@tangle-network/agent-bench 0.3.6 → 0.3.8

package/dist/chunk-5H5XV76F.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/benchmarks/humaneval.ts"],"sourcesContent":["/**\n * HumanEval adapter — the deployable-checker domain as a `BenchmarkAdapter`, so the\n * gate runner (`runGate`) can A/B the STEERING regime on it: a real rollout\n * through the `Supervisor` that self-corrects across rounds, vs blind\n * random@k. This is the experiment `humaneval-gate.mts` names as \"the next one\" —\n * the gate measures SELECTION over stateless single completions; this measures\n * whether observe→steer (self-correction) beats blind compute at equal k.\n *\n * Worker artifact = the model's reply (a Python function, fenced or raw). The\n * DETERMINISTIC judge runs the candidate against the task's own `test` in an\n * isolated `--network=none` python container — exit 0 = pass. No gold\n * `canonical_solution` is ever shown to the model; `goldArtifact` returns it only\n * to self-verify the judge before spending tokens.\n *\n * The primitives (loader / extractor / Docker checker) live here and are reused by\n * `humaneval-gate.mts` — one home, no duplication.\n */\n\nimport { execFile } from 'node:child_process'\nimport { chmodSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs'\nimport { tmpdir } from 'node:os'\nimport { join } from 'node:path'\nimport { gunzipSync } from 'node:zlib'\nimport type { BenchmarkAdapter, BenchScore, BenchTask, LoadOptions } from './types'\n\nconst humanevalUrl = 'https://github.com/openai/human-eval/raw/master/data/HumanEval.jsonl.gz'\nconst dockerImage = 'python:3.12-slim'\nconst dockerTimeoutMs = Number(process.env.DOCKER_TIMEOUT_MS ?? 20000)\n\nexport interface HumanEvalTask {\n  taskId: string\n  prompt: string\n  test: string\n  entryPoint: string\n  /** Reference solution body — used ONLY to self-verify the judge, never shown to the worker. */\n  canonicalSolution?: string\n}\n\n/** Pull the 164-task HumanEval JSONL.gz and parse it. Fail loud on a non-OK fetch\n *  or a malformed line — a silently-short task set would poison the gate. `offset`\n *  selects a deeper slice (the later tasks are harder) so the worker has a\n *  correctable middle band rather than a saturated easy prefix. */\nexport async function loadHumanEval(limit: number, offset = 0): Promise<HumanEvalTask[]> {\n  // Prefer a locally-cached .jsonl.gz (HUMANEVAL_GZ) — the GitHub raw URL rate-limits\n  // (429) under repeated runs. Fall back to the network fetch when unset.\n  const localGz = process.env.HUMANEVAL_GZ\n  let gz: Buffer\n  if (localGz) {\n    gz = readFileSync(localGz)\n  } else {\n    const res = await fetch(humanevalUrl)\n    if (!res.ok) throw new Error(`HumanEval fetch HTTP ${res.status}: ${humanevalUrl}`)\n    gz = Buffer.from(await res.arrayBuffer())\n  }\n  const text = gunzipSync(gz).toString('utf8')\n  const tasks: HumanEvalTask[] = []\n  for (const line of text.split('\\n')) {\n    if (line.trim() === '') continue\n    const d = JSON.parse(line) as {\n      task_id?: string\n      prompt?: string\n      test?: string\n      entry_point?: string\n      canonical_solution?: string\n    }\n    if (!d.task_id || !d.prompt || !d.test || !d.entry_point) {\n      throw new Error(`malformed HumanEval record: ${line.slice(0, 120)}`)\n    }\n    tasks.push({\n      taskId: d.task_id,\n      prompt: d.prompt,\n      test: d.test,\n      entryPoint: d.entry_point,\n      ...(d.canonical_solution ? { canonicalSolution: d.canonical_solution } : {}),\n    })\n  }\n  if (tasks.length === 0) throw new Error('HumanEval parsed to 0 tasks')\n  if (offset >= tasks.length) throw new Error(`OFFSET ${offset} >= dataset size ${tasks.length}`)\n  return tasks.slice(offset, offset + limit)\n}\n\nconst solveInstruction =\n  'Complete the following Python function. Output the COMPLETE function definition (signature, docstring optional, body) inside a single ```python code block. Include any imports the function needs. Do not write tests or example calls.'\n\nexport function basePrompt(task: HumanEvalTask): string {\n  return `${solveInstruction}\\n\\n\\`\\`\\`python\\n${task.prompt}\\`\\`\\``\n}\n\n/** Extract the function source from a model reply: prefer a fenced ```python (or\n *  bare ```) block, else fall back to the raw text. The deployable program adds the\n *  prompt header (imports + signature context), so a candidate that returns only a\n *  body still runs; a candidate that re-defines the function shadows the header. */\nexport function extractCode(reply: string): string {\n  const fenced = reply.match(/```(?:python|py)?\\s*\\n([\\s\\S]*?)```/i)\n  if (fenced && typeof fenced[1] === 'string') return fenced[1].trim()\n  return reply.trim()\n}\n\n/** The deployable test program: the prompt header (imports + signature/docstring the\n *  model was given), then the candidate (its def shadows the header's stub), then the\n *  task's own check() suite and the call. No gold solution anywhere. */\nfunction buildProgram(task: HumanEvalTask, candidate: string): string {\n  return `${task.prompt}\\n${candidate}\\n\\n${task.test}\\n\\ncheck(${task.entryPoint})\\n`\n}\n\nexport interface CheckResult {\n  /** {0,1} pass-count for this candidate (1 = the check() suite passed). */\n  pass: number\n  /** On failure: the interpreter stderr tail (traceback / failing assertion). The\n   *  execution-grounded feedback a self-repair loop steers on; ignored by selection. */\n  detail?: string\n}\n\nexport interface PythonProgramResult {\n  exitCode: number\n  stdout: string\n  stderr: string\n}\n\n/** Run one candidate's deployable test program in an isolated container:\n *  read-only root, bounded writable `/tmp`, no network, one CPU, and 512 MiB.\n *  Exit 0 → pass. A docker invocation error (binary missing, daemon down, image\n *  unavailable) is NOT a test failure — it throws so the harness fails loud rather\n *  than scoring every candidate 0 from a broken checker. */\nlet dockerRunSeq = 0\n\n/** Run arbitrary Python in the same isolated container used by the HumanEval\n * checker. This is also the execution-tool primitive for experiments that let\n * a model test its own snippets; model-written code must never run on the host. */\nexport function runPythonProgram(\n  program: string,\n  timeoutMs = dockerTimeoutMs,\n): Promise<PythonProgramResult> {\n  const dir = mkdtempSync(join(tmpdir(), 'hev-'))\n  // Rootless Docker maps container root to a subordinate host uid that cannot\n  // traverse mkdtemp's default 0700 directory. The directory contains only the\n  // disposable candidate script and is mounted read-only into the container.