@tangle-network/agent-runtime 0.68.0 → 0.70.0

package/dist/chunk-BYZCXQHF.js

@@ -0,0 +1,474 @@
+import {
+  createKbGate
+} from "./chunk-FNMGYYSS.js";
+import {
+  detachedSessionDelegate
+} from "./chunk-L5ZFBVT6.js";
+import {
+  ConfigError,
+  createExecutorRegistry,
+  definePersona,
+  runPersonified,
+  worktreeFanout
+} from "./chunk-QXWGSDAQ.js";
+
+// src/loop-runner.ts
+import {
+  selfImprove
+} from "@tangle-network/agent-eval/contract";
+
+// src/analyst-loop/iterations-to-trace-store.ts
+import {
+  DEFAULT_TRACE_ANALYST_BUDGETS,
+  TRACE_ANALYST_TRUNCATION_MARKER_PREFIX
+} from "@tangle-network/agent-eval";
+
+// src/analyst-loop/run-analyst-loop.ts
+import { diffFindings } from "@tangle-network/agent-eval";
+async function runAnalystLoop(opts) {
+  const log = opts.log ?? defaultLog;
+  const strategy = opts.priorFindingsStrategy ?? "per-kind";
+  const emit = makeEmitter(opts.onEvent);
+  const startedAt = Date.now();
+  const baselineRunId = resolveBaselineRunId(opts);
+  const priorAll = baselineRunId ? opts.findingsStore?.loadRun(baselineRunId) ?? [] : [];
+  log("baseline resolved", { baselineRunId, prior_findings: priorAll.length });
+  await emit({
+    type: "baseline-resolved",
+    runId: opts.runId,
+    baselineRunId,
+    priorFindingCount: priorAll.length
+  });
+  const priorFindings = buildPriorFindingsInput(priorAll, strategy, opts.registry.list());
+  const analystResult = await runRegistry(opts, priorFindings, emit);
+  log("analyst run complete", {
+    findings: analystResult.findings.length,
+    cost_usd: analystResult.total_cost_usd,
+    per_analyst: analystResult.per_analyst.map((s) => ({
+      id: s.analyst_id,
+      status: s.status,
+      n: s.findings_count
+    }))
+  });
+  if (opts.findingsStore && analystResult.findings.length > 0) {
+    await opts.findingsStore.append(opts.runId, analystResult.findings);
+    await emit({
+      type: "findings-persisted",
+      runId: opts.runId,
+      count: analystResult.findings.length
+    });
+  }
+  let diff = null;
+  if (baselineRunId && analystResult.findings.length > 0) {
+    diff = diffFindings(
+      priorAll.map((f) => ({ ...f })),
+      analystResult.findings.map((f) => ({ ...f, run_id: opts.runId }))
+    );
+    log("diff vs baseline", {
+      appeared: diff.appeared.length,
+      disappeared: diff.disappeared.length,
+      persisted: diff.persisted.length,
+      changed: diff.changed.length
+    });
+    await emit({
+      type: "diff-computed",
+      runId: opts.runId,
+      baselineRunId,
+      appeared: diff.appeared.length,
+      disappeared: diff.disappeared.length,
+      persisted: diff.persisted.length,
+      changed: diff.changed.length
+    });
+  }
+  let knowledge = null;
+  if (opts.knowledgeAdapter) {
+    knowledge = await runKnowledgeAdapter(opts, analystResult.findings, log, emit);
+  }
+  let improvement = null;
+  if (opts.improvementAdapter) {
+    improvement = await runImprovementAdapter(opts, analystResult.findings, log, emit);
+  }
+  await emit({
+    type: "loop-completed",
+    runId: opts.runId,
+    durationMs: Date.now() - startedAt
+  });
+  return {
+    runId: opts.runId,
+    baselineRunId,
+    analystResult,
+    diff,
+    knowledge,
+    improvement
+  };
+}
+function makeEmitter(onEvent) {
+  if (!onEvent) return async () => {
+  };
+  return async (event) => {
+    await onEvent(event);
+  };
+}
+async function runRegistry(opts, priorFindings, emit) {
+  const reg = opts.registry;
+  if (typeof reg.runStream === "function" && opts.onEvent) {
+    let final = null;
+    for await (const ev of reg.runStream(opts.runId, opts.inputs, { priorFindings })) {
+      await emit({ type: "analyst", runId: opts.runId, event: ev });
+      if (ev.type === "run-completed") final = ev.result;
+    }
+    if (!final) {
+      throw new Error("runAnalystLoop: registry.runStream ended without run-completed event");
+    }
+    return final;
+  }
+  return opts.registry.run(opts.runId, opts.inputs, { priorFindings });
+}
+function resolveBaselineRunId(opts) {
+  if (opts.baselineRunId === null) return null;
+  if (typeof opts.baselineRunId === "string") return opts.baselineRunId;
+  if (!opts.findingsStore) return null;
+  const all = opts.findingsStore.loadAll();
+  let last = null;
+  for (const row of all) {
+    if (row.run_id === opts.runId) continue;
+    last = row.run_id;
+  }
+  return last;
+}
+function buildPriorFindingsInput(prior, strategy, registry) {
+  if (strategy === "none" || prior.length === 0) return void 0;
+  const stripped = prior.map(({ run_id: _run_id, ...rest }) => rest);
+  if (strategy === "wildcard") {
+    return { "*": stripped };
+  }
+  void registry;
+  return stripped;
+}
+async function runKnowledgeAdapter(opts, findings, log, emit) {
+  const adapter = opts.knowledgeAdapter;
+  const batch = await adapter.proposeFromFindings(findings);
+  log("knowledge.proposeFromFindings", {
+    proposals: batch.proposals.length,
+    skipped: batch.skipped,
+    errors: batch.errors.length
+  });
+  await emit({
+    type: "knowledge-proposed",
+    runId: opts.runId,
+    proposalCount: batch.proposals.length,
+    skipped: batch.skipped,
+    errors: batch.errors.length
+  });
+  const auto = opts.autoApply?.knowledge ?? false;
+  const threshold = opts.autoApply?.knowledgeConfidenceThreshold ?? 0.85;
+  if (!auto || !adapter.apply) {
+    await emit({
+      type: "knowledge-applied",
+      runId: opts.runId,
+      writtenCount: 0,
+      withheldForReview: batch.proposals.length
+    });
+    return {
+      proposals: batch.proposals,
+      applied: [],
+      skipped: batch.skipped,
+      errors: batch.errors,
+      withheld_for_review: batch.proposals.length
+    };
+  }
+  const findingsById = new Map(findings.map((f) => [f.finding_id, f]));
+  const safe = [];
+  let withheld = 0;
+  for (const p of batch.proposals) {
+    const src = p.sourceFindingId ? findingsById.get(p.sourceFindingId) : void 0;
+    if (!src) {
+      withheld += 1;
+      continue;
+    }
+    if (src.confidence < threshold) {
+      withheld += 1;
+      continue;
+    }
+    safe.push(p);
+  }
+  const result = await adapter.apply(safe);
+  log("knowledge.apply", {
+    applied: result.written.length,
+    withheld_for_review: withheld,
+    warnings: result.warnings.length
+  });
+  await emit({
+    type: "knowledge-applied",
+    runId: opts.runId,
+    writtenCount: result.written.length,
+    withheldForReview: withheld
+  });
+  return {
+    proposals: batch.proposals,
+    applied: result.written,
+    skipped: batch.skipped,
+    errors: batch.errors,
+    withheld_for_review: withheld
+  };
+}
+async function runImprovementAdapter(opts, findings, log, emit) {
+  const adapter = opts.improvementAdapter;
+  const batch = await adapter.proposeFromFindings(findings);
+  log("improvement.proposeFromFindings", {
+    edits: batch.edits.length,
+    skipped: batch.skipped,
+    errors: batch.errors.length
+  });
+  await emit({
+    type: "improvement-proposed",
+    runId: opts.runId,
+    editCount: batch.edits.length,
+    skipped: batch.skipped,
+    errors: batch.errors.length
+  });
+  const auto = opts.autoApply?.improvement ?? false;
+  const threshold = opts.autoApply?.improvementConfidenceThreshold ?? 0.9;
+  if (!auto || !adapter.apply) {
+    await emit({
+      type: "improvement-applied",
+      runId: opts.runId,
+      appliedCount: 0,
+      withheldForReview: batch.edits.length
+    });
+    return {
+      edits: batch.edits,
+      applied: [],
+      skipped: batch.skipped,
+      errors: batch.errors,
+      withheld_for_review: batch.edits.length
+    };
+  }
+  const findingsById = new Map(findings.map((f) => [f.finding_id, f]));
+  const safe = [];
+  let withheld = 0;
+  for (const e of batch.edits) {
+    const src = e.sourceFindingId ? findingsById.get(e.sourceFindingId) : void 0;
+    if (!src || src.confidence < threshold) {
+      withheld += 1;
+      continue;
+    }
+    safe.push(e);
+  }
+  const result = await adapter.apply(safe);
+  log("improvement.apply", {
+    applied: result.applied.length,
+    withheld_for_review: withheld,
+    warnings: result.warnings.length
+  });
+  await emit({
+    type: "improvement-applied",
+    runId: opts.runId,
+    appliedCount: result.applied.length,
+    withheldForReview: withheld
+  });
+  return {
+    edits: batch.edits,
+    applied: result.applied,
+    skipped: batch.skipped,
+    errors: batch.errors,
+    withheld_for_review: withheld
+  };
+}
+function defaultLog(msg, fields) {
+  if (fields) console.log(`[analyst-loop] ${msg}`, fields);
+  else console.log(`[analyst-loop] ${msg}`);
+}
+
+// src/loop-runner.ts
+var DELEGATED_LOOP_MODES = ["code", "review", "research", "audit", "self-improve"];
+function isDelegatedLoopMode(value) {
+  return typeof value === "string" && DELEGATED_LOOP_MODES.includes(value);
+}
+async function runDelegatedLoop(mode, registry, options = {}) {
+  const runner = registry[mode];
+  if (!runner) {
+    throw new ConfigError(
+      `runDelegatedLoop: no runner registered for mode '${mode}' (registered: ${Object.keys(registry).join(", ") || "none"})`
+    );
+  }
