interface SliceOptions {
     /** If true, include scenarios marked as holdout. Default false. */
     includeHoldout?: boolean;
 }
+
 /** Locked holdouts — throws on mutate. Callers that need a mutable dataset fork it. */
-declare class HoldoutLockedError extends Error {
+declare class HoldoutLockedError extends ValidationError {
     constructor(datasetName: string);
 }
 declare class Dataset {
@@ -109,4 +112,4 @@ declare class Dataset {
 }
 declare function hashScenarios(scenarios: DatasetScenario[]): Promise<string>;
 
-export { type DatasetSplit as D, HoldoutLockedError as H, type SliceOptions as S, type DatasetScenario as a, Dataset as b, type DatasetManifest as c, type DatasetDifficulty as d, type DatasetProvenance as e, hashScenarios as h };
+export { type DatasetSplit as D, HoldoutLockedError as H, type SliceOptions as S, type DatasetScenario as a, type DatasetManifest as b, Dataset as c, type DatasetDifficulty as d, type DatasetProvenance as e, hashScenarios as h };

package/dist/{emitter-B2XqDKFU.d.ts → emitter-DP_cSSiw.d.ts}

@@ -1,4 +1,4 @@
-import { T as TraceStore, R as RunOutcome, a as Run, S as Span, b as SpanKind, L as LlmSpan, c as ToolSpan, d as RetrievalSpan, J as JudgeSpan, e as SandboxSpan, E as EventKind, f as TraceEvent, B as BudgetLedgerEntry, A as Artifact, M as Message } from './store-u47QaJ9G.js';
+import { T as TraceStore, c as RunOutcome, R as Run, S as Span, d as SpanKind, L as LlmSpan, a as ToolSpan, e as RetrievalSpan, J as JudgeSpan, f as SandboxSpan, E as EventKind, b as TraceEvent, B as BudgetLedgerEntry, A as Artifact, M as Message } from './store-Db2Bv8Cf.js';
 
 /**
  * TraceEmitter — hierarchical span builder that auto-parents using an

package/dist/errors-BZ9sTdz7.d.ts

@@ -0,0 +1,70 @@
+/**
+ * Error taxonomy for `@tangle-network/agent-eval`.
+ *
+ * Every error this package throws as part of its *public contract* extends
+ * `AgentEvalError`. Consumers can pattern-match by `instanceof <Subclass>` or
+ * by the stable string `code` carried on the base class.
+ *
+ * The codes are stable across minor versions; new codes can be added, but
+ * existing codes never change meaning. New subclasses are non-breaking.
+ *
+ * Internal invariant guards (`throw new Error('this should never happen')`)
+ * remain plain `Error`s on purpose — they're programmer-mistake assertions,
+ * not consumer-catchable contract failures.
+ */
+type AgentEvalErrorCode = 'validation' | 'not_found' | 'config' | 'capture_integrity' | 'judge' | 'verification' | 'replay';
+declare class AgentEvalError extends Error {
+    /** Stable string code. Survives minification; safe to switch on. */
+    readonly code: AgentEvalErrorCode;
+    constructor(code: AgentEvalErrorCode, message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** Caller passed invalid arguments (out of range, mutually-exclusive options, bad shape). */
+declare class ValidationError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** A named resource (run, span, rubric, scenario, dataset row, route) does not exist. */
+declare class NotFoundError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** Configuration missing or malformed (`HOME` unset, required image not supplied, env var absent). */
+declare class ConfigError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/**
+ * A run is missing the artifacts a launch-grade check requires:
+ * raw HTTP capture absent, no LLM spans, route assertion failed, run-end
+ * assertion tripped. Block ship on this; do not catch and move on.
+ */
+declare class CaptureIntegrityError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** A judge call failed in a way that's not retryable: schema parse failure, bad rubric, conflicting dimensions. */
+declare class JudgeError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** A verifier signalled a hard failure (compile, test, schema) — distinct from a low judge score. */
+declare class VerificationError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+/** Replay cache cannot satisfy a request: miss with no fallback, sink lacks list(), unsupported URL. */
+declare class ReplayError extends AgentEvalError {
+    constructor(message: string, options?: {
+        cause?: unknown;
+    });
+}
+
+export { AgentEvalError as A, CaptureIntegrityError as C, JudgeError as J, NotFoundError as N, ReplayError as R, ValidationError as V, type AgentEvalErrorCode as a, ConfigError as b, VerificationError as c };