\n  chmodSync(dir, 0o755)\n  writeFileSync(join(dir, 'p.py'), program, { mode: 0o644 })\n  // Unique container name so we can force-reap it regardless of the docker client's state.\n  const name = `hev-${process.pid}-${dockerRunSeq++}`\n  const startMarker = `__AGENT_RUNTIME_CANDIDATE_STARTED_${name}__`\n  return new Promise<PythonProgramResult>((resolvePromise, reject) => {\n    let settled = false\n    const cleanup = () => {\n      rmSync(dir, { recursive: true, force: true })\n      // `execFile`'s `timeout` kills the docker CLIENT, not the container — a hung\n      // `python` would otherwise pin a CPU forever. Force-reap by name (fire-and-forget;\n      // the name is unique, so no reuse race).\n      execFile('docker', ['rm', '-f', name], () => {})\n    }\n    const finish = (res: PythonProgramResult) => {\n      if (settled) return\n      settled = true\n      clearTimeout(backstop)\n      cleanup()\n      resolvePromise(res)\n    }\n    const fail = (e: Error) => {\n      if (settled) return\n      settled = true\n      clearTimeout(backstop)\n      cleanup()\n      reject(e)\n    }\n    // Candidate timeouts are enforced inside the container. If the outer Docker\n    // client or daemon misses this larger deadline, that is infrastructure failure,\n    // not a wrong answer.\n    const outerTimeoutMs = timeoutMs + 3_000\n    const backstop = setTimeout(\n      () => fail(new Error(`docker checker did not return within ${outerTimeoutMs + 3_000}ms`)),\n      outerTimeoutMs + 3_000,\n    )\n    const inContainerSeconds = Math.max(1, Math.ceil(timeoutMs / 1_000))\n    execFile(\n      'docker',\n      [\n        'run',\n        '--rm',\n        '--pull=never',\n        '--name',\n        name,\n        '--network=none',\n        '--cpus=1',\n        '--memory=512m',\n        '--pids-limit=64',\n        '--cap-drop=ALL',\n        '--security-opt=no-new-privileges',\n        '--read-only',\n        '--tmpfs',\n        '/tmp:rw,nosuid,nodev,noexec,size=64m,mode=1777',\n        '--user',\n        '65534:65534',\n        '-v',\n        `${dir}:/w:ro`,\n        '-w',\n        '/w',\n        dockerImage,\n        'sh',\n        '-c',\n        'command -v timeout >/dev/null 2>&1 && command -v python >/dev/null 2>&1 && [ -r /w/p.py ] || exit 126; printf \"%s\\\\n\" \"$1\"; exec timeout -s KILL \"$2\" python /w/p.py',\n        'agent-runtime-checker',\n        startMarker,\n        `${inContainerSeconds}s`,\n      ],\n      { timeout: outerTimeoutMs, killSignal: 'SIGKILL', maxBuffer: 4 * 1024 * 1024 },\n      (err, stdout, stderr) => {\n        const markerLine = `${startMarker}\\n`\n        const candidateStarted = stdout?.startsWith(markerLine) === true\n        const candidateStdout = candidateStarted ? stdout.slice(markerLine.length) : (stdout ?? '')\n        if (err) {\n          const e = err as Error & { killed?: boolean; code?: number | string }\n          if (e.code === 'ENOENT') {\n            fail(new Error('docker binary not found on PATH — cannot run the deployable checker'))\n            return\n          }\n          if (e.killed) {\n            fail(new Error(`docker checker exceeded its ${outerTimeoutMs}ms outer timeout`))\n            return\n          }\n          if (!candidateStarted) {\n            if (\n              /cannot connect to the docker daemon|is the docker daemon running|permission denied while trying to connect/i.test(\n                stderr,\n              )\n            ) {\n              fail(new Error(`docker daemon unreachable: ${stderr.slice(0, 200)}`))\n              return\n            }\n            if (/unable to find image|no such image|pull access denied|manifest unknown/i.test(stderr)) {\n              fail(new Error(`docker image ${dockerImage} unavailable: ${stderr.slice(0, 200)}`))\n              return\n            }\n            fail(\n              new Error(\n                `docker checker did not start the candidate: ${stderr.slice(0, 200) || e.message}`,\n              ),\n            )\n            return\n          }\n          const exitCode = typeof e.code === 'number' ? e.code : 1\n          finish({\n            exitCode,\n            stdout: candidateStdout,\n            stderr: stderr ?? '',\n          })\n          return\n        }\n        if (!candidateStarted) {\n          fail(new Error('docker checker exited without starting the candidate'))\n          return\n        }\n        finish({ exitCode: 0, stdout: candidateStdout, stderr: stderr ?? '' })\n      },\n    )\n  })\n}\n\nexport async function runChecker(task: HumanEvalTask, candidate: string): Promise<CheckResult> {\n  const result = await runPythonProgram(buildProgram(task, candidate))\n  return result.exitCode === 0\n    ? { pass: 1 }\n    : { pass: 0, detail: result.stderr.slice(-600) || 'timed out (no output)' }\n}\n\n/** A HumanEval task carries its checker inputs in metadata so the deterministic\n *  judge can rebuild the deployable program from a `BenchTask` alone. */\ninterface HumanEvalMeta extends Record<string, unknown> {\n  promptHeader: string\n  test: string\n  entryPoint: string\n  canonicalSolution?: string\n}\n\nfunction toBenchTask(t: HumanEvalTask): BenchTask {\n  const metadata: HumanEvalMeta = {\n    promptHeader: t.prompt,\n    test: t.test,\n    entryPoint: t.entryPoint,\n    ...(t.canonicalSolution ? { canonicalSolution: t.canonicalSolution } : {}),\n  }\n  return { id: t.taskId, prompt: basePrompt(t), metadata }\n}\n\nfunction taskFromMeta(task: BenchTask): HumanEvalTask {\n  const m = task.metadata as HumanEvalMeta | undefined\n  if (!m?.promptHeader || !m.test || !m.entryPoint) {\n    throw new Error(`HumanEval judge: task ${task.id} missing checker metadata`)\n  }\n  return { taskId: task.id, prompt: m.promptHeader, test: m.test, entryPoint: m.entryPoint }\n}\n\n/** The HumanEval `BenchmarkAdapter`. OFFSET (env) selects the correctable middle\n *  band; loadTasks honors `limit`/`ids`. The judge is the Docker deployable checker. */\nexport function createHumanEvalAdapter(): BenchmarkAdapter {\n  return {\n    name: 'humaneval',\n    async preflight() {\n      // Prove the daemon and exact local image before any model call. Scoring uses\n      // --pull=never so a registry or cold pull cannot become a candidate outcome.\n      await new Promise<void>((resolve, reject) => {\n        execFile('docker', ['version', '--format', '{{.Server.Version}}'], (err) => {\n          if (err) reject(new Error('HumanEval judge needs a running Docker daemon (python:3.12-slim, --network=none)'))\n          else resolve()\n        })\n      })\n      await new Promise<void>((resolve, reject) => {\n        execFile('docker', ['image', 'inspect', dockerImage], (err) => {\n          if (err) reject(new Error(`HumanEval judge needs the cached Docker image ${dockerImage}`))\n          else resolve()\n        })\n      })\n    },\n    async loadTasks(opts?: LoadOptions) {\n      const offset = Number(process.env.OFFSET ?? 