+  const now = options.now ?? Date.now;
+  const signal = options.signal ?? new AbortController().signal;
+  const start = now();
+  try {
+    const output = await runner(signal);
+    return { mode, ok: true, output, durationMs: now() - start };
+  } catch (err) {
+    return {
+      mode,
+      ok: false,
+      error: err instanceof Error ? err.message : String(err),
+      durationMs: now() - start
+    };
+  }
+}
+function coderLoopRunner(options) {
+  const delegate = detachedSessionDelegate({
+    sandboxClient: options.sandboxClient,
+    ...options.reviewer ? { reviewer: options.reviewer } : {},
+    ...options.winnerSelection ? { winnerSelection: options.winnerSelection } : {},
+    ...options.fanoutHarnesses ? { fanoutHarnesses: options.fanoutHarnesses } : {}
+  });
+  return async (signal) => {
+    const ctx = { signal, report: () => {
+    } };
+    return delegate(options.args, ctx);
+  };
+}
+function worktreeLoopRunner(options) {
+  const shape = worktreeFanout({
+    repoRoot: options.repoRoot,
+    taskPrompt: options.taskPrompt,
+    harnesses: options.harnesses,
+    ...options.testCmd !== void 0 ? { testCmd: options.testCmd } : {},
+    ...options.typecheckCmd !== void 0 ? { typecheckCmd: options.typecheckCmd } : {},
+    ...options.require !== void 0 ? { require: options.require } : {},
+    ...options.maxDiffLines !== void 0 ? { maxDiffLines: options.maxDiffLines } : {},
+    ...options.forbiddenPaths !== void 0 ? { forbiddenPaths: options.forbiddenPaths } : {},
+    ...options.winnerStrategy !== void 0 ? { winnerStrategy: options.winnerStrategy } : {},
+    ...options.runGit ? { runGit: options.runGit } : {},
+    ...options.runHarness ? { runHarness: options.runHarness } : {},
+    ...options.runCommand ? { runCommand: options.runCommand } : {}
+  });
+  const persona = definePersona({
+    name: "worktree-coder",
+    root: { profile: { name: "worktree-coder" }, harness: null },
+    directive: "deliver a minimal validated patch on a fresh worktree",
+    context: { role: "coder" },
+    executors: { registry: createExecutorRegistry() }
+  });
+  return async (signal) => {
+    const result = await runPersonified({
+      persona,
+      shape,
+      task: options.taskPrompt,
+      budget: options.budget,
+      signal
+    });
+    if (result.kind !== "winner" || result.out.kind !== "done") {
+      const blockers = result.kind === "winner" && result.out.kind === "blocked" ? result.out.blockers.join("; ") : `supervisor settled ${result.kind}`;
+      throw new Error(`worktreeLoopRunner: no delivered patch (${blockers})`);
+    }
+    return result.out.deliverable;
+  };
+}
+function researchLoopRunner(o) {
+  const gate = createKbGate(o.gate);
+  const maxRounds = Math.max(1, Math.trunc(o.maxRounds ?? 1));
+  return async (signal) => {
+    const accepted = [];
+    let vetoed = [];
+    let rounds = 0;
+    for (let round = 0; round < maxRounds; round += 1) {
+      if (signal.aborted) break;
+      rounds += 1;
+      const candidates = await o.research(round, vetoed);
+      if (candidates.length === 0) break;
+      vetoed = [];
+      for (const c of candidates) {
+        const v = await gate(c);
+        if (v.accepted) accepted.push(c);
+        else vetoed.push({ candidate: c, vetoedBy: v.vetoedBy, reason: v.reason });
+      }
+      if (vetoed.length === 0) break;
+    }
+    return { accepted, vetoed, rounds };
+  };
+}
+function selfImproveLoopRunner(options) {
+  return async () => selfImprove(options);
+}
+function auditLoopRunner(options) {
+  return async () => runAnalystLoop(options);
+}
+
+// src/loop-runner-bin.ts
+async function runLoopRunnerCli(args) {
+  if (!isDelegatedLoopMode(args.mode)) {
+    return {
+      exitCode: 2,
+      error: `unknown mode '${args.mode}' (expected one of: ${DELEGATED_LOOP_MODES.join(", ")})`
+    };
+  }
+  let registry;
+  try {
+    registry = await args.loadRegistry();
+  } catch (err) {
+    return { exitCode: 2, error: `failed to load registry: ${errMsg(err)}` };
+  }
+  if (!registry[args.mode]) {
+    return {
+      exitCode: 2,
+      error: `config registers no runner for mode '${args.mode}' (registered: ${Object.keys(registry).join(", ") || "none"})`
+    };
+  }
+  const result = await runDelegatedLoop(args.mode, registry, {
+    ...args.now ? { now: args.now } : {}
+  });
+  return { exitCode: result.ok ? 0 : 1, result };
+}
+function parseLoopRunnerArgv(argv) {
+  const out = {};
+  for (let i = 0; i < argv.length; i += 1) {
+    const a = argv[i];
+    if (a === "--mode") out.mode = argv[++i];
+    else if (a === "--config") out.config = argv[++i];
+    else if (a?.startsWith("--mode=")) out.mode = a.slice("--mode=".length);
+    else if (a?.startsWith("--config=")) out.config = a.slice("--config=".length);
+  }
+  return out;
+}
+function resolveRegistry(mod) {
+  const def = mod?.default ?? mod;
+  const value = typeof def === "function" ? def() : def;
+  return value;
+}
+function errMsg(err) {
+  return err instanceof Error ? err.message : String(err);
+}
+async function main() {
+  const { mode, config } = parseLoopRunnerArgv(process.argv.slice(2));
+  if (!mode || !config) {
+    process.stderr.write(
+      `usage: agent-runtime-loop --mode <mode> --config <module>
+  modes: ${DELEGATED_LOOP_MODES.join(" | ")}
+  config: a JS/TS module default-exporting a DelegatedLoopRegistry (or a factory)
+`
+    );
+    process.exit(2);
+  }
+  const { pathToFileURL } = await import("url");
+  const { resolve } = await import("path");
+  const cli = await runLoopRunnerCli({
+    mode,
+    loadRegistry: async () => resolveRegistry(await import(pathToFileURL(resolve(config)).href))
+  });
+  process.stdout.write(`${JSON.stringify(cli.result ?? { error: cli.error }, null, 2)}
+`);
+  if (cli.error) process.stderr.write(`${cli.error}
+`);
+  process.exit(cli.exitCode);
+}
+var invokedScript = typeof process !== "undefined" ? process.argv?.[1] : void 0;
+if (invokedScript && /loop-runner-bin\.(js|ts|mjs)$/.test(invokedScript)) {
+  void main();
+}
+
+export {
+  DELEGATED_LOOP_MODES,
+  isDelegatedLoopMode,
+  runDelegatedLoop,
+  coderLoopRunner,
+  worktreeLoopRunner,
+  researchLoopRunner,
+  selfImproveLoopRunner,
+  auditLoopRunner,
+  runLoopRunnerCli,
+  parseLoopRunnerArgv
+};
+//# sourceMappingURL=chunk-BYZCXQHF.js.map

package/dist/chunk-BYZCXQHF.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/loop-runner.ts","../src/analyst-loop/iterations-to-trace-store.ts","../src/analyst-loop/run-analyst-loop.ts","../src/loop-runner-bin.ts"],"sourcesContent":["/**\n * @experimental\n *\n * `runDelegatedLoop` — the configured delegated loop-runner.\n *\n * One typed entrypoint a worker agent (or a scheduled routine) calls to run a\n * disciplined loop in a chosen MODE, over agent-runtime's hardened engines:\n *\n *   code         → build-in-a-loop via the coder delegate (no-op + secret floor,\n *                  optional reviewer gate, winner-selection)\n *   review       → code mode with a REQUIRED reviewer (the gate is the point)\n *   research     → research-in-a-loop with valid-only KB growth (createKbGate)\n *   audit        → analyze trace/run data → findings (runAnalystLoop, caller-wired)\n *   self-improve → closed-loop text/config optimization (selfImprove, held-out gated)\n *\n * It is intentionally a thin façade: the value is that EVERY product reuses the\n * one hardened engine instead of forking delegation logic. The dispatcher owns\n * mode routing, timing, fail-loud on an unregistered mode, and a uniform result\n * shape; each mode's engine is a pre-configured runner in the registry (build it\n * with the factories below, or inject your own / a stub).\n */\n\nimport type { Scenario } from '@tangle-network/agent-eval/campaign'\nimport {\n  type SelfImproveOptions,\n  type SelfImproveResult,\n  selfImprove,\n} from '@tangle-network/agent-eval/contract'\nimport { runAnalystLoop } from './analyst-loop'\nimport type { RunAnalystLoopOpts, RunAnalystLoopResult } from './analyst-loop/types'\nimport { ConfigError } from './errors'\nimport {\n  type CoderReviewer,\n  type DelegateRunCtx,\n  type DetachedWinnerSelection,\n  detachedSessionDelegate,\n} from './mcp/delegates'\nimport type { CoderOutput } from './mcp/detached-coder'\nimport { type CreateKbGateOptions, createKbGate, type FactCandidate } from './mcp/kb-gate'\nimport type { DelegateCodeArgs } from './mcp/types'\nimport {\n  type AuthoredHarness,\n  type Budget,\n  createExecutorRegistry,\n  definePersona,\n  runPersonified,\n  type SandboxClient,\n  type WinnerStrategy,\n  type WorktreeFanoutOptions,\n  type WorktreePatchArtifact,\n  worktreeFanout,\n} from './runtime'\n\n/** @experimental Every delegated-loop mode, for validation + CLI surfaces. */\nexport const DELEGATED_LOOP_MODES = ['code', 'review', 'research', 'audit', 'self-improve'] as const\n\n/** @experimental */\nexport type DelegatedLoopMode = (typeof DELEGATED_LOOP_MODES)[number]\n\n/** @experimental Type guard for an untrusted mode string (CLI / config input). */\nexport function isDelegatedLoopMode(value: unknown): value is DelegatedLoopMode {\n  return typeof value === 'string' && (DELEGATED_LOOP_MODES as readonly string[]).includes(value)\n}\n\n/** @experimental A pre-configured loop for one mode. Returns the mode's raw\n *  output; the dispatcher wraps it in a {@link DelegatedLoopResult}. */\nexport type DelegatedLoopRunner<T = unknown> = (signal: AbortSignal) => Promise<T>\n\n/** @experimental Mode → configured runner. Partial: only register the modes a\n *  given product/routine actually uses. */\nexport type DelegatedLoopRegistry = Partial<Record<DelegatedLoopMode, DelegatedLoopRunner>>\n\n/** @experimental Uniform result — never throws from a registered runner; a\n *  thrown engine becomes `{ ok: false, error }` so a routine can record + move on. */\nexport interface DelegatedLoopResult<T = unknown> {\n  mode: DelegatedLoopMode\n  ok: boolean\n  output?: T\n  error?: string\n  durationMs: number\n}\n\n/** @experimental */\nexport interface RunDelegatedLoopOptions {\n  signal?: AbortSignal\n  /** Clock override for deterministic tests. */\n  now?: () => number\n}\n\n/**\n * @experimental\n *\n * Dispatch a configured loop by mode. Fails loud (throws `ConfigError`) when no\n * runner is registered for the mode — a routine pointed at an unwired mode is a\n * config bug, not a silent no-op. A runner that throws is captured as\n * `{ ok: false }` so unattended runs record the failure rather than crash.