0)\n      // Pull a generous window when filtering by id, else exactly `limit` from offset.\n      const all = await loadHumanEval(opts?.ids ? 164 : (opts?.limit ?? 8), offset)\n      const picked = opts?.ids ? all.filter((t) => opts.ids?.includes(t.taskId)) : all\n      return picked.map(toBenchTask)\n    },\n    async judge(task: BenchTask, artifact: string): Promise<BenchScore> {\n      const { pass } = await runChecker(taskFromMeta(task), extractCode(artifact))\n      return { resolved: pass === 1, score: pass, detail: pass === 1 ? 'tests passed' : 'tests failed' }\n    },\n    async goldArtifact(task: BenchTask) {\n      const m = task.metadata as HumanEvalMeta | undefined\n      const sol = m?.canonicalSolution\n      // Return the COMPLETE function (signature header + canonical body), i.e. what a\n      // real worker emits — NOT the body alone. The judge runs `extractCode`, whose\n      // unfenced fallback is `reply.trim()`; trimming a body-only string strips its\n      // leading indent and breaks it, so a body-only gold fails its own judge. A full\n      // def starts at column 0, trims safely, and self-verifies.\n      return sol ? `${m!.promptHeader}${sol}` : undefined\n    },\n  }\n}\n"],"mappings":";AAkBA,SAAS,gBAAgB;AACzB,SAAS,WAAW,aAAa,cAAc,QAAQ,qBAAqB;AAC5E,SAAS,cAAc;AACvB,SAAS,YAAY;AACrB,SAAS,kBAAkB;AAG3B,IAAM,eAAe;AACrB,IAAM,cAAc;AACpB,IAAM,kBAAkB,OAAO,QAAQ,IAAI,qBAAqB,GAAK;AAerE,eAAsB,cAAc,OAAe,SAAS,GAA6B;AAGvF,QAAM,UAAU,QAAQ,IAAI;AAC5B,MAAI;AACJ,MAAI,SAAS;AACX,SAAK,aAAa,OAAO;AAAA,EAC3B,OAAO;AACL,UAAM,MAAM,MAAM,MAAM,YAAY;AACpC,QAAI,CAAC,IAAI,GAAI,OAAM,IAAI,MAAM,wBAAwB,IAAI,MAAM,KAAK,YAAY,EAAE;AAClF,SAAK,OAAO,KAAK,MAAM,IAAI,YAAY,CAAC;AAAA,EAC1C;AACA,QAAM,OAAO,WAAW,EAAE,EAAE,SAAS,MAAM;AAC3C,QAAM,QAAyB,CAAC;AAChC,aAAW,QAAQ,KAAK,MAAM,IAAI,GAAG;AACnC,QAAI,KAAK,KAAK,MAAM,GAAI;AACxB,UAAM,IAAI,KAAK,MAAM,IAAI;AAOzB,QAAI,CAAC,EAAE,WAAW,CAAC,EAAE,UAAU,CAAC,EAAE,QAAQ,CAAC,EAAE,aAAa;AACxD,YAAM,IAAI,MAAM,+BAA+B,KAAK,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,IACrE;AACA,UAAM,KAAK;AAAA,MACT,QAAQ,EAAE;AAAA,MACV,QAAQ,EAAE;AAAA,MACV,MAAM,EAAE;AAAA,MACR,YAAY,EAAE;AAAA,MACd,GAAI,EAAE,qBAAqB,EAAE,mBAAmB,EAAE,mBAAmB,IAAI,CAAC;AAAA,IAC5E,CAAC;AAAA,EACH;AACA,MAAI,MAAM,WAAW,EAAG,OAAM,IAAI,MAAM,6BAA6B;AACrE,MAAI,UAAU,MAAM,OAAQ,OAAM,IAAI,MAAM,UAAU,MAAM,oBAAoB,MAAM,MAAM,EAAE;AAC9F,SAAO,MAAM,MAAM,QAAQ,SAAS,KAAK;AAC3C;AAEA,IAAM,mBACJ;AAEK,SAAS,WAAW,MAA6B;AACtD,SAAO,GAAG,gBAAgB;AAAA;AAAA;AAAA,EAAqB,KAAK,MAAM;AAC5D;AAMO,SAAS,YAAY,OAAuB;AACjD,QAAM,SAAS,MAAM,MAAM,sCAAsC;AACjE,MAAI,UAAU,OAAO,OAAO,CAAC,MAAM,SAAU,QAAO,OAAO,CAAC,EAAE,KAAK;AACnE,SAAO,MAAM,KAAK;AACpB;AAKA,SAAS,aAAa,MAAqB,WAA2B;AACpE,SAAO,GAAG,KAAK,MAAM;AAAA,EAAK,SAAS;AAAA;AAAA,EAAO,KAAK,IAAI;AAAA;AAAA,QAAa,KAAK,UAAU;AAAA;AACjF;AAqBA,IAAI,eAAe;AAKZ,SAAS,iBACd,SACA,YAAY,iBACkB;AAC9B,QAAM,MAAM,YAAY,KAAK,OAAO,GAAG,MAAM,CAAC;AAI9C,YAAU,KAAK,GAAK;AACpB,gBAAc,KAAK,KAAK,MAAM,GAAG,SAAS,EAAE,MAAM,IAAM,CAAC;AAEzD,QAAM,OAAO,OAAO,QAAQ,GAAG,IAAI,cAAc;AACjD,QAAM,cAAc,qCAAqC,IAAI;AAC7D,SAAO,IAAI,QAA6B,CAAC,gBAAgB,WAAW;AAClE,QAAI,UAAU;AACd,UAAM,UAAU,MAAM;AACpB,aAAO,KAAK,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC;AAI5C,eAAS,UAAU,CAAC,MAAM,MAAM,IAAI,GAAG,MAAM;AAAA,MAAC,CAAC;AAAA,IACjD;AACA,UAAM,SAAS,CAAC,QAA6B;AAC3C,UAAI,QAAS;AACb,gBAAU;AACV,mBAAa,QAAQ;AACrB,cAAQ;AACR,qBAAe,GAAG;AAAA,IACpB;AACA,UAAM,OAAO,CAAC,MAAa;AACzB,UAAI,QAAS;AACb,gBAAU;AACV,mBAAa,QAAQ;AACrB,cAAQ;AACR,aAAO,CAAC;AAAA,IACV;AAIA,UAAM,iBAAiB,YAAY;AACnC,UAAM,WAAW;AAAA,MACf,MAAM,KAAK,IAAI,MAAM,wCAAwC,iBAAiB,GAAK,IAAI,CAAC;AAAA,MACxF,iBAAiB;AAAA,IACnB;AACA,UAAM,qBAAqB,KAAK,IAAI,GAAG,KAAK,KAAK,YAAY,GAAK,CAAC;AACnE;AAAA,MACE;AAAA,MACA;AAAA,QACE;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,GAAG,GAAG;AAAA,QACN;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,GAAG,kBAAkB;AAAA,MACvB;AAAA,MACA,EAAE,SAAS,gBAAgB,YAAY,WAAW,WAAW,IAAI,OAAO,KAAK;AAAA,MAC7E,CAAC,KAAK,QAAQ,WAAW;AACvB,cAAM,aAAa,GAAG,WAAW;AAAA;AACjC,cAAM,mBAAmB,QAAQ,WAAW,UAAU,MAAM;AAC5D,cAAM,kBAAkB,mBAAmB,OAAO,MAAM,WAAW,MAAM,IAAK,UAAU;AACxF,YAAI,KAAK;AACP,gBAAM,IAAI;AACV,cAAI,EAAE,SAAS,UAAU;AACvB,iBAAK,IAAI,MAAM,0EAAqE,CAAC;AACrF;AAAA,UACF;AACA,cAAI,EAAE,QAAQ;AACZ,iBAAK,IAAI,MAAM,+BAA+B,cAAc,kBAAkB,CAAC;AAC/E;AAAA,UACF;AACA,cAAI,CAAC,kBAAkB;AACrB,gBACE,8GAA8G;AAAA,cAC5G;AAAA,YACF,GACA;AACA,mBAAK,IAAI,MAAM,8BAA8B,OAAO,MAAM,GAAG,GAAG,CAAC,EAAE,CAAC;AACpE;AAAA,YACF;AACA,gBAAI,0EAA0E,KAAK,MAAM,GAAG;AAC1F,mBAAK,IAAI,MAAM,gBAAgB,WAAW,iBAAiB,OAAO,MAAM,GAAG,GAAG,CAAC,EAAE,CAAC;AAClF;AAAA,YACF;AACA;AAAA,cACE,IAAI;AAAA,gBACF,+CAA+C,OAAO,MAAM,GAAG,GAAG,KAAK,EAAE,OAAO;AAAA,cAClF;AAAA,YACF;AACA;AAAA,UACF;AACA,gBAAM,WAAW,OAAO,EAAE,SAAS,WAAW,EAAE,OAAO;AACvD,iBAAO;AAAA,YACL;AAAA,YACA,QAAQ;AAAA,YACR,QAAQ,UAAU;AAAA,UACpB,CAAC;AACD;AAAA,QACF;AACA,YAAI,CAAC,kBAAkB;AACrB,eAAK,IAAI,MAAM,sDAAsD,CAAC;AACtE;AAAA,QACF;AACA,eAAO,EAAE,UAAU,GAAG,QAAQ,iBAAiB,QAAQ,UAAU,GAAG,CAAC;AAAA,MACvE;AAAA,IACF;AAAA,EACF,CAAC;AACH;AAEA,eAAsB,WAAW,MAAqB,WAAyC;AAC7F,QAAM,SAAS,MAAM,iBAAiB,aAAa,MAAM,SAAS,CAAC;AACnE,SAAO,OAAO,aAAa,IACvB,EAAE,MAAM,EAAE,IACV,EAAE,MAAM,GAAG,QAAQ,OAAO,OAAO,MAAM,IAAI,KAAK,wBAAwB;AAC9E;AAWA,SAAS,YAAY,GAA6B;AAChD,QAAM,WAA0B;AAAA,IAC9B,cAAc,EAAE;AAAA,IAChB,MAAM,EAAE;AAAA,IACR,YAAY,EAAE;AAAA,IACd,GAAI,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,kBAAkB,IAAI,CAAC;AAAA,EAC1E;AACA,SAAO,EAAE,IAAI,EAAE,QAAQ,QAAQ,WAAW,CAAC,GAAG,SAAS;AACzD;AAEA,SAAS,aAAa,MAAgC;AACpD,QAAM,IAAI,KAAK;AACf,MAAI,CAAC,GAAG,gBAAgB,CAAC,EAAE,QAAQ,CAAC,EAAE,YAAY;AAChD,UAAM,IAAI,MAAM,yBAAyB,KAAK,EAAE,2BAA2B;AAAA,EAC7E;AACA,SAAO,EAAE,QAAQ,KAAK,IAAI,QAAQ,EAAE,cAAc,MAAM,EAAE,MAAM,YAAY,EAAE,WAAW;AAC3F;AAIO,SAAS,yBAA2C;AACzD,SAAO;AAAA,IACL,MAAM;AAAA,IACN,MAAM,YAAY;AAGhB,YAAM,IAAI,QAAc,CAAC,SAAS,WAAW;AAC3C,iBAAS,UAAU,CAAC,WAAW,YAAY,qBAAqB,GAAG,CAAC,QAAQ;AAC1E,cAAI,IAAK,QAAO,IAAI,MAAM,kFAAkF,CAAC;AAAA,cACxG,SAAQ;AAAA,QACf,CAAC;AAAA,MACH,CAAC;AACD,YAAM,IAAI,QAAc,CAAC,SAAS,WAAW;AAC3C,iBAAS,UAAU,CAAC,SAAS,WAAW,WAAW,GAAG,CAAC,QAAQ;AAC7D,cAAI,IAAK,QAAO,IAAI,MAAM,iDAAiD,WAAW,EAAE,CAAC;AAAA,cACpF,SAAQ;AAAA,QACf,CAAC;AAAA,MACH,CAAC;AAAA,IACH;AAAA,IACA,MAAM,UAAU,MAAoB;AAClC,YAAM,SAAS,OAAO,QAAQ,IAAI,UAAU,CAAC;AAE7C,YAAM,MAAM,MAAM,cAAc,MAAM,MAAM,MAAO,MAAM,SAAS,GAAI,MAAM;AAC5E,YAAM,SAAS,MAAM,MAAM,IAAI,OAAO,CAAC,MAAM,KAAK,KAAK,SAAS,EAAE,MAAM,CAAC,IAAI;AAC7E,aAAO,OAAO,IAAI,WAAW;AAAA,IAC/B;AAAA,IACA,MAAM,MAAM,MAAiB,UAAuC;AAClE,YAAM,EAAE,KAAK,IAAI,MAAM,WAAW,aAAa,IAAI,GAAG,YAAY,QAAQ,CAAC;AAC3E,aAAO,EAAE,UAAU,SAAS,GAAG,OAAO,MAAM,QAAQ,SAAS,IAAI,iBAAiB,eAAe;AAAA,IACnG;AAAA,IACA,MAAM,aAAa,MAAiB;AAClC,YAAM,IAAI,KAAK;AACf,YAAM,MAAM,GAAG;AAMf,aAAO,MAAM,GAAG,EAAG,YAAY,GAAG,GAAG,KAAK;AAAA,IAC5C;AAAA,EACF;AACF;","names":[]}