\n */\nexport async function runDelegatedLoop<T = unknown>(\n  mode: DelegatedLoopMode,\n  registry: DelegatedLoopRegistry,\n  options: RunDelegatedLoopOptions = {},\n): Promise<DelegatedLoopResult<T>> {\n  const runner = registry[mode] as DelegatedLoopRunner<T> | undefined\n  if (!runner) {\n    throw new ConfigError(\n      `runDelegatedLoop: no runner registered for mode '${mode}' (registered: ${\n        Object.keys(registry).join(', ') || 'none'\n      })`,\n    )\n  }\n  const now = options.now ?? Date.now\n  const signal = options.signal ?? new AbortController().signal\n  const start = now()\n  try {\n    const output = await runner(signal)\n    return { mode, ok: true, output, durationMs: now() - start }\n  } catch (err) {\n    return {\n      mode,\n      ok: false,\n      error: err instanceof Error ? err.message : String(err),\n      durationMs: now() - start,\n    }\n  }\n}\n\n/** @experimental Options for the default `code`/`review` runner. */\nexport interface CoderLoopRunnerOptions {\n  sandboxClient: SandboxClient\n  /** What to build — the delegate args (goal, repoRoot, variants, config, …). */\n  args: DelegateCodeArgs\n  /** Adversarial reviewer. Pass one to run `review` mode (an approval gate over the candidate). */\n  reviewer?: CoderReviewer\n  /** Winner-selection strategy. Default `highest-score`. */\n  winnerSelection?: DetachedWinnerSelection\n  /** Harnesses for `variants > 1` fanout. */\n  fanoutHarnesses?: string[]\n}\n\n/**\n * @experimental Build a `code`/`review`-mode runner over the sandbox-session coder delegate. Pass a\n * `reviewer` to run `review` mode — an approval gate over the validated candidate.\n */\nexport function coderLoopRunner(options: CoderLoopRunnerOptions): DelegatedLoopRunner<CoderOutput> {\n  const delegate = detachedSessionDelegate({\n    sandboxClient: options.sandboxClient,\n    ...(options.reviewer ? { reviewer: options.reviewer } : {}),\n    ...(options.winnerSelection ? { winnerSelection: options.winnerSelection } : {}),\n    ...(options.fanoutHarnesses ? { fanoutHarnesses: options.fanoutHarnesses } : {}),\n  })\n  return async (signal) => {\n    const ctx: DelegateRunCtx = { signal, report: () => {} }\n    return delegate(options.args, ctx)\n  }\n}\n\n/** @experimental Options for the local-repo `code` runner over the GENERIC recursive path. */\nexport interface WorktreeLoopRunnerOptions {\n  /** Absolute path to the local git checkout each worktree is cut from. */\n  repoRoot: string\n  /** The instruction handed to every authored harness (composed under each profile's systemPrompt). */\n  taskPrompt: string\n  /** The supervisor-authored harness profiles — one fanout item (one worktree-CLI leaf) each. */\n  harnesses: ReadonlyArray<AuthoredHarness>\n  /** Conserved budget pool bounding the fanout (equal-k holds by construction). */\n  budget: Budget\n  /** Shell command run in each worktree to derive the tests-PASS signal. */\n  testCmd?: string\n  /** Shell command run in each worktree to derive the typecheck-PASS signal. */\n  typecheckCmd?: string\n  /** Which verification signals the deliverable REQUIRES present-and-passing (default none). */\n  require?: ReadonlyArray<'tests' | 'typecheck'>\n  /** Diff-size cap (lines). */\n  maxDiffLines?: number\n  /** Literal path prefixes the patch must not touch (the secret-floor is always on regardless). */\n  forbiddenPaths?: string[]\n  /** Winner-selection strategy among gated candidates. Default `highest-score`. */\n  winnerStrategy?: WinnerStrategy\n  /** Test seams forwarded to the worktree-CLI leaves so the runner drives offline. */\n  runGit?: WorktreeFanoutOptions['runGit']\n  runHarness?: WorktreeFanoutOptions['runHarness']\n  runCommand?: WorktreeFanoutOptions['runCommand']\n}\n\n/**\n * @experimental\n *\n * `code` mode on the GENERIC recursive path: author one `AgentProfile` per harness, run them as a\n * `worktreeFanout` (N `createWorktreeCliExecutor` leaves, each `gateOnDeliverable`) through\n * `runPersonified` on the keystone Supervisor. This is the local-repo counterpart to\n * {@link coderLoopRunner} (which drives the in-box harness over a `SandboxClient`): no `runLoop`\n * driver, no role-coupled delegate — the harness list is the fanout, the gate is `patchDelivered`,\n * the winner is the shared valid-only selector (NOT `defaultSelectWinner`, whose non-valid fallback\n * would surface an ungated patch). Equal-k holds by the conserved budget pool. Returns the winning\n * patch artifact, or throws when no candidate is delivered (fail loud, never a vacuous done).\n */\nexport function worktreeLoopRunner(\n  options: WorktreeLoopRunnerOptions,\n): DelegatedLoopRunner<WorktreePatchArtifact> {\n  const shape = worktreeFanout<string>({\n    repoRoot: options.repoRoot,\n    taskPrompt: options.taskPrompt,\n    harnesses: options.harnesses,\n    ...(options.testCmd !== undefined ? { testCmd: options.testCmd } : {}),\n    ...(options.typecheckCmd !== undefined ? { typecheckCmd: options.typecheckCmd } : {}),\n    ...(options.require !== undefined ? { require: options.require } : {}),\n    ...(options.maxDiffLines !== undefined ? { maxDiffLines: options.maxDiffLines } : {}),\n    ...(options.forbiddenPaths !== undefined ? { forbiddenPaths: options.forbiddenPaths } : {}),\n    ...(options.winnerStrategy !== undefined ? { winnerStrategy: options.winnerStrategy } : {}),\n    ...(options.runGit ? { runGit: options.runGit } : {}),\n    ...(options.runHarness ? { runHarness: options.runHarness } : {}),\n    ...(options.runCommand ? { runCommand: options.runCommand } : {}),\n  })\n  // The persona's only role here is to carry the fanout shape onto the Supervisor; each item's\n  // executor is BYO (the gated worktree-CLI leaf), so the registry only needs to pass BYO through.\n  const persona = definePersona<WorktreePatchArtifact>({\n    name: 'worktree-coder',\n    root: { profile: { name: 'worktree-coder' }, harness: null },\n    directive: 'deliver a minimal validated patch on a fresh worktree',\n    context: { role: 'coder' },\n    executors: { registry: createExecutorRegistry() },\n  })\n  return async (signal) => {\n    const result = await runPersonified<string, WorktreePatchArtifact>({\n      persona,\n      shape,\n      task: options.taskPrompt,\n      budget: options.budget,\n      signal,\n    })\n    if (result.kind !== 'winner' || result.out.kind !== 'done') {\n      const blockers =\n        result.kind === 'winner' && result.out.kind === 'blocked'\n          ? result.out.blockers.join('; ')\n          : `supervisor settled ${result.kind}`\n      throw new Error(`worktreeLoopRunner: no delivered patch (${blockers})`)\n    }\n    return result.out.deliverable\n  }\n}\n\n/** @experimental A fact rejected at the KB gate — surfaced, never dropped. */\nexport interface VetoedFact {\n  candidate: FactCandidate\n  vetoedBy?: string\n  reason?: string\n}\n\n/** @experimental */\nexport interface ResearchLoopResult {\n  /** Facts that passed the fail-closed gate — safe to write to the KB. */\n  accepted: FactCandidate[]\n  /** Facts the gate vetoed in the final round — escalate, do not silently drop. */\n  vetoed: VetoedFact[]\n  /** Research rounds actually run. */\n  rounds: number\n}\n\n/** @experimental Options for the default `research` runner. */\nexport interface ResearchLoopRunnerOptions {\n  /**\n   * The research engine (the consumer's web/doc searcher + extractor). Called\n   * each round with the prior round's vetoes so it can re-research the gaps.\n   * Returns fact candidates carrying their grounding (`verbatimPassage` +\n   * `sourceText`).\n   */\n  research: (round: number, vetoed: VetoedFact[]) => Promise<FactCandidate[]>\n  /** Gate config (extra judges, self-artifact kinds, …). The floor is always on. */\n  gate?: CreateKbGateOptions\n  /** Max research rounds (correct-on-veto remediation). Default 1. */\n  maxRounds?: number\n}\n\n/**\n * @experimental `research` mode — research-in-a-loop with valid-only KB growth.