package/dist/{chunk-YCGY7UIZ.js → chunk-7GRVHU22.js}

@@ -1,6 +1,6 @@
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/finresearchbench.ts
 import { readFile, stat } from "fs/promises";
@@ -205,4 +205,4 @@ function createFinResearchBenchAdapter() {
 export {
   createFinResearchBenchAdapter
 };
-//# sourceMappingURL=chunk-YCGY7UIZ.js.map
+//# sourceMappingURL=chunk-7GRVHU22.js.map

package/dist/{chunk-Z7ML6L77.js → chunk-HWST3SED.js}

@@ -1,7 +1,7 @@
 import {
   benchRoot,
   runVenvPython
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/webarena-verified.ts
 import { access, readFile, stat } from "fs/promises";
@@ -159,4 +159,4 @@ export {
   webarenaOutputDirOutput,
   createWebArenaVerifiedAdapter
 };
-//# sourceMappingURL=chunk-Z7ML6L77.js.map
+//# sourceMappingURL=chunk-HWST3SED.js.map

package/dist/{chunk-SYDW647C.js → chunk-IA2FBTWC.js}

@@ -3,7 +3,7 @@ import {
   preflightVenvImports,
   runVenvScriptStdin,
   venvPython
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/appworld.ts
 import { spawn } from "child_process";
@@ -315,4 +315,4 @@ export {
   appworldToolLoopClient,
   createAppWorldReactAdapter
 };
-//# sourceMappingURL=chunk-SYDW647C.js.map
+//# sourceMappingURL=chunk-IA2FBTWC.js.map

package/dist/{chunk-R67DFVLO.js → chunk-IFVINJ4B.js}

@@ -14,7 +14,7 @@ import {
 } from "./chunk-X3BTXCJ4.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/ragbench.ts
 import { readFile } from "fs/promises";
@@ -139,4 +139,4 @@ function createRagBenchAdapter() {
 export {
   createRagBenchAdapter
 };
-//# sourceMappingURL=chunk-R67DFVLO.js.map
+//# sourceMappingURL=chunk-IFVINJ4B.js.map

package/dist/{chunk-R36V2VP7.js → chunk-IZ5M6OAC.js}

@@ -1,6 +1,6 @@
 import {
   runVenvPython
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/tau-bench-shared.ts
 import { readFile, stat } from "fs/promises";
@@ -166,4 +166,4 @@ export {
   tauResultsOutput,
   createTauBenchAdapter
 };
-//# sourceMappingURL=chunk-R36V2VP7.js.map
+//# sourceMappingURL=chunk-IZ5M6OAC.js.map

package/dist/{chunk-ODT47UAY.js → chunk-K3BQGZCT.js}

@@ -5,7 +5,7 @@ import {
   runStagedJudge,
   stageFile,
   venvPython
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/aec-bench.ts
 import { readFile } from "fs/promises";
@@ -218,4 +218,4 @@ Fix: ensure network access to raw.githubusercontent.com, or set AEC_FIXTURES=1 t
 export {
   createAecBenchAdapter
 };
-//# sourceMappingURL=chunk-ODT47UAY.js.map
+//# sourceMappingURL=chunk-K3BQGZCT.js.map

package/dist/{chunk-IFAV6KEM.js → chunk-KP5KD6EN.js}

@@ -1,6 +1,6 @@
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/bfcl.ts
 import { readFile, stat } from "fs/promises";
@@ -273,4 +273,4 @@ export {
   bfclOutput,
   createBfclAdapter
 };
-//# sourceMappingURL=chunk-IFAV6KEM.js.map
+//# sourceMappingURL=chunk-KP5KD6EN.js.map

package/dist/{chunk-ZEWMTR5M.js → chunk-MQMRLGOG.js}

@@ -2,7 +2,7 @@ import {
   benchRoot,
   preflightVenvImports,
   runVenvScriptStdin
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/programbench.ts
 import { join } from "path";
@@ -133,4 +133,4 @@ export {
   programbenchSubmissionOutput,
   createProgrambenchAdapter
 };
-//# sourceMappingURL=chunk-ZEWMTR5M.js.map
+//# sourceMappingURL=chunk-MQMRLGOG.js.map

package/dist/{chunk-TSWPNOYM.js → chunk-NQG5XDSB.js}

@@ -10,7 +10,7 @@ import {
 } from "./chunk-X3BTXCJ4.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/nomiracl.ts
 import { readFile } from "fs/promises";
@@ -144,4 +144,4 @@ function createNoMiraclAdapter() {
 export {
   createNoMiraclAdapter
 };
-//# sourceMappingURL=chunk-TSWPNOYM.js.map
+//# sourceMappingURL=chunk-NQG5XDSB.js.map