\n *\n * Each round: research → gate every candidate (fail-closed; passage MUST be in\n * the source) → accept the clean ones → re-research the vetoed ones next round,\n * up to `maxRounds`. Vetoed facts in the final round are RETURNED (escalate,\n * never silently dropped) so the caller audits vs retries.\n */\nexport function researchLoopRunner(\n  o: ResearchLoopRunnerOptions,\n): DelegatedLoopRunner<ResearchLoopResult> {\n  const gate = createKbGate(o.gate)\n  const maxRounds = Math.max(1, Math.trunc(o.maxRounds ?? 1))\n  return async (signal) => {\n    const accepted: FactCandidate[] = []\n    let vetoed: VetoedFact[] = []\n    let rounds = 0\n    for (let round = 0; round < maxRounds; round += 1) {\n      if (signal.aborted) break\n      rounds += 1\n      const candidates = await o.research(round, vetoed)\n      if (candidates.length === 0) break\n      vetoed = []\n      for (const c of candidates) {\n        const v = await gate(c)\n        if (v.accepted) accepted.push(c)\n        else vetoed.push({ candidate: c, vetoedBy: v.vetoedBy, reason: v.reason })\n      }\n      if (vetoed.length === 0) break\n    }\n    return { accepted, vetoed, rounds }\n  }\n}\n\n/** @experimental `self-improve` mode — agent-eval's one-call closed loop (held-out gated). */\nexport function selfImproveLoopRunner<TScenario extends Scenario, TArtifact>(\n  options: SelfImproveOptions<TScenario, TArtifact>,\n): DelegatedLoopRunner<SelfImproveResult<TScenario, TArtifact>> {\n  return async () => selfImprove<TScenario, TArtifact>(options)\n}\n\n/** @experimental `audit` mode — analyst loop over captured trace/run data. */\nexport function auditLoopRunner<TProposal = unknown, TEdit = unknown>(\n  options: RunAnalystLoopOpts,\n): DelegatedLoopRunner<RunAnalystLoopResult<TProposal, TEdit>> {\n  return async () => runAnalystLoop<TProposal, TEdit>(options)\n}\n","/**\n * @experimental\n *\n * The read seam that closes the autonomous loop: project a round's `Iteration[]`\n * (each carrying its raw `SandboxEvent[]`) into an in-memory `TraceAnalysisStore`, the\n * read interface the trace analysts query. `runAnalystLoop` has had zero consumers\n * because nothing turned a loop's iterations into a store — this is that bridge.\n *\n * One iteration → one trace. The iteration is the root AGENT span; each `SandboxEvent`\n * becomes a child span (LLM for llm_call events, TOOL for tool events, else SPAN), so an\n * analyst can walk a shot's trace, cluster its errors, and emit findings the driver\n * steers on. Projection is best-effort over the FLAT SandboxEvent shape (no per-event\n * lineage yet — that's the richer-trace gap); it never fabricates — an errored iteration\n * surfaces a real ERROR span carrying the real message.\n */\n\nimport {\n  type DatasetOverview,\n  DEFAULT_TRACE_ANALYST_BUDGETS,\n  type QueryTracesPage,\n  type SearchSpanResult,\n  type SearchTraceResult,\n  type SpanMatchRecord,\n  TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,\n  type TraceAnalysisStore,\n  type TraceAnalystByteBudgets,\n  type TraceAnalystFilters,\n  type TraceAnalystSpan,\n  type TraceAnalystSpanKind,\n  type TraceAnalystTraceSummary,\n  type ViewSpansResult,\n  type ViewTraceResult,\n} from '@tangle-network/agent-eval'\nimport type { SandboxEvent } from '@tangle-network/sandbox'\nimport { AnalystError } from '../errors'\nimport { extractLlmCallEvent } from '../runtime/sandbox-events'\nimport type { Iteration } from '../runtime/types'\n\n/** ErrorCluster isn't re-exported from the agent-eval root; derive it from the overview. */\ntype ErrorCluster = DatasetOverview['error_clusters'][number]\n\ninterface ProjectedTrace {\n  summary: TraceAnalystTraceSummary\n  spans: TraceAnalystSpan[]\n  /** raw JSONL bytes of this trace's spans — the byte-budget accounting unit. */\n  rawBytes: number\n}\n\nconst bytesOf = (v: unknown): number => Buffer.byteLength(JSON.stringify(v) ?? '', 'utf8')\nconst iso = (ms: number): string => new Date(ms).toISOString()\n\n/** Normalize volatile tokens out of a status message so semantically identical failures\n *  collapse to one signature (digits, hex/uuids, paths, durations → placeholders). */\nfunction normalizeSignature(message: string): string {\n  return message\n    .replace(/0x[0-9a-fA-F]+/g, 'HEX')\n    .replace(/[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}/g, 'UUID')\n    .replace(/(\\/[\\w.-]+){2,}/g, 'PATH')\n    .replace(/\\b\\d+(\\.\\d+)?(ms|s|m|h)\\b/g, 'DUR')\n    .replace(/\\b\\d+\\b/g, '#')\n    .replace(/\\s+/g, ' ')\n    .trim()\n    .slice(0, 200)\n}\n\nfunction spanKindFor(\n  event: SandboxEvent,\n  agentRunName: string,\n): {\n  kind: TraceAnalystSpanKind\n  model: string | null\n  tool: string | null\n} {\n  const llm = extractLlmCallEvent(event, agentRunName)\n  if (llm) return { kind: 'LLM', model: llm.model ?? null, tool: null }\n  const type = String(event?.type ?? '')\n  if (/tool/i.test(type)) {\n    const d = event?.data as { name?: unknown; tool?: unknown } | undefined\n    const tool = typeof d?.name === 'string' ? d.name : typeof d?.tool === 'string' ? d.tool : type\n    return { kind: 'TOOL', model: null, tool }\n  }\n  return { kind: 'SPAN', model: null, tool: null }\n}\n\nfunction errorMessageOf(event: SandboxEvent): string | undefined {\n  const type = String(event?.type ?? '')\n  const d = event?.data as { error?: unknown; message?: unknown } | undefined\n  if (/error|fail/i.test(type) || d?.error) {\n    const m = d?.error ?? d?.message\n    return typeof m === 'string' ? m : `${type} error`\n  }\n  return undefined\n}\n\n/** Project one iteration → one trace: root AGENT span + a child span per event. */\nfunction projectIteration<Task, Output>(iter: Iteration<Task, Output>): ProjectedTrace {\n  const traceId =\n    (\n      iter.events.find((e) => (e?.data as { sandboxId?: string } | undefined)?.sandboxId)?.data as\n        | { sandboxId?: string }\n        | undefined\n    )?.sandboxId ?? `iter-${iter.index}`\n  const start = iso(iter.startedAt)\n  const end = iso(iter.endedAt || iter.startedAt)\n  const durationMs = Math.max(0, (iter.endedAt || iter.startedAt) - iter.startedAt)\n  const rootId = `${traceId}:root`\n  const iterErrored = Boolean(iter.error) || iter.verdict?.valid === false\n  const spans: TraceAnalystSpan[] = [\n    {\n      trace_id: traceId,\n      span_id: rootId,\n      parent_span_id: null,\n      name: iter.agentRunName,\n      kind: 'AGENT',\n      start_time: start,\n      end_time: end,\n      duration_ms: durationMs,\n      status: iter.error ? 'ERROR' : 'OK',\n      status_message: iter.error?.message,\n      service_name: 'agent-runtime',\n      agent_name: iter.agentRunName,\n      model_name: null,\n      tool_name: null,\n      attributes: {\n        'iteration.index': iter.index,\n        'verdict.valid': iter.verdict?.valid,\n        'verdict.score': iter.verdict?.score,\n        'output.preview':\n          iter.output === undefined ? undefined : String(iter.output).slice(0, 2000),\n      },\n    },\n  ]\n  const models = new Set<string>()\n  const tools = new Set<string>()\n  iter.events.forEach((event, i) => {\n    const { kind, model, tool } = spanKindFor(event, iter.agentRunName)\n    if (model) models.add(model)\n    if (tool) tools.add(tool)\n    const errMsg = errorMessageOf(event)\n    spans.push({\n      trace_id: traceId,\n      span_id: `${traceId}:e${i}`,\n      parent_span_id: rootId,\n      name: String(event?.type ?? 'event'),\n      kind,\n      start_time: start,\n      end_time: end,\n      duration_ms: 0,\n      status: errMsg ? 'ERROR' : 'OK',\n      status_message: errMsg,\n      service_name: 'agent-runtime',\n      agent_name: iter.agentRunName,\n      model_name: model,\n      tool_name: tool,\n      attributes: (event?.data as Record<string, unknown> | undefined) ?? {},\n    })\n  })\n  const hasErrors = spans.some((s) => s.status === 'ERROR') || iterErrored\n  const summary: TraceAnalystTraceSummary = {\n    trace_id: traceId,\n    service_name: 'agent-runtime',\n    agent_name: iter.agentRunName,\n    span_count: spans.length,\n    has_errors: hasErrors,\n    start_time: start,\n    end_time: end,\n    duration_ms: durationMs,\n    raw_jsonl_bytes: bytesOf(spans),\n    models: [...models],\n    tools: [...tools],\n  }\n  return { summary, spans, rawBytes: summary.raw_jsonl_bytes }\n}\n\nfunction matchesFilters(t: ProjectedTrace, f?: TraceAnalystFilters): boolean {\n  if (!f) return true\n  if (f.has_errors !== undefined && t.summary.has_errors !== f.has_errors) return false\n  if (f.service_names?.length && !f.service_names.includes(t.summary.service_name ?? ''))\n    return false\n  if (f.agent_names?.length && !f.agent_names.includes(t.summary.agent_name ?? '')) return false\n  if (f.model_names?.length && !f.model_names.some((m) => t.summary.models.includes(m)))\n    return false\n  if (f.tool_names?.length && !f.tool_names.some((tn) => t.summary.tools.includes(tn))) return false\n  if (f.start_time_after && t.summary.start_time < f.start_time_after) return false\n  if (f.start_time_before && t.summary.start_time > f.start_time_before) return false\n  if (f.regex_pattern && !new RegExp(f.regex_pattern).test(JSON.stringify(t.spans))) return false\n  return true\n}\n\nfunction capAttributes(\n  attributes: Record<string, unknown>,\n  perAttrCap: number,\n): { capped: Record<string, unknown>; truncated: number } {\n  let truncated = 0\n  const capped: Record<string, unknown> = {}\n  for (const [k, v] of Object.entries(attributes)) {\n    const s = typeof v === 'string' ? v : JSON.stringify(v)\n    if (typeof s === 'string' && s.length > perAttrCap) {\n      truncated += 1\n      capped[k] = `${TRACE_ANALYST_TRUNCATION_MARKER_PREFIX} ${s.length}b]${s.slice(0, perAttrCap)}`\n    } else {\n      capped[k] = v\n    }\n  }\n  return { capped, truncated }\n}\n\n/**\n * Build an in-memory `TraceAnalysisStore` over a loop round's iterations. Fail-loud on an\n * empty round — there is nothing for an analyst to read, and a silent empty store would\n * mask a broken capture path.\n */\nexport function iterationsToTraceStore<Task, Output>(\n  iterations: ReadonlyArray<Iteration<Task, Output>>,\n  budgets: TraceAnalystByteBudgets = DEFAULT_TRACE_ANALYST_BUDGETS,\n): TraceAnalysisStore {\n  if (iterations.length === 0) {\n    throw new AnalystError('iterationsToTraceStore: no iterations to analyze (empty round)')\n  }\n  const traces = iterations.map((it) => projectIteration(it))\n  const byId = new Map(traces.map((t) => [t.summary.trace_id, t]))\n\n  const buildClusters = (set: ProjectedTrace[]): ErrorCluster[] => {\n    const map = new Map<string, ErrorCluster>()\n    for (const t of set) {\n      for (const s of t.spans) {\n        if (s.status !== 'ERROR' || !s.status_message) continue\n        const sig = normalizeSignature(s.status_message)\n        const c = map.get(sig) ?? {\n          signature: sig,\n          status_message_sample: s.status_message,\n          span_name: s.name,\n          tool_name: s.tool_name,\n          trace_count: 0,\n          span_count: 0,\n          prevalence: 0,\n          exemplar_trace_ids: [],\n          exemplar_span_ids: [],\n        }\n        c.span_count += 1\n        if (\n          !c.exemplar_trace_ids.includes(t.summary.trace_id) &&\n          c.exemplar_trace_ids.length < 10\n        ) {\n          c.exemplar_trace_ids.push(t.summary.trace_id)\n          c.trace_count += 1\n        }\n        if (c.exemplar_span_ids.length < 10) c.exemplar_span_ids.push(s.span_id)\n        map.set(sig, c)\n      }\n    }\n    const errorTraces = set.filter((t) => t.summary.has_errors).length || 1\n    const clusters = [...map.values()].map((c) => ({\n      ...c,\n      prevalence: c.trace_count / errorTraces,\n    }))\n    return clusters.sort((a, b) => b.trace_count - a.trace_count)\n  }\n\n  return {\n    async getOverview(filters?: TraceAnalystFilters): Promise<DatasetOverview> {\n      const set = traces.filter((t) => matchesFilters(t, filters))\n      const services = new Set<string>()\n      const agents = new Set<string>()\n      const models = new Set<string>()\n      const tools = new Set<string>()\n      let errorSpans = 0\n      for (const t of set) {\n        if (t.summary.service_name) services.add(t.summary.service_name)\n        if (t.summary.agent_name) agents.add(t.summary.agent_name)\n        for (const m of t.summary.models) models.add(m)\n        for (const tn of t.summary.tools) tools.add(tn)\n        errorSpans += t.spans.filter((s) => s.status === 'ERROR').length\n      }\n      const times = set.map((t) => t.summary.start_time).sort()\n      return {\n        total_traces: set.length,\n        raw_jsonl_bytes: set.reduce((n, t) => n + t.rawBytes, 0),\n        services: [...services],\n        agents: [...agents],\n        models: [...models],\n        tool_names: [...tools],\n        sample_trace_ids: set.slice(0, 20).map((t) => t.summary.trace_id),\n        errors: {\n          trace_count: set.filter((t) => t.summary.has_errors).length,\n          span_count: errorSpans,\n        },\n        error_clusters: buildClusters(set),\n        time_range: times.length ? { earliest: times[0]!, latest: times[times.length - 1]! } : null,\n      }\n    },\n\n    async queryTraces(opts): Promise<QueryTracesPage> {\n      const set = traces.filter((t) => matchesFilters(t, opts.filters))\n      const offset = opts.offset ?? 0\n      const page = set.slice(offset, offset + opts.limit)\n      return {\n        traces: page.map((t) => t.summary),\n        total: set.length,\n        has_more: offset + opts.limit < set.length,\n      }\n    },\n\n    async countTraces(filters?: TraceAnalystFilters): Promise<number> {\n      return traces.filter((t) => matchesFilters(t, filters)).length\n    },\n\n    async viewTrace(opts): Promise<ViewTraceResult> {\n      const t = byId.get(opts.trace_id)\n      if (!t) return { trace_id: opts.trace_id, spans: [] }\n      const cap = opts.per_attribute_byte_cap ?? budgets.perAttributeViewBudget\n      const projected = t.spans.map((s) => ({\n        ...s,\n        attributes: capAttributes(s.attributes, cap).capped,\n      }))\n      if (bytesOf(projected) > budgets.perCallByteCeiling) {\n        const names = new Map<string, number>()\n        for (const s of t.spans) names.set(s.name, (names.get(s.name) ?? 0) + 1)\n        return {\n          trace_id: opts.trace_id,\n          oversized: {\n            span_count: t.spans.length,\n            top_span_names: [...names.entries()].sort((a, b) => b[1] - a[1]).slice(0, 20),\n            span_response_bytes_max: Math.max(...t.spans.map((s) => bytesOf(s))),\n            error_span_count: t.spans.filter((s) => s.status === 'ERROR').length,\n          },\n        }\n      }\n      return { trace_id: opts.trace_id, spans: projected }\n    },\n\n    async viewSpans(opts): Promise<ViewSpansResult> {\n      const t = byId.get(opts.trace_id)\n      const cap = opts.per_attribute_byte_cap ?? budgets.perAttributeSpanBudget\n      const want = new Set(opts.span_ids)\n      const found = (t?.spans ?? []).filter((s) => want.has(s.span_id))\n      let truncated = 0\n      const spans = found.map((s) => {\n        const { capped, truncated: n } = capAttributes(s.attributes, cap)\n        truncated += n\n        return { ...s, attributes: capped }\n      })\n      const foundIds = new Set(found.map((s) => s.span_id))\n      return {\n        trace_id: opts.trace_id,\n        spans,\n        missing_span_ids: opts.span_ids.filter((id) => !foundIds.has(id)),\n        truncated_attribute_count: truncated,\n      }\n    },\n\n    async searchTrace(opts): Promise<SearchTraceResult> {\n      const t = byId.get(opts.trace_id)\n      const max = opts.max_matches ?? 50\n      const hits: SpanMatchRecord[] = []\n      for (const s of t?.spans ?? []) {\n        for (const hit of searchSpanAttrs(s, opts.regex_pattern, budgets.perMatchTextBudget)) {\n          if (hits.length >= max) break\n          hits.push(hit)\n        }\n      }\n      return {\n        trace_id: opts.trace_id,\n        hits,\n        total_matches: hits.length,\n        has_more: hits.length >= max,\n      }\n    },\n\n    async searchSpan(opts): Promise<SearchSpanResult> {\n      const t = byId.get(opts.trace_id)\n      const max = opts.max_matches ?? 50\n      const span = (t?.spans ?? []).find((s) => s.span_id === opts.span_id)\n      const hits = span\n        ? searchSpanAttrs(span, opts.regex_pattern, budgets.perMatchTextBudget).slice(0, max)\n        : []\n      return {\n        trace_id: opts.trace_id,\n        span_id: opts.span_id,\n        hits,\n        total_matches: hits.length,\n        has_more: false,\n      }\n    },\n  }\n}\n\nfunction searchSpanAttrs(\n  span: TraceAnalystSpan,\n  pattern: string,\n  textCap: number,\n): SpanMatchRecord[] {\n  const re = new RegExp(pattern, 'g')\n  const hits: SpanMatchRecord[] = []\n  for (const [k, v] of Object.entries(span.attributes)) {\n    const text = typeof v === 'string' ? v : JSON.stringify(v)\n    if (typeof text !== 'string') continue\n    re.lastIndex = 0\n    const m = re.exec(text)\n    if (!m) continue\n    const at = m.index\n    hits.push({\n      trace_id: span.trace_id,\n      span_id: span.span_id,\n      span_name: span.name,\n      span_kind: span.kind,\n      attribute_path: `attributes.${k}`,\n      matched_text: m[0].slice(0, textCap),\n      context_before: text.slice(Math.max(0, at - textCap / 2), at),\n      context_after: text.slice(at + m[0].length, at + m[0].length + textCap / 2),\n      match_offset: at,\n    })\n  }\n  return hits\n}\n","/**\n * `runAnalystLoop` — the one call agent apps reach for to close the\n * recursive-self-improvement loop.\n *\n *   1. Load baseline findings (last run, or the slice the caller specifies)\n *   2. Run the analyst registry with priorFindings injected\n *   3. Persist the new run's findings to the ledger\n *   4. Diff the new run against the baseline\n *   5. Hand the findings to the knowledge adapter → proposals (and\n *      optionally apply them) → wiki edits\n *   6. Hand the findings to the improvement adapter → prompt / tool /\n *      scaffolding edits (review-only by default)\n *   7. Return a single report the consumer renders / persists / acts on.