package/dist/{chunk-7WSD27QQ.js → chunk-PB64GYIG.js}

@@ -1,6 +1,6 @@
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/agentbench.ts
 import { readFile } from "fs/promises";
@@ -115,4 +115,4 @@ export {
   agentbenchAnswerOutput,
   createAgentBenchAdapter
 };
-//# sourceMappingURL=chunk-7WSD27QQ.js.map
+//# sourceMappingURL=chunk-PB64GYIG.js.map

package/dist/{chunk-HBSWHQNJ.js → chunk-RCYQEFNX.js}

@@ -1,10 +1,10 @@
 import {
   createTauBenchAdapter,
   tauResultsOutput
-} from "./chunk-R36V2VP7.js";
+} from "./chunk-IZ5M6OAC.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/tau2-bench.ts
 import { join } from "path";
@@ -27,4 +27,4 @@ export {
   tau2ResultsOutput,
   createTau2BenchAdapter
 };
-//# sourceMappingURL=chunk-HBSWHQNJ.js.map
+//# sourceMappingURL=chunk-RCYQEFNX.js.map

package/dist/{chunk-J3KDJNX2.js → chunk-RH5F53JT.js}

@@ -2,7 +2,7 @@ import {
   benchRoot,
   runVenvPython,
   runVenvScriptStdin
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/dabstep.ts
 import { join } from "path";
@@ -179,4 +179,4 @@ export {
   dabstepAnswerOutput,
   createDabstepAdapter
 };
-//# sourceMappingURL=chunk-J3KDJNX2.js.map
+//# sourceMappingURL=chunk-RH5F53JT.js.map

package/dist/{chunk-UAIOHCUK.js → chunk-SFLA7OH3.js}

@@ -1,9 +1,9 @@
 import {
   createTauBenchAdapter
-} from "./chunk-R36V2VP7.js";
+} from "./chunk-IZ5M6OAC.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/tau3-banking.ts
 import { join } from "path";
@@ -24,4 +24,4 @@ function createTau3BankingAdapter() {
 export {
   createTau3BankingAdapter
 };
-//# sourceMappingURL=chunk-UAIOHCUK.js.map
+//# sourceMappingURL=chunk-SFLA7OH3.js.map

package/dist/{chunk-Y6O2OCUO.js → chunk-SHM6MRRF.js}

@@ -14,7 +14,7 @@ import {
 } from "./chunk-X3BTXCJ4.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/t2-ragbench.ts
 import { readFile } from "fs/promises";
@@ -127,4 +127,4 @@ function createT2RagBenchAdapter() {
 export {
   createT2RagBenchAdapter
 };
-//# sourceMappingURL=chunk-Y6O2OCUO.js.map
+//# sourceMappingURL=chunk-SHM6MRRF.js.map

package/dist/{chunk-KDIKRJGB.js → chunk-SHYIRB7I.js}

@@ -14,7 +14,7 @@ import {
 } from "./chunk-X3BTXCJ4.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/open-rag-bench.ts
 import { readFile } from "fs/promises";
@@ -117,4 +117,4 @@ function createOpenRagBenchAdapter() {
 export {
   createOpenRagBenchAdapter
 };
-//# sourceMappingURL=chunk-KDIKRJGB.js.map
+//# sourceMappingURL=chunk-SHYIRB7I.js.map

package/dist/{chunk-HHXFIHXC.js → chunk-SVR2LKYI.js}

@@ -12,7 +12,7 @@ import {
 } from "./chunk-X3BTXCJ4.js";
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/crag.ts
 import { readFile } from "fs/promises";
@@ -113,4 +113,4 @@ function createCragAdapter() {
 export {
   createCragAdapter
 };
-//# sourceMappingURL=chunk-HHXFIHXC.js.map
+//# sourceMappingURL=chunk-SVR2LKYI.js.map

package/dist/{chunk-5SBJCB6W.js → chunk-V7AEBY6U.js}

@@ -1,33 +1,33 @@
 import {
   createToolLlmAdapter
-} from "./chunk-2PVVP7GN.js";
+} from "./chunk-XKEFIFIC.js";
 import {
   createTrataHedgeAdapter
 } from "./chunk-VQRS7VUC.js";
 import {
   createWebArenaVerifiedAdapter
-} from "./chunk-Z7ML6L77.js";
+} from "./chunk-HWST3SED.js";
 import {
   createRagBenchAdapter
-} from "./chunk-R67DFVLO.js";
+} from "./chunk-IFVINJ4B.js";
 import {
   createSimpleQaAdapter
 } from "./chunk-SEVJPLZC.js";
 import {
   createSweBenchAdapter
-} from "./chunk-CKUVRZ2T.js";
+} from "./chunk-3U5TXJZS.js";
 import {
   createT2RagBenchAdapter
-} from "./chunk-Y6O2OCUO.js";
+} from "./chunk-SHM6MRRF.js";
 import {
   createTau2BenchAdapter
-} from "./chunk-HBSWHQNJ.js";
+} from "./chunk-RCYQEFNX.js";
 import {
   createTau3BankingAdapter
-} from "./chunk-UAIOHCUK.js";
+} from "./chunk-SFLA7OH3.js";
 import {
   createTerminalBenchAdapter
-} from "./chunk-JRWWGMK7.js";
+} from "./chunk-XYA4XSNU.js";
 import {
   createFramesAdapter
 } from "./chunk-INNOYXCP.js";
@@ -36,19 +36,19 @@ import {
 } from "./chunk-PA2ZKHJC.js";
 import {
   createHumanEvalAdapter
-} from "./chunk-PPYSEKFM.js";
+} from "./chunk-5H5XV76F.js";
 import {
   createMind2WebAdapter
 } from "./chunk-TBKU5XQI.js";
 import {
   createNoMiraclAdapter
-} from "./chunk-TSWPNOYM.js";
+} from "./chunk-NQG5XDSB.js";
 import {
   createOpenRagBenchAdapter
-} from "./chunk-KDIKRJGB.js";
+} from "./chunk-SHYIRB7I.js";
 import {
   createProgrambenchAdapter
-} from "./chunk-ZEWMTR5M.js";
+} from "./chunk-MQMRLGOG.js";
 import {
   createCadBenchAdapter
 } from "./chunk-PUIRNYI7.js";
@@ -57,35 +57,35 @@ import {
 } from "./chunk-C7T7WEK2.js";
 import {
   createCommit0Adapter
-} from "./chunk-2XU6OGEN.js";
+} from "./chunk-Z4TZ76N7.js";
 import {
   createCragAdapter
-} from "./chunk-HHXFIHXC.js";
+} from "./chunk-SVR2LKYI.js";
 import {
   createDabstepAdapter
-} from "./chunk-J3KDJNX2.js";
+} from "./chunk-RH5F53JT.js";
 import {
   createEnterpriseOpsGymAdapter
-} from "./chunk-X5YKXC6V.js";
+} from "./chunk-YSMEKBTD.js";
 import {
   createFinResearchBenchAdapter
-} from "./chunk-YCGY7UIZ.js";
+} from "./chunk-7GRVHU22.js";
 import {
   createFinsearchcompAdapter
 } from "./chunk-UPAMRDX4.js";
 import {
   createAecBenchAdapter
-} from "./chunk-ODT47UAY.js";
+} from "./chunk-K3BQGZCT.js";
 import {
   createAgentBenchAdapter
-} from "./chunk-7WSD27QQ.js";
+} from "./chunk-PB64GYIG.js";
 import {
   createAppWorldAdapter,
   createAppWorldReactAdapter
-} from "./chunk-SYDW647C.js";
+} from "./chunk-IA2FBTWC.js";
 import {
   createBfclAdapter
-} from "./chunk-IFAV6KEM.js";
+} from "./chunk-KP5KD6EN.js";
 import {
   createCadDesignAdapter
 } from "./chunk-53UPUNBZ.js";
@@ -141,4 +141,4 @@ export {
   ADAPTERS,
   resolveAdapter
 };
-//# sourceMappingURL=chunk-5SBJCB6W.js.map
+//# sourceMappingURL=chunk-V7AEBY6U.js.map

package/dist/{chunk-LRRD7NAG.js → chunk-WSKWVEQB.js}

@@ -1,6 +1,7 @@
 // src/benchmarks/_harness.ts
 import { execFile, spawn } from "child_process";
 import { createHash } from "crypto";
+import { existsSync, readFileSync } from "fs";
 import {
   cp,
   lstat,
@@ -18,7 +19,22 @@ import { basename, dirname, isAbsolute, join, relative, resolve, sep } from "pat
 import { fileURLToPath } from "url";
 import { promisify } from "util";
 var execFileAsync = promisify(execFile);
-var benchRoot = fileURLToPath(new URL("../..", import.meta.url));
+function resolveBenchRoot(moduleUrl) {
+  let current = dirname(fileURLToPath(moduleUrl));
+  while (true) {
+    const manifestPath = join(current, "package.json");
+    if (existsSync(manifestPath)) {
+      const manifest = JSON.parse(readFileSync(manifestPath, "utf8"));
+      if (manifest.name === "@tangle-network/agent-bench") return current;
+    }
+    const parent = dirname(current);
+    if (parent === current) {
+      throw new Error(`Unable to locate @tangle-network/agent-bench from ${moduleUrl}`);
+    }
+    current = parent;
+  }
+}
+var benchRoot = resolveBenchRoot(import.meta.url);
 function resolveBenchPython(env = process.env, root = benchRoot) {
   const configured = env.AGENT_BENCH_PYTHON;
   if (configured === void 0) return join(root, ".venv", "bin", "python");
@@ -298,4 +314,4 @@ export {
   stageFile,
   safeRunId
 };
-//# sourceMappingURL=chunk-LRRD7NAG.js.map
+//# sourceMappingURL=chunk-WSKWVEQB.js.map