\n *\n * Adapters are optional: the loop works as a \"run + diff + report\"\n * primitive when no adapters are wired; it closes end-to-end when\n * both adapters are wired.\n */\n\nimport type { AnalystFinding, AnalystRunResult, FindingsDiff } from '@tangle-network/agent-eval'\nimport { diffFindings } from '@tangle-network/agent-eval'\n\nimport type {\n  AnalystLoopEvent,\n  AnalystRegistryStreamingLike,\n  ImprovementReport,\n  KnowledgeReport,\n  RunAnalystLoopOpts,\n  RunAnalystLoopResult,\n} from './types'\n\nexport async function runAnalystLoop<TProposal = unknown, TEdit = unknown>(\n  opts: RunAnalystLoopOpts,\n): Promise<RunAnalystLoopResult<TProposal, TEdit>> {\n  const log = opts.log ?? defaultLog\n  const strategy = opts.priorFindingsStrategy ?? 'per-kind'\n  const emit = makeEmitter(opts.onEvent)\n  const startedAt = Date.now()\n\n  // 1. Resolve baseline + load prior findings.\n  const baselineRunId = resolveBaselineRunId(opts)\n  const priorAll: ReadonlyArray<AnalystFinding & { run_id: string }> = baselineRunId\n    ? (opts.findingsStore?.loadRun(baselineRunId) ?? [])\n    : []\n  log('baseline resolved', { baselineRunId, prior_findings: priorAll.length })\n  await emit({\n    type: 'baseline-resolved',\n    runId: opts.runId,\n    baselineRunId,\n    priorFindingCount: priorAll.length,\n  })\n\n  // 2. Run the registry. Strategy controls how analysts see priors.\n  //    When the registry exposes runStream, forward each event verbatim\n  //    so subscribers see per-analyst progress in real time.\n  const priorFindings = buildPriorFindingsInput(priorAll, strategy, opts.registry.list())\n  const analystResult = await runRegistry(opts, priorFindings, emit)\n  log('analyst run complete', {\n    findings: analystResult.findings.length,\n    cost_usd: analystResult.total_cost_usd,\n    per_analyst: analystResult.per_analyst.map((s) => ({\n      id: s.analyst_id,\n      status: s.status,\n      n: s.findings_count,\n    })),\n  })\n\n  // 3. Persist the new run before any side-effecting adapter runs so\n  //    the ledger is the source of truth even if an adapter throws.\n  if (opts.findingsStore && analystResult.findings.length > 0) {\n    await opts.findingsStore.append(opts.runId, analystResult.findings)\n    await emit({\n      type: 'findings-persisted',\n      runId: opts.runId,\n      count: analystResult.findings.length,\n    })\n  }\n\n  // 4. Diff vs baseline.\n  let diff: FindingsDiff | null = null\n  if (baselineRunId && analystResult.findings.length > 0) {\n    diff = diffFindings(\n      priorAll.map((f) => ({ ...f })),\n      analystResult.findings.map((f) => ({ ...f, run_id: opts.runId })),\n    )\n    log('diff vs baseline', {\n      appeared: diff.appeared.length,\n      disappeared: diff.disappeared.length,\n      persisted: diff.persisted.length,\n      changed: diff.changed.length,\n    })\n    await emit({\n      type: 'diff-computed',\n      runId: opts.runId,\n      baselineRunId,\n      appeared: diff.appeared.length,\n      disappeared: diff.disappeared.length,\n      persisted: diff.persisted.length,\n      changed: diff.changed.length,\n    })\n  }\n\n  // 5. Knowledge adapter — proposals + optional auto-apply.\n  let knowledge: KnowledgeReport<TProposal> | null = null\n  if (opts.knowledgeAdapter) {\n    knowledge = await runKnowledgeAdapter(opts, analystResult.findings, log, emit)\n  }\n\n  // 6. Improvement adapter — prompt / tool / scaffolding edits.\n  let improvement: ImprovementReport<TEdit> | null = null\n  if (opts.improvementAdapter) {\n    improvement = await runImprovementAdapter(opts, analystResult.findings, log, emit)\n  }\n\n  await emit({\n    type: 'loop-completed',\n    runId: opts.runId,\n    durationMs: Date.now() - startedAt,\n  })\n\n  return {\n    runId: opts.runId,\n    baselineRunId,\n    analystResult,\n    diff,\n    knowledge,\n    improvement,\n  }\n}\n\ntype Emitter = (event: AnalystLoopEvent) => Promise<void>\n\nfunction makeEmitter(onEvent: RunAnalystLoopOpts['onEvent']): Emitter {\n  if (!onEvent) return async () => {}\n  return async (event) => {\n    await onEvent(event)\n  }\n}\n\nasync function runRegistry(\n  opts: RunAnalystLoopOpts,\n  priorFindings: ReturnType<typeof buildPriorFindingsInput>,\n  emit: Emitter,\n): Promise<AnalystRunResult> {\n  const reg = opts.registry as AnalystRegistryStreamingLike\n  if (typeof reg.runStream === 'function' && opts.onEvent) {\n    let final: AnalystRunResult | null = null\n    for await (const ev of reg.runStream(opts.runId, opts.inputs, { priorFindings })) {\n      await emit({ type: 'analyst', runId: opts.runId, event: ev })\n      if (ev.type === 'run-completed') final = ev.result\n    }\n    if (!final) {\n      throw new Error('runAnalystLoop: registry.runStream ended without run-completed event')\n    }\n    return final\n  }\n  return opts.registry.run(opts.runId, opts.inputs, { priorFindings })\n}\n\nfunction resolveBaselineRunId(opts: RunAnalystLoopOpts): string | null {\n  if (opts.baselineRunId === null) return null\n  if (typeof opts.baselineRunId === 'string') return opts.baselineRunId\n  if (!opts.findingsStore) return null\n  const all = opts.findingsStore.loadAll()\n  let last: string | null = null\n  for (const row of all) {\n    if (row.run_id === opts.runId) continue\n    last = row.run_id\n  }\n  return last\n}\n\nfunction buildPriorFindingsInput(\n  prior: ReadonlyArray<AnalystFinding & { run_id: string }>,\n  strategy: 'per-kind' | 'wildcard' | 'none',\n  registry: ReadonlyArray<{ id: string }>,\n): ReadonlyArray<AnalystFinding> | Record<string, ReadonlyArray<AnalystFinding>> | undefined {\n  if (strategy === 'none' || prior.length === 0) return undefined\n  const stripped = prior.map(({ run_id: _run_id, ...rest }) => rest as AnalystFinding)\n  if (strategy === 'wildcard') {\n    return { '*': stripped }\n  }\n  void registry\n  return stripped\n}\n\nasync function runKnowledgeAdapter<TProposal>(\n  opts: RunAnalystLoopOpts,\n  findings: ReadonlyArray<AnalystFinding>,\n  log: NonNullable<RunAnalystLoopOpts['log']>,\n  emit: Emitter,\n): Promise<KnowledgeReport<TProposal>> {\n  const adapter = opts.knowledgeAdapter!\n  const batch = await adapter.proposeFromFindings(findings)\n  log('knowledge.proposeFromFindings', {\n    proposals: batch.proposals.length,\n    skipped: batch.skipped,\n    errors: batch.errors.length,\n  })\n  await emit({\n    type: 'knowledge-proposed',\n    runId: opts.runId,\n    proposalCount: batch.proposals.length,\n    skipped: batch.skipped,\n    errors: batch.errors.length,\n  })\n\n  const auto = opts.autoApply?.knowledge ?? false\n  const threshold = opts.autoApply?.knowledgeConfidenceThreshold ?? 0.85\n\n  if (!auto || !adapter.apply) {\n    await emit({\n      type: 'knowledge-applied',\n      runId: opts.runId,\n      writtenCount: 0,\n      withheldForReview: batch.proposals.length,\n    })\n    return {\n      proposals: batch.proposals as TProposal[],\n      applied: [],\n      skipped: batch.skipped,\n      errors: batch.errors,\n      withheld_for_review: batch.proposals.length,\n    }\n  }\n\n  const findingsById = new Map(findings.map((f) => [f.finding_id, f]))\n  const safe: TProposal[] = []\n  let withheld = 0\n  for (const p of batch.proposals as Array<TProposal & { sourceFindingId?: string }>) {\n    const src = p.sourceFindingId ? findingsById.get(p.sourceFindingId) : undefined\n    if (!src) {\n      withheld += 1\n      continue\n    }\n    if (src.confidence < threshold) {\n      withheld += 1\n      continue\n    }\n    safe.push(p)\n  }\n  const result = await adapter.apply(safe)\n  log('knowledge.apply', {\n    applied: result.written.length,\n    withheld_for_review: withheld,\n    warnings: result.warnings.length,\n  })\n  await emit({\n    type: 'knowledge-applied',\n    runId: opts.runId,\n    writtenCount: result.written.length,\n    withheldForReview: withheld,\n  })\n  return {\n    proposals: batch.proposals as TProposal[],\n    applied: result.written,\n    skipped: batch.skipped,\n    errors: batch.errors,\n    withheld_for_review: withheld,\n  }\n}\n\nasync function runImprovementAdapter<TEdit>(\n  opts: RunAnalystLoopOpts,\n  findings: ReadonlyArray<AnalystFinding>,\n  log: NonNullable<RunAnalystLoopOpts['log']>,\n  emit: Emitter,\n): Promise<ImprovementReport<TEdit>> {\n  const adapter = opts.improvementAdapter!\n  const batch = await adapter.proposeFromFindings(findings)\n  log('improvement.proposeFromFindings', {\n    edits: batch.edits.length,\n    skipped: batch.skipped,\n    errors: batch.errors.length,\n  })\n  await emit({\n    type: 'improvement-proposed',\n    runId: opts.runId,\n    editCount: batch.edits.length,\n    skipped: batch.skipped,\n    errors: batch.errors.length,\n  })\n\n  const auto = opts.autoApply?.improvement ?? false\n  const threshold = opts.autoApply?.improvementConfidenceThreshold ?? 