package/dist/chunk-WSKWVEQB.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/benchmarks/_harness.ts"],"sourcesContent":["/**\n * Shared code-bench harness. The \"stage the artifact → run the benchmark's own\n * evaluator in an external process (mkdtemp / execFile / .venv python / Docker)\n * → read its JSON report → { resolved, score }\" spine that swe-bench,\n * terminal-bench, commit0, programbench and aec-bench all need. Factored out so\n * the Docker/venv/report-reading logic lives in ONE place instead of being\n * copy-pasted per adapter.\n *\n * It owns NO benchmark policy: each adapter passes the argv for its evaluator\n * and a `parseReport` that maps that evaluator's report JSON → a BenchScore. The\n * harness owns process spawning, temp-dir lifecycle, large-buffer/timeout config\n * and fail-loud diagnostics.\n *\n * Fail-loud: an absent .venv / harness / Docker daemon THROWS from `preflight`\n * (the adapter passes the import line + the exact fix). A staged run that exits\n * nonzero throws with the captured stderr — never a fabricated score.\n */\n\nimport { execFile, spawn } from 'node:child_process'\nimport { createHash } from 'node:crypto'\nimport { existsSync, readFileSync } from 'node:fs'\nimport {\n  cp,\n  lstat,\n  mkdir,\n  mkdtemp,\n  readFile,\n  readlink,\n  readdir,\n  rename,\n  rm,\n  writeFile,\n} from 'node:fs/promises'\nimport { tmpdir } from 'node:os'\nimport { basename, dirname, isAbsolute, join, relative, resolve, sep } from 'node:path'\nimport { fileURLToPath } from 'node:url'\nimport { promisify } from 'node:util'\nimport type {\n  BenchScore,\n  JudgeArtifactFileReceipt,\n  JudgeArtifactReceipt,\n} from './types'\n\nconst execFileAsync = promisify(execFile)\n\n/** Locate the package by identity because source files and compiled chunks have different depths. */\nfunction resolveBenchRoot(moduleUrl: string): string {\n  let current = dirname(fileURLToPath(moduleUrl))\n  while (true) {\n    const manifestPath = join(current, 'package.json')\n    if (existsSync(manifestPath)) {\n      const manifest = JSON.parse(readFileSync(manifestPath, 'utf8')) as { name?: unknown }\n      if (manifest.name === '@tangle-network/agent-bench') return current\n    }\n    const parent = dirname(current)\n    if (parent === current) {\n      throw new Error(`Unable to locate @tangle-network/agent-bench from ${moduleUrl}`)\n    }\n    current = parent\n  }\n}\n\n/** Package root for agent-bench, so `.venv`, scripts, and fixtures resolve. */\nexport const benchRoot = resolveBenchRoot(import.meta.url)\n\n/** Resolve the shared interpreter without requiring an installed package to contain a venv. */\nexport function resolveBenchPython(\n  env: Readonly<{ AGENT_BENCH_PYTHON?: string }> = process.env,\n  root: string = benchRoot,\n): string {\n  const configured = env.AGENT_BENCH_PYTHON\n  if (configured === undefined) return join(root, '.venv', 'bin', 'python')\n  if (!isAbsolute(configured)) {\n    throw new Error('AGENT_BENCH_PYTHON must be an absolute path')\n  }\n  return configured\n}\n\n/** The shared interpreter every Python-backed evaluator runs through. */\nexport const venvPython = resolveBenchPython()\n\n/** Interpreter for a NAMED isolated venv (e.g. `.venv-commit0`). Benches whose pip\n *  deps conflict with the shared `.venv` (commit0 downgrades pydantic/sqlalchemy)\n *  get their own venv and pass its python explicitly — keeping the shared one clean. */\nexport const venvPythonAt = (venvDir: string): string => join(benchRoot, venvDir, 'bin', 'python')\n/** Report/transcript reads are large; 256 MiB matches the SWE harness budget. */\nexport const bigBuffer = 1024 * 1024 * 256\n\n/** Path to a named executable inside the bench venv (e.g. `venvBin('tb')`). */\nexport function venvBin(name: string): string {\n  return join(benchRoot, '.venv', 'bin', name)\n}\n\n/**\n * Run the bench venv python with an inline script (`-c`); return stdout. Throws\n * (with stderr) on a nonzero exit — the loaders rely on this to fail loud rather\n * than parse a partial dump.\n */\nexport async function runVenvPython(\n  script: string,\n  args: string[] = [],\n  timeoutMs = 0,\n  python: string = venvPython,\n): Promise<string> {\n  const { stdout } = await execFileAsync(python, ['-c', script, ...args], {\n    maxBuffer: bigBuffer,\n    timeout: timeoutMs,\n  })\n  return stdout\n}\n\n/**\n * Preflight a python-backed harness: import the module(s) and (optionally) ping\n * Docker, all inside the bench venv. On failure THROWS the captured error joined\n * to the adapter's `fix` guidance — the contract every code-bench preflight wants.\n */\nexport async function preflightVenvImports(opts: {\n  /** Module names to `import` (e.g. ['swebench']); '' entries are ignored. */\n  modules: string[]\n  /** Also `docker.from_env().ping()` — true for Docker-backed evaluators. */\n  requireDocker?: boolean\n  /** Actionable remediation appended to the thrown message. */\n  fix: string\n  /** Override the interpreter (e.g. an isolated `.venv-commit0`). Default: shared `.venv`. */\n  python?: string\n}): Promise<void> {\n  const imports = opts.modules.filter((m) => m.length > 0)\n  const lines = [...imports.map((m) => `import ${m}`)]\n  if (opts.requireDocker) lines.push('import docker', 'docker.from_env().ping()')\n  lines.push(\"print('ok')\")\n  try {\n    await runVenvPython(lines.join('\\n'), [], 0, opts.python ?? venvPython)\n  } catch (err) {\n    const msg = err instanceof Error ? err.message : String(err)\n    throw new Error(`${msg}\\n${opts.fix}`)\n  }\n}\n\n/**\n * Run a bench-local python driver script (in the bench venv) while piping\n * `input` to its stdin, returning stdout. The driver's verdict JSON is its LAST\n * stdout line; callers parse that and inspect an `error` field (fail loud).\n *\n * Uses spawn + an explicit `stdin.end(input)` rather than promisify(execFile)'s\n * `input` option, because that option is NOT honored by async execFile — stdin\n * is left open and a driver that does `sys.stdin.read()` blocks forever. The\n * artifact-piping judges (commit0, appworld) MUST go through this.\n */\nexport function runVenvScriptStdin(\n  scriptPath: string,\n  args: string[],\n  input: string,\n  opts: { cwd?: string; timeoutMs?: number; python?: string } = {},\n): Promise<string> {\n  return new Promise<string>((resolve, reject) => {\n    const child = spawn(opts.python ?? venvPython, [scriptPath, ...args], {\n      cwd: opts.cwd ?? benchRoot,\n      ...(opts.timeoutMs ? { timeout: opts.timeoutMs } : {}),\n    })\n    let stdout = ''\n    let stderr = ''\n    let bytes = 0\n    child.stdout.on('data', (c: Buffer) => {\n      bytes += c.length\n      if (bytes <= bigBuffer) stdout += c.toString('utf8')\n    })\n    child.stderr.on('data', (c: Buffer) => {\n      stderr += c.toString('utf8')\n    })\n    child.on('error', (err) => reject(err))\n    child.on('close', (code) => {\n      if (code === 0) resolve(stdout)\n      else reject(new Error((stderr || stdout || `exit ${code}`).slice(0, 1500)))\n    })\n    child.stdin.end(input)\n  })\n}\n\nexport interface StagedRunSpec {\n  /** mkdtemp prefix, e.g. 'swebench-' / 'commit0-'. */\n  tmpPrefix: string\n  /**\n   * Write the artifact + any harness inputs into the temp dir. Returns nothing;\n   * `argv`/`cwd` consume `dir` to point the evaluator at what was written.\n   */\n  stage(dir: string): Promise<void>\n  /** The external evaluator to spawn. `bin` defaults to the bench venv python. */\n  bin?: string\n  /** argv for the evaluator, computed from the temp `dir`. */\n  argv(dir: string): string[]\n  /** Working directory for the evaluator. Defaults to the temp `dir`. */\n  cwd?(dir: string): string\n  /** Hard timeout for the evaluator (ms); 0 = none. */\n  timeoutMs?: number\n  /**\n   * Read the evaluator's report(s) out of `dir` and map to a BenchScore. Throws\n   * if the expected report is absent/malformed (fail loud — no default score).\n   */\n  parseReport(dir: string): Promise<BenchScore>\n  /**\n   * Copy the complete evaluator directory plus raw process stdout/stderr to this\n   * caller-owned directory before cleanup. The destination must not exist.\n   */\n  capture?: StagedRunCaptureSpec\n  /** Keep the temp dir on disk (debugging). Default false → always cleaned up. */\n  keepTmp?: boolean\n}\n\nexport interface StagedRunCaptureSpec {\n  /** Destination for `evaluator/`, `process/`, and the hashed `receipt.json`. */\n  destination: string\n}\n\n/** A staged run failed after any requested evidence was durably retained. */\nexport class StagedJudgeError extends Error {\n  readonly judgeArtifacts?: JudgeArtifactReceipt\n\n  constructor(message: string, judgeArtifacts?: JudgeArtifactReceipt, options?: ErrorOptions) {\n    super(message, options)\n    this.name = 'StagedJudgeError'\n    this.judgeArtifacts = judgeArtifacts\n  }\n}\n\nfunction sha256(bytes: Uint8Array): `sha256:${string}` {\n  return `sha256:${createHash('sha256').update(bytes).digest('hex')}`\n}\n\nfunction portablePath(path: string): string {\n  return path.split(sep).join('/')\n}\n\nfunction compareText(left: string, right: string): number {\n  return left < right ? -1 : left > right ? 