0.9\n\n  if (!auto || !adapter.apply) {\n    await emit({\n      type: 'improvement-applied',\n      runId: opts.runId,\n      appliedCount: 0,\n      withheldForReview: batch.edits.length,\n    })\n    return {\n      edits: batch.edits as TEdit[],\n      applied: [],\n      skipped: batch.skipped,\n      errors: batch.errors,\n      withheld_for_review: batch.edits.length,\n    }\n  }\n\n  const findingsById = new Map(findings.map((f) => [f.finding_id, f]))\n  const safe: TEdit[] = []\n  let withheld = 0\n  for (const e of batch.edits as Array<TEdit & { sourceFindingId?: string }>) {\n    const src = e.sourceFindingId ? findingsById.get(e.sourceFindingId) : undefined\n    if (!src || src.confidence < threshold) {\n      withheld += 1\n      continue\n    }\n    safe.push(e)\n  }\n  const result = await adapter.apply(safe)\n  log('improvement.apply', {\n    applied: result.applied.length,\n    withheld_for_review: withheld,\n    warnings: result.warnings.length,\n  })\n  await emit({\n    type: 'improvement-applied',\n    runId: opts.runId,\n    appliedCount: result.applied.length,\n    withheldForReview: withheld,\n  })\n  return {\n    edits: batch.edits as TEdit[],\n    applied: result.applied,\n    skipped: batch.skipped,\n    errors: batch.errors,\n    withheld_for_review: withheld,\n  }\n}\n\nfunction defaultLog(msg: string, fields?: Record<string, unknown>): void {\n  if (fields) console.log(`[analyst-loop] ${msg}`, fields)\n  else console.log(`[analyst-loop] ${msg}`)\n}\n","#!/usr/bin/env node\n/**\n * @experimental\n *\n * `agent-runtime-loop` — the schedulable entrypoint for the configured\n * delegated loop-runner. A cron job / routine / Makefile target invokes:\n *\n *   agent-runtime-loop --mode research --config ./loops.config.js\n *\n * The config module wires the registry (with full access to env / creds —\n * which is why the deps live there, not in this generic bin). It must default-\n * export a `DelegatedLoopRegistry`, or a `() => DelegatedLoopRegistry | Promise<…>`.\n * The bin runs the selected mode, prints the `DelegatedLoopResult` as JSON, and\n * exits 0 on `ok`, 1 on a recorded failure, 2 on a usage/config error.\n */\n\nimport {\n  DELEGATED_LOOP_MODES,\n  type DelegatedLoopMode,\n  type DelegatedLoopRegistry,\n  type DelegatedLoopResult,\n  isDelegatedLoopMode,\n  runDelegatedLoop,\n} from './loop-runner'\n\n/** @experimental Parsed CLI invocation. */\nexport interface LoopRunnerCliArgs {\n  mode: string\n  /** Loads the registry — the bin wires this from `--config`; tests inject a stub. */\n  loadRegistry: () => Promise<DelegatedLoopRegistry> | DelegatedLoopRegistry\n  now?: () => number\n}\n\n/** @experimental */\nexport interface LoopRunnerCliResult {\n  exitCode: number\n  result?: DelegatedLoopResult\n  error?: string\n}\n\n/**\n * @experimental\n *\n * Pure CLI core (no process / argv / IO) so it's unit-testable: validate the\n * mode, load the registry, dispatch, map to an exit code (0 ok / 1 failed /\n * 2 usage). Exported for embedding in custom runners + tests.\n */\nexport async function runLoopRunnerCli(args: LoopRunnerCliArgs): Promise<LoopRunnerCliResult> {\n  if (!isDelegatedLoopMode(args.mode)) {\n    return {\n      exitCode: 2,\n      error: `unknown mode '${args.mode}' (expected one of: ${DELEGATED_LOOP_MODES.join(', ')})`,\n    }\n  }\n  let registry: DelegatedLoopRegistry\n  try {\n    registry = await args.loadRegistry()\n  } catch (err) {\n    return { exitCode: 2, error: `failed to load registry: ${errMsg(err)}` }\n  }\n  if (!registry[args.mode]) {\n    return {\n      exitCode: 2,\n      error: `config registers no runner for mode '${args.mode}' (registered: ${\n        Object.keys(registry).join(', ') || 'none'\n      })`,\n    }\n  }\n  // runDelegatedLoop throws only on a missing runner (guarded above); a failing\n  // engine is captured as { ok: false } → exit 1, not a crash.\n  const result = await runDelegatedLoop(args.mode as DelegatedLoopMode, registry, {\n    ...(args.now ? { now: args.now } : {}),\n  })\n  return { exitCode: result.ok ? 0 : 1, result }\n}\n\n/** Parse `--mode X --config Y` from an argv tail (`process.argv.slice(2)`). */\nexport function parseLoopRunnerArgv(argv: string[]): { mode?: string; config?: string } {\n  const out: { mode?: string; config?: string } = {}\n  for (let i = 0; i < argv.length; i += 1) {\n    const a = argv[i]\n    if (a === '--mode') out.mode = argv[++i]\n    else if (a === '--config') out.config = argv[++i]\n    else if (a?.startsWith('--mode=')) out.mode = a.slice('--mode='.length)\n    else if (a?.startsWith('--config=')) out.config = a.slice('--config='.length)\n  }\n  return out\n}\n\n/** Normalize a config module's default export → a registry. */\nfunction resolveRegistry(mod: unknown): DelegatedLoopRegistry {\n  const def = (mod as { default?: unknown })?.default ?? mod\n  const value = typeof def === 'function' ? (def as () => unknown)() : def\n  return value as DelegatedLoopRegistry\n}\n\nfunction errMsg(err: unknown): string {\n  return err instanceof Error ? err.message : String(err)\n}\n\n/** The argv → IO → exit shell. Kept thin; logic lives in `runLoopRunnerCli`. */\nasync function main(): Promise<void> {\n  const { mode, config } = parseLoopRunnerArgv(process.argv.slice(2))\n  if (!mode || !config) {\n    process.stderr.write(\n      'usage: agent-runtime-loop --mode <mode> --config <module>\\n' +\n        `  modes: ${DELEGATED_LOOP_MODES.join(' | ')}\\n` +\n        '  config: a JS/TS module default-exporting a DelegatedLoopRegistry (or a factory)\\n',\n    )\n    process.exit(2)\n  }\n  const { pathToFileURL } = await import('node:url')\n  const { resolve } = await import('node:path')\n  const cli = await runLoopRunnerCli({\n    mode,\n    loadRegistry: async () => resolveRegistry(await import(pathToFileURL(resolve(config)).href)),\n  })\n  process.stdout.write(`${JSON.stringify(cli.result ?? { error: cli.error }, null, 2)}\\n`)\n  if (cli.error) process.stderr.write(`${cli.error}\\n`)\n  process.exit(cli.exitCode)\n}\n\n// Run only when executed as the bin — never when imported for the testable\n// core, and never when bundled into a runtime that has no `process.argv`\n// (e.g. Cloudflare Workers, where `process` is a shim without `argv`). Reading\n// `process.argv[1]` directly would throw at module load there; `process.argv?.`\n// keeps the guard a no-op instead of crashing the Worker on startup.\nconst invokedScript = typeof process !== 'undefined' ? process.argv?.[1] : undefined\nif (invokedScript && /loop-runner-bin\\.(js|ts|mjs)$/.test(invokedScript)) {\n  void main()\n}\n"],"mappings":";;;;;;;;;;;;;;;AAuBA;AAAA,EAGE;AAAA,OACK;;;ACXP;AAAA,EAEE;AAAA,EAKA;AAAA,OASK;;;ACZP,SAAS,oBAAoB;AAW7B,eAAsB,eACpB,MACiD;AACjD,QAAM,MAAM,KAAK,OAAO;AACxB,QAAM,WAAW,KAAK,yBAAyB;AAC/C,QAAM,OAAO,YAAY,KAAK,OAAO;AACrC,QAAM,YAAY,KAAK,IAAI;AAG3B,QAAM,gBAAgB,qBAAqB,IAAI;AAC/C,QAAM,WAA+D,gBAChE,KAAK,eAAe,QAAQ,aAAa,KAAK,CAAC,IAChD,CAAC;AACL,MAAI,qBAAqB,EAAE,eAAe,gBAAgB,SAAS,OAAO,CAAC;AAC3E,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ;AAAA,IACA,mBAAmB,SAAS;AAAA,EAC9B,CAAC;AAKD,QAAM,gBAAgB,wBAAwB,UAAU,UAAU,KAAK,SAAS,KAAK,CAAC;AACtF,QAAM,gBAAgB,MAAM,YAAY,MAAM,eAAe,IAAI;AACjE,MAAI,wBAAwB;AAAA,IAC1B,UAAU,cAAc,SAAS;AAAA,IACjC,UAAU,cAAc;AAAA,IACxB,aAAa,cAAc,YAAY,IAAI,CAAC,OAAO;AAAA,MACjD,IAAI,EAAE;AAAA,MACN,QAAQ,EAAE;AAAA,MACV,GAAG,EAAE;AAAA,IACP,EAAE;AAAA,EACJ,CAAC;AAID,MAAI,KAAK,iBAAiB,cAAc,SAAS,SAAS,GAAG;AAC3D,UAAM,KAAK,cAAc,OAAO,KAAK,OAAO,cAAc,QAAQ;AAClE,UAAM,KAAK;AAAA,MACT,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,OAAO,cAAc,SAAS;AAAA,IAChC,CAAC;AAAA,EACH;AAGA,MAAI,OAA4B;AAChC,MAAI,iBAAiB,cAAc,SAAS,SAAS,GAAG;AACtD,WAAO;AAAA,MACL,SAAS,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,EAAE;AAAA,MAC9B,cAAc,SAAS,IAAI,CAAC,OAAO,EAAE,GAAG,GAAG,QAAQ,KAAK,MAAM,EAAE;AAAA,IAClE;AACA,QAAI,oBAAoB;AAAA,MACtB,UAAU,KAAK,SAAS;AAAA,MACxB,aAAa,KAAK,YAAY;AAAA,MAC9B,WAAW,KAAK,UAAU;AAAA,MAC1B,SAAS,KAAK,QAAQ;AAAA,IACxB,CAAC;AACD,UAAM,KAAK;AAAA,MACT,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ;AAAA,MACA,UAAU,KAAK,SAAS;AAAA,MACxB,aAAa,KAAK,YAAY;AAAA,MAC9B,WAAW,KAAK,UAAU;AAAA,MAC1B,SAAS,KAAK,QAAQ;AAAA,IACxB,CAAC;AAAA,EACH;AAGA,MAAI,YAA+C;AACnD,MAAI,KAAK,kBAAkB;AACzB,gBAAY,MAAM,oBAAoB,MAAM,cAAc,UAAU,KAAK,IAAI;AAAA,EAC/E;AAGA,MAAI,cAA+C;AACnD,MAAI,KAAK,oBAAoB;AAC3B,kBAAc,MAAM,sBAAsB,MAAM,cAAc,UAAU,KAAK,IAAI;AAAA,EACnF;AAEA,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,YAAY,KAAK,IAAI,IAAI;AAAA,EAC3B,CAAC;AAED,SAAO;AAAA,IACL,OAAO,KAAK;AAAA,IACZ;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AACF;AAIA,SAAS,YAAY,SAAiD;AACpE,MAAI,CAAC,QAAS,QAAO,YAAY;AAAA,EAAC;AAClC,SAAO,OAAO,UAAU;AACtB,UAAM,QAAQ,KAAK;AAAA,EACrB;AACF;AAEA,eAAe,YACb,MACA,eACA,MAC2B;AAC3B,QAAM,MAAM,KAAK;AACjB,MAAI,OAAO,IAAI,cAAc,cAAc,KAAK,SAAS;AACvD,QAAI,QAAiC;AACrC,qBAAiB,MAAM,IAAI,UAAU,KAAK,OAAO,KAAK,QAAQ,EAAE,cAAc,CAAC,GAAG;AAChF,YAAM,KAAK,EAAE,MAAM,WAAW,OAAO,KAAK,OAAO,OAAO,GAAG,CAAC;AAC5D,UAAI,GAAG,SAAS,gBAAiB,SAAQ,GAAG;AAAA,IAC9C;AACA,QAAI,CAAC,OAAO;AACV,YAAM,IAAI,MAAM,sEAAsE;AAAA,IACxF;AACA,WAAO;AAAA,EACT;AACA,SAAO,KAAK,SAAS,IAAI,KAAK,OAAO,KAAK,QAAQ,EAAE,cAAc,CAAC;AACrE;AAEA,SAAS,qBAAqB,MAAyC;AACrE,MAAI,KAAK,kBAAkB,KAAM,QAAO;AACxC,MAAI,OAAO,KAAK,kBAAkB,SAAU,QAAO,KAAK;AACxD,MAAI,CAAC,KAAK,cAAe,QAAO;AAChC,QAAM,MAAM,KAAK,cAAc,QAAQ;AACvC,MAAI,OAAsB;AAC1B,aAAW,OAAO,KAAK;AACrB,QAAI,IAAI,WAAW,KAAK,MAAO;AAC/B,WAAO,IAAI;AAAA,EACb;AACA,SAAO;AACT;AAEA,SAAS,wBACP,OACA,UACA,UAC2F;AAC3F,MAAI,aAAa,UAAU,MAAM,WAAW,EAAG,QAAO;AACtD,QAAM,WAAW,MAAM,IAAI,CAAC,EAAE,QAAQ,SAAS,GAAG,KAAK,MAAM,IAAsB;AACnF,MAAI,aAAa,YAAY;AAC3B,WAAO,EAAE,KAAK,SAAS;AAAA,EACzB;AACA,OAAK;AACL,SAAO;AACT;AAEA,eAAe,oBACb,MACA,UACA,KACA,MACqC;AACrC,QAAM,UAAU,KAAK;AACrB,QAAM,QAAQ,MAAM,QAAQ,oBAAoB,QAAQ;AACxD,MAAI,iCAAiC;AAAA,IACnC,WAAW,MAAM,UAAU;AAAA,IAC3B,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM,OAAO;AAAA,EACvB,CAAC;AACD,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,eAAe,MAAM,UAAU;AAAA,IAC/B,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM,OAAO;AAAA,EACvB,CAAC;AAED,QAAM,OAAO,KAAK,WAAW,aAAa;AAC1C,QAAM,YAAY,KAAK,WAAW,gCAAgC;AAElE,MAAI,CAAC,QAAQ,CAAC,QAAQ,OAAO;AAC3B,UAAM,KAAK;AAAA,MACT,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,cAAc;AAAA,MACd,mBAAmB,MAAM,UAAU;AAAA,IACrC,CAAC;AACD,WAAO;AAAA,MACL,WAAW,MAAM;AAAA,MACjB,SAAS,CAAC;AAAA,MACV,SAAS,MAAM;AAAA,MACf,QAAQ,MAAM;AAAA,MACd,qBAAqB,MAAM,UAAU;AAAA,IACvC;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,YAAY,CAAC,CAAC,CAAC;AACnE,QAAM,OAAoB,CAAC;AAC3B,MAAI,WAAW;AACf,aAAW,KAAK,MAAM,WAA8D;AAClF,UAAM,MAAM,EAAE,kBAAkB,aAAa,IAAI,EAAE,eAAe,IAAI;AACtE,QAAI,CAAC,KAAK;AACR,kBAAY;AACZ;AAAA,IACF;AACA,QAAI,IAAI,aAAa,WAAW;AAC9B,kBAAY;AACZ;AAAA,IACF;AACA,SAAK,KAAK,CAAC;AAAA,EACb;AACA,QAAM,SAAS,MAAM,QAAQ,MAAM,IAAI;AACvC,MAAI,mBAAmB;AAAA,IACrB,SAAS,OAAO,QAAQ;AAAA,IACxB,qBAAqB;AAAA,IACrB,UAAU,OAAO,SAAS;AAAA,EAC5B,CAAC;AACD,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,cAAc,OAAO,QAAQ;AAAA,IAC7B,mBAAmB;AAAA,EACrB,CAAC;AACD,SAAO;AAAA,IACL,WAAW,MAAM;AAAA,IACjB,SAAS,OAAO;AAAA,IAChB,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM;AAAA,IACd,qBAAqB;AAAA,EACvB;AACF;AAEA,eAAe,sBACb,MACA,UACA,KACA,MACmC;AACnC,QAAM,UAAU,KAAK;AACrB,QAAM,QAAQ,MAAM,QAAQ,oBAAoB,QAAQ;AACxD,MAAI,mCAAmC;AAAA,IACrC,OAAO,MAAM,MAAM;AAAA,IACnB,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM,OAAO;AAAA,EACvB,CAAC;AACD,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,WAAW,MAAM,MAAM;AAAA,IACvB,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM,OAAO;AAAA,EACvB,CAAC;AAED,QAAM,OAAO,KAAK,WAAW,eAAe;AAC5C,QAAM,YAAY,KAAK,WAAW,kCAAkC;AAEpE,MAAI,CAAC,QAAQ,CAAC,QAAQ,OAAO;AAC3B,UAAM,KAAK;AAAA,MACT,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,cAAc;AAAA,MACd,mBAAmB,MAAM,MAAM;AAAA,IACjC,CAAC;AACD,WAAO;AAAA,MACL,OAAO,MAAM;AAAA,MACb,SAAS,CAAC;AAAA,MACV,SAAS,MAAM;AAAA,MACf,QAAQ,MAAM;AAAA,MACd,qBAAqB,MAAM,MAAM;AAAA,IACnC;AAAA,EACF;AAEA,QAAM,eAAe,IAAI,IAAI,SAAS,IAAI,CAAC,MAAM,CAAC,EAAE,YAAY,CAAC,CAAC,CAAC;AACnE,QAAM,OAAgB,CAAC;AACvB,MAAI,WAAW;AACf,aAAW,KAAK,MAAM,OAAsD;AAC1E,UAAM,MAAM,EAAE,kBAAkB,aAAa,IAAI,EAAE,eAAe,IAAI;AACtE,QAAI,CAAC,OAAO,IAAI,aAAa,WAAW;AACtC,kBAAY;AACZ;AAAA,IACF;AACA,SAAK,KAAK,CAAC;AAAA,EACb;AACA,QAAM,SAAS,MAAM,QAAQ,MAAM,IAAI;AACvC,MAAI,qBAAqB;AAAA,IACvB,SAAS,OAAO,QAAQ;AAAA,IACxB,qBAAqB;AAAA,IACrB,UAAU,OAAO,SAAS;AAAA,EAC5B,CAAC;AACD,QAAM,KAAK;AAAA,IACT,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,cAAc,OAAO,QAAQ;AAAA,IAC7B,mBAAmB;AAAA,EACrB,CAAC;AACD,SAAO;AAAA,IACL,OAAO,MAAM;AAAA,IACb,SAAS,OAAO;AAAA,IAChB,SAAS,MAAM;AAAA,IACf,QAAQ,MAAM;AAAA,IACd,qBAAqB;AAAA,EACvB;AACF;AAEA,SAAS,WAAW,KAAa,QAAwC;AACvE,MAAI,OAAQ,SAAQ,IAAI,kBAAkB,GAAG,IAAI,MAAM;AAAA,MAClD,SAAQ,IAAI,kBAAkB,GAAG,EAAE;AAC1C;;;AF3RO,IAAM,uBAAuB,CAAC,QAAQ,UAAU,YAAY,SAAS,cAAc;AAMnF,SAAS,oBAAoB,OAA4C;AAC9E,SAAO,OAAO,UAAU,YAAa,qBAA2C,SAAS,KAAK;AAChG;AAmCA,eAAsB,iBACpB,MACA,UACA,UAAmC,CAAC,GACH;AACjC,QAAM,SAAS,SAAS,IAAI;AAC5B,MAAI,CAAC,QAAQ;AACX,UAAM,IAAI;AAAA,MACR,oDAAoD,IAAI,kBACtD,OAAO,KAAK,QAAQ,EAAE,KAAK,IAAI,KAAK,MACtC;AAAA,IACF;AAAA,EACF;AACA,QAAM,MAAM,QAAQ,OAAO,KAAK;AAChC,QAAM,SAAS,QAAQ,UAAU,IAAI,gBAAgB,EAAE;AACvD,QAAM,QAAQ,IAAI;AAClB,MAAI;AACF,UAAM,SAAS,MAAM,OAAO,MAAM;AAClC,WAAO,EAAE,MAAM,IAAI,MAAM,QAAQ,YAAY,IAAI,IAAI,MAAM;AAAA,EAC7D,SAAS,KAAK;AACZ,WAAO;AAAA,MACL;AAAA,MACA,IAAI;AAAA,MACJ,OAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAAA,MACtD,YAAY,IAAI,IAAI;AAAA,IACtB;AAAA,EACF;AACF;AAmBO,SAAS,gBAAgB,SAAmE;AACjG,QAAM,WAAW,wBAAwB;AAAA,IACvC,eAAe,QAAQ;AAAA,IACvB,GAAI,QAAQ,WAAW,EAAE,UAAU,QAAQ,SAAS,IAAI,CAAC;AAAA,IACzD,GAAI,QAAQ,kBAAkB,EAAE,iBAAiB,QAAQ,gBAAgB,IAAI,CAAC;AAAA,IAC9E,GAAI,QAAQ,kBAAkB,EAAE,iBAAiB,QAAQ,gBAAgB,IAAI,CAAC;AAAA,EAChF,CAAC;AACD,SAAO,OAAO,WAAW;AACvB,UAAM,MAAsB,EAAE,QAAQ,QAAQ,MAAM;AAAA,IAAC,EAAE;AACvD,WAAO,SAAS,QAAQ,MAAM,GAAG;AAAA,EACnC;AACF;AA0CO,SAAS,mBACd,SAC4C;AAC5C,QAAM,QAAQ,eAAuB;AAAA,IACnC,UAAU,QAAQ;AAAA,IAClB,YAAY,QAAQ;AAAA,IACpB,WAAW,QAAQ;AAAA,IACnB,GAAI,QAAQ,YAAY,SAAY,EAAE,SAAS,QAAQ,QAAQ,IAAI,CAAC;AAAA,IACpE,GAAI,QAAQ,iBAAiB,SAAY,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;AAAA,IACnF,GAAI,QAAQ,YAAY,SAAY,EAAE,SAAS,QAAQ,QAAQ,IAAI,CAAC;AAAA,IACpE,GAAI,QAAQ,iBAAiB,SAAY,EAAE,cAAc,QAAQ,aAAa,IAAI,CAAC;AAAA,IACnF,GAAI,QAAQ,mBAAmB,SAAY,EAAE,gBAAgB,QAAQ,eAAe,IAAI,CAAC;AAAA,IACzF,GAAI,QAAQ,mBAAmB,SAAY,EAAE,gBAAgB,QAAQ,eAAe,IAAI,CAAC;AAAA,IACzF,GAAI,QAAQ,SAAS,EAAE,QAAQ,QAAQ,OAAO,IAAI,CAAC;AAAA,IACnD,GAAI,QAAQ,aAAa,EAAE,YAAY,QAAQ,WAAW,IAAI,CAAC;AAAA,IAC/D,GAAI,QAAQ,aAAa,EAAE,YAAY,QAAQ,WAAW,IAAI,CAAC;AAAA,EACjE,CAAC;AAGD,QAAM,UAAU,cAAqC;AAAA,IACnD,MAAM;AAAA,IACN,MAAM,EAAE,SAAS,EAAE,MAAM,iBAAiB,GAAG,SAAS,KAAK;AAAA,IAC3D,WAAW;AAAA,IACX,SAAS,EAAE,MAAM,QAAQ;AAAA,IACzB,WAAW,EAAE,UAAU,uBAAuB,EAAE;AAAA,EAClD,CAAC;AACD,SAAO,OAAO,WAAW;AACvB,UAAM,SAAS,MAAM,eAA8C;AAAA,MACjE;AAAA,MACA;AAAA,MACA,MAAM,QAAQ;AAAA,MACd,QAAQ,QAAQ;AAAA,MAChB;AAAA,IACF,CAAC;AACD,QAAI,OAAO,SAAS,YAAY,OAAO,IAAI,SAAS,QAAQ;AAC1D,YAAM,WACJ,OAAO,SAAS,YAAY,OAAO,IAAI,SAAS,YAC5C,OAAO,IAAI,SAAS,KAAK,IAAI,IAC7B,sBAAsB,OAAO,IAAI;AACvC,YAAM,IAAI,MAAM,2CAA2C,QAAQ,GAAG;AAAA,IACxE;AACA,WAAO,OAAO,IAAI;AAAA,EACpB;AACF;AA0CO,SAAS,mBACd,GACyC;AACzC,QAAM,OAAO,aAAa,EAAE,IAAI;AAChC,QAAM,YAAY,KAAK,IAAI,GAAG,KAAK,MAAM,EAAE,aAAa,CAAC,CAAC;AAC1D,SAAO,OAAO,WAAW;AACvB,UAAM,WAA4B,CAAC;AACnC,QAAI,SAAuB,CAAC;AAC5B,QAAI,SAAS;AACb,aAAS,QAAQ,GAAG,QAAQ,WAAW,SAAS,GAAG;AACjD,UAAI,OAAO,QAAS;AACpB,gBAAU;AACV,YAAM,aAAa,MAAM,EAAE,SAAS,OAAO,MAAM;AACjD,UAAI,WAAW,WAAW,EAAG;AAC7B,eAAS,CAAC;AACV,iBAAW,KAAK,YAAY;AAC1B,cAAM,IAAI,MAAM,KAAK,CAAC;AACtB,YAAI,EAAE,SAAU,UAAS,KAAK,CAAC;AAAA,YAC1B,QAAO,KAAK,EAAE,WAAW,GAAG,UAAU,EAAE,UAAU,QAAQ,EAAE,OAAO,CAAC;AAAA,MAC3E;AACA,UAAI,OAAO,WAAW,EAAG;AAAA,IAC3B;AACA,WAAO,EAAE,UAAU,QAAQ,OAAO;AAAA,EACpC;AACF;AAGO,SAAS,sBACd,SAC8D;AAC9D,SAAO,YAAY,YAAkC,OAAO;AAC9D;AAGO,SAAS,gBACd,SAC6D;AAC7D,SAAO,YAAY,eAAiC,OAAO;AAC7D;;;AGhRA,eAAsB,iBAAiB,MAAuD;AAC5F,MAAI,CAAC,oBAAoB,KAAK,IAAI,GAAG;AACnC,WAAO;AAAA,MACL,UAAU;AAAA,MACV,OAAO,iBAAiB,KAAK,IAAI,uBAAuB,qBAAqB,KAAK,IAAI,CAAC;AAAA,IACzF;AAAA,EACF;AACA,MAAI;AACJ,MAAI;AACF,eAAW,MAAM,KAAK,aAAa;AAAA,EACrC,SAAS,KAAK;AACZ,WAAO,EAAE,UAAU,GAAG,OAAO,4BAA4B,OAAO,GAAG,CAAC,GAAG;AAAA,EACzE;AACA,MAAI,CAAC,SAAS,KAAK,IAAI,GAAG;AACxB,WAAO;AAAA,MACL,UAAU;AAAA,MACV,OAAO,wCAAwC,KAAK,IAAI,kBACtD,OAAO,KAAK,QAAQ,EAAE,KAAK,IAAI,KAAK,MACtC;AAAA,IACF;AAAA,EACF;AAGA,QAAM,SAAS,MAAM,iBAAiB,KAAK,MAA2B,UAAU;AAAA,IAC9E,GAAI,KAAK,MAAM,EAAE,KAAK,KAAK,IAAI,IAAI,CAAC;AAAA,EACtC,CAAC;AACD,SAAO,EAAE,UAAU,OAAO,KAAK,IAAI,GAAG,OAAO;AAC/C;AAGO,SAAS,oBAAoB,MAAoD;AACtF,QAAM,MAA0C,CAAC;AACjD,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,GAAG;AACvC,UAAM,IAAI,KAAK,CAAC;AAChB,QAAI,MAAM,SAAU,KAAI,OAAO,KAAK,EAAE,CAAC;AAAA,aAC9B,MAAM,WAAY,KAAI,SAAS,KAAK,EAAE,CAAC;AAAA,aACvC,GAAG,WAAW,SAAS,EAAG,KAAI,OAAO,EAAE,MAAM,UAAU,MAAM;AAAA,aAC7D,GAAG,WAAW,WAAW,EAAG,KAAI,SAAS,EAAE,MAAM,YAAY,MAAM;AAAA,EAC9E;AACA,SAAO;AACT;AAGA,SAAS,gBAAgB,KAAqC;AAC5D,QAAM,MAAO,KAA+B,WAAW;AACvD,QAAM,QAAQ,OAAO,QAAQ,aAAc,IAAsB,IAAI;AACrE,SAAO;AACT;AAEA,SAAS,OAAO,KAAsB;AACpC,SAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AACxD;AAGA,eAAe,OAAsB;AACnC,QAAM,EAAE,MAAM,OAAO,IAAI,oBAAoB,QAAQ,KAAK,MAAM,CAAC,CAAC;AAClE,MAAI,CAAC,QAAQ,CAAC,QAAQ;AACpB,YAAQ,OAAO;AAAA,MACb;AAAA,WACc,qBAAqB,KAAK,KAAK,CAAC;AAAA;AAAA;AAAA,IAEhD;AACA,YAAQ,KAAK,CAAC;AAAA,EAChB;AACA,QAAM,EAAE,cAAc,IAAI,MAAM,OAAO,KAAU;AACjD,QAAM,EAAE,QAAQ,IAAI,MAAM,OAAO,MAAW;AAC5C,QAAM,MAAM,MAAM,iBAAiB;AAAA,IACjC;AAAA,IACA,cAAc,YAAY,gBAAgB,MAAM,OAAO,cAAc,QAAQ,MAAM,CAAC,EAAE,KAAK;AAAA,EAC7F,CAAC;AACD,UAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,IAAI,UAAU,EAAE,OAAO,IAAI,MAAM,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACvF,MAAI,IAAI,MAAO,SAAQ,OAAO,MAAM,GAAG,IAAI,KAAK;AAAA,CAAI;AACpD,UAAQ,KAAK,IAAI,QAAQ;AAC3B;AAOA,IAAM,gBAAgB,OAAO,YAAY,cAAc,QAAQ,OAAO,CAAC,IAAI;AAC3E,IAAI,iBAAiB,gCAAgC,KAAK,aAAa,GAAG;AACxE,OAAK,KAAK;AACZ;","names":[]}

package/dist/{chunk-OYU5LR4T.js → chunk-EDCVUZZC.js}

@@ -14,7 +14,7 @@ import {
   DELEGATION_STATUS_DESCRIPTION,
   DELEGATION_STATUS_INPUT_SCHEMA,
   DELEGATION_STATUS_TOOL_NAME
-} from "./chunk-VKUHUFX7.js";
+} from "./chunk-QXWGSDAQ.js";
 
 // src/mcp/openai-tools.ts
 function buildTool(name, description, parameters) {
@@ -61,4 +61,4 @@ export {
   mcpToolsForRuntimeMcp,
   mcpToolsForRuntimeMcpSubset
 };
-//# sourceMappingURL=chunk-OYU5LR4T.js.map
+//# sourceMappingURL=chunk-EDCVUZZC.js.map

package/dist/{chunk-QNUEFNU7.js → chunk-L5ZFBVT6.js}

@@ -1,7 +1,3 @@
-import {
-  coderProfile,
-  coderTaskToPrompt
-} from "./chunk-DVQGYDN5.js";
 import {
   composeLoopTraceEmitters,
   detachedTurnEvents,
@@ -11,7 +7,11 @@ import {
   runDetachedTurn,
   runLoop,
   selectValidWinner
-} from "./chunk-VKUHUFX7.js";
+} from "./chunk-QXWGSDAQ.js";
+import {
+  coderProfile,
+  coderTaskToPrompt
+} from "./chunk-DVQGYDN5.js";
 
 // src/mcp/executor.ts
 function createSiblingSandboxExecutor(options) {
@@ -472,4 +472,4 @@ export {
   coderTaskFromArgs,
   settleDetachedCoderTurn
 };
-//# sourceMappingURL=chunk-QNUEFNU7.js.map
+//# sourceMappingURL=chunk-L5ZFBVT6.js.map