1 : 0\n}\n\nasync function collectArtifactFiles(\n  root: string,\n  current: string,\n): Promise<JudgeArtifactFileReceipt[]> {\n  const absolute = join(root, current)\n  const entries = await readdir(absolute, { withFileTypes: true })\n  const files: JudgeArtifactFileReceipt[] = []\n  for (const entry of entries.sort((left, right) => compareText(left.name, right.name))) {\n    const relativePath = join(current, entry.name)\n    const path = join(root, relativePath)\n    if (entry.isDirectory()) {\n      files.push(...await collectArtifactFiles(root, relativePath))\n      continue\n    }\n    if (entry.isFile()) {\n      const bytes = await readFile(path)\n      files.push({\n        path: portablePath(relativePath),\n        byteLength: bytes.byteLength,\n        sha256: sha256(bytes),\n        kind: 'file',\n      })\n      continue\n    }\n    if (entry.isSymbolicLink()) {\n      const targetBytes = await readlink(path, { encoding: 'buffer' })\n      files.push({\n        path: portablePath(relativePath),\n        byteLength: targetBytes.byteLength,\n        sha256: sha256(targetBytes),\n        kind: 'symlink',\n      })\n      continue\n    }\n    throw new Error(`staged judge capture does not support ${relativePath}`)\n  }\n  return files\n}\n\nfunction isWithin(parent: string, candidate: string): boolean {\n  const path = relative(parent, candidate)\n  return path === '' || (!path.startsWith(`..${sep}`) && path !== '..' && !isAbsolute(path))\n}\n\nasync function assertDestinationAbsent(destination: string): Promise<void> {\n  try {\n    await lstat(destination)\n  } catch (error) {\n    if ((error as NodeJS.ErrnoException).code === 'ENOENT') return\n    throw error\n  }\n  throw new Error(`staged judge capture destination already exists: ${destination}`)\n}\n\nasync function captureStagedRun(\n  sourceDirectory: string,\n  spec: StagedRunCaptureSpec,\n  processOutput: Readonly<{ stdout: Buffer; stderr: Buffer }>,\n  evaluatorSucceeded: boolean,\n): Promise<JudgeArtifactReceipt> {\n  const source = resolve(sourceDirectory)\n  const destination = resolve(spec.destination)\n  if (isWithin(source, destination)) {\n    throw new Error('staged judge capture destination must be outside the evaluator directory')\n  }\n  await mkdir(dirname(destination), { recursive: true })\n  await assertDestinationAbsent(destination)\n  const staging = await mkdtemp(join(dirname(destination), `.${basename(destination)}-capture-`))\n  try {\n    await cp(source, join(staging, 'evaluator'), {\n      recursive: true,\n      errorOnExist: true,\n      force: false,\n      preserveTimestamps: true,\n      verbatimSymlinks: true,\n    })\n    await mkdir(join(staging, 'process'))\n    await writeFile(join(staging, 'process', 'stdout.bin'), processOutput.stdout)\n    await writeFile(join(staging, 'process', 'stderr.bin'), processOutput.stderr)\n\n    const files = [\n      ...await collectArtifactFiles(staging, 'evaluator'),\n      ...await collectArtifactFiles(staging, 'process'),\n    ].sort((left, right) => compareText(left.path, right.path))\n    const byteLength = files.reduce((total, file) => total + file.byteLength, 0)\n    const treeBytes = Buffer.from(\n      files\n        .map((file) => `${file.path}\\0${file.kind}\\0${file.byteLength}\\0${file.sha256}\\n`)\n        .join(''),\n      'utf8',\n    )\n    const receipt: JudgeArtifactReceipt = {\n      schema: 'agent-bench/judge-artifacts/v1',\n      directory: destination,\n      evaluatorDirectory: join(destination, 'evaluator'),\n      manifestPath: join(destination, 'receipt.json'),\n      evaluatorSucceeded,\n      files,\n      fileCount: files.length,\n      byteLength,\n      treeSha256: sha256(treeBytes),\n    }\n    await writeFile(join(staging, 'receipt.json'), `${JSON.stringify(receipt, null, 2)}\\n`)\n    await rename(staging, destination)\n    return receipt\n  } catch (error) {\n    await rm(staging, { recursive: true, force: true }).catch(() => {})\n    throw error\n  }\n}\n\nfunction processBytes(value: unknown): Buffer {\n  if (Buffer.isBuffer(value)) return value\n  if (value === undefined || value === null) return Buffer.alloc(0)\n  return Buffer.from(String(value), 'utf8')\n}\n\n/**\n * The shared judge body: mkdtemp → stage → spawn evaluator → parseReport →\n * cleanup. The evaluator's stdout/stderr is surfaced on failure; the temp dir is\n * always removed in `finally` unless `keepTmp`.\n */\nexport async function runStagedJudge(spec: StagedRunSpec): Promise<BenchScore> {\n  const dir = await mkdtemp(join(tmpdir(), spec.tmpPrefix))\n  let stdout: Buffer<ArrayBufferLike> = Buffer.alloc(0)\n  let stderr: Buffer<ArrayBufferLike> = Buffer.alloc(0)\n  let evaluatorSucceeded = false\n  let score: BenchScore | undefined\n  let failure: unknown\n  try {\n    try {\n      await spec.stage(dir)\n      const bin = spec.bin ?? venvPython\n      const argv = spec.argv(dir)\n      try {\n        const output = await execFileAsync(bin, argv, {\n          cwd: spec.cwd ? spec.cwd(dir) : dir,\n          encoding: 'buffer',\n          maxBuffer: bigBuffer,\n          ...(spec.timeoutMs ? { timeout: spec.timeoutMs } : {}),\n        })\n        stdout = processBytes(output.stdout)\n        stderr = processBytes(output.stderr)\n        evaluatorSucceeded = true\n      } catch (err) {\n        const e = err as { stderr?: unknown; stdout?: unknown; message?: string }\n        stdout = processBytes(e.stdout)\n        stderr = processBytes(e.stderr)\n        const detail = processBytes(e.stderr ?? e.stdout ?? e.message ?? String(err))\n          .toString('utf8')\n          .slice(0, 2000)\n        throw new Error(`${spec.tmpPrefix.replace(/-$/, '')} evaluator failed (${bin} ${argv.join(' ')}):\\n${detail}`)\n      }\n      score = await spec.parseReport(dir)\n    } catch (err) {\n      failure = err\n    }\n  } finally {\n    let judgeArtifacts: JudgeArtifactReceipt | undefined\n    if (spec.capture) {\n      try {\n        judgeArtifacts = await captureStagedRun(\n          dir,\n          spec.capture,\n          { stdout, stderr },\n          evaluatorSucceeded,\n        )\n      } catch (captureError) {\n        failure = new Error(\n          `staged judge failed to capture evaluator artifacts: ${captureError instanceof Error ? captureError.message : captureError}`,\n          { cause: failure ?? captureError },\n        )\n      }\n    }\n    if (!spec.keepTmp) await rm(dir, { recursive: true, force: true }).catch(() => {})\n    if (failure) {\n      throw new StagedJudgeError(\n        failure instanceof Error ? failure.message : String(failure),\n        judgeArtifacts,\n        { cause: failure },\n      )\n    }\n    if (!score) throw new StagedJudgeError('staged judge completed without a score', judgeArtifacts)\n    return judgeArtifacts ? { ...score, judgeArtifacts } : score\n  }\n}\n\n/** Read + JSON.parse a report file from a staged run; throws with the path on failure. */\nexport async function readJsonReport<T>(path: string): Promise<T> {\n  let raw: string\n  try {\n    raw = await readFile(path, 'utf8')\n  } catch (err) {\n    throw new Error(`expected report not written: ${path} (${err instanceof Error ? err.message : err})`)\n  }\n  try {\n    return JSON.parse(raw) as T\n  } catch (err) {\n    throw new Error(`report not valid JSON at ${path}: ${err instanceof Error ? err.message : err}`)\n  }\n}\n\n/** Write a UTF-8 file into a staged dir (artifact / preds.json / attempt.sh / …). */\nexport async function stageFile(path: string, content: string): Promise<void> {\n  await writeFile(path, content)\n}\n\n/** Sanitize an instance id into a filesystem/run-id-safe token. */\nexport function safeRunId(prefix: string, id: string): string {\n  return `${prefix}-${id}`.replace(/[^a-zA-Z0-9_.-]/g, '_')\n}\n"],"mappings":";AAkBA,SAAS,UAAU,aAAa;AAChC,SAAS,kBAAkB;AAC3B,SAAS,YAAY,oBAAoB;AACzC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,cAAc;AACvB,SAAS,UAAU,SAAS,YAAY,MAAM,UAAU,SAAS,WAAW;AAC5E,SAAS,qBAAqB;AAC9B,SAAS,iBAAiB;AAO1B,IAAM,gBAAgB,UAAU,QAAQ;AAGxC,SAAS,iBAAiB,WAA2B;AACnD,MAAI,UAAU,QAAQ,cAAc,SAAS,CAAC;AAC9C,SAAO,MAAM;AACX,UAAM,eAAe,KAAK,SAAS,cAAc;AACjD,QAAI,WAAW,YAAY,GAAG;AAC5B,YAAM,WAAW,KAAK,MAAM,aAAa,cAAc,MAAM,CAAC;AAC9D,UAAI,SAAS,SAAS,8BAA+B,QAAO;AAAA,IAC9D;AACA,UAAM,SAAS,QAAQ,OAAO;AAC9B,QAAI,WAAW,SAAS;AACtB,YAAM,IAAI,MAAM,qDAAqD,SAAS,EAAE;AAAA,IAClF;AACA,cAAU;AAAA,EACZ;AACF;AAGO,IAAM,YAAY,iBAAiB,YAAY,GAAG;AAGlD,SAAS,mBACd,MAAiD,QAAQ,KACzD,OAAe,WACP;AACR,QAAM,aAAa,IAAI;AACvB,MAAI,eAAe,OAAW,QAAO,KAAK,MAAM,SAAS,OAAO,QAAQ;AACxE,MAAI,CAAC,WAAW,UAAU,GAAG;AAC3B,UAAM,IAAI,MAAM,6CAA6C;AAAA,EAC/D;AACA,SAAO;AACT;AAGO,IAAM,aAAa,mBAAmB;AAKtC,IAAM,eAAe,CAAC,YAA4B,KAAK,WAAW,SAAS,OAAO,QAAQ;AAE1F,IAAM,YAAY,OAAO,OAAO;AAGhC,SAAS,QAAQ,MAAsB;AAC5C,SAAO,KAAK,WAAW,SAAS,OAAO,IAAI;AAC7C;AAOA,eAAsB,cACpB,QACA,OAAiB,CAAC,GAClB,YAAY,GACZ,SAAiB,YACA;AACjB,QAAM,EAAE,OAAO,IAAI,MAAM,cAAc,QAAQ,CAAC,MAAM,QAAQ,GAAG,IAAI,GAAG;AAAA,IACtE,WAAW;AAAA,IACX,SAAS;AAAA,EACX,CAAC;AACD,SAAO;AACT;AAOA,eAAsB,qBAAqB,MASzB;AAChB,QAAM,UAAU,KAAK,QAAQ,OAAO,CAAC,MAAM,EAAE,SAAS,CAAC;AACvD,QAAM,QAAQ,CAAC,GAAG,QAAQ,IAAI,CAAC,MAAM,UAAU,CAAC,EAAE,CAAC;AACnD,MAAI,KAAK,cAAe,OAAM,KAAK,iBAAiB,0BAA0B;AAC9E,QAAM,KAAK,aAAa;AACxB,MAAI;AACF,UAAM,cAAc,MAAM,KAAK,IAAI,GAAG,CAAC,GAAG,GAAG,KAAK,UAAU,UAAU;AAAA,EACxE,SAAS,KAAK;AACZ,UAAM,MAAM,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AAC3D,UAAM,IAAI,MAAM,GAAG,GAAG;AAAA,EAAK,KAAK,GAAG,EAAE;AAAA,EACvC;AACF;AAYO,SAAS,mBACd,YACA,MACA,OACA,OAA8D,CAAC,GAC9C;AACjB,SAAO,IAAI,QAAgB,CAACA,UAAS,WAAW;AAC9C,UAAM,QAAQ,MAAM,KAAK,UAAU,YAAY,CAAC,YAAY,GAAG,IAAI,GAAG;AAAA,MACpE,KAAK,KAAK,OAAO;AAAA,MACjB,GAAI,KAAK,YAAY,EAAE,SAAS,KAAK,UAAU,IAAI,CAAC;AAAA,IACtD,CAAC;AACD,QAAI,SAAS;AACb,QAAI,SAAS;AACb,QAAI,QAAQ;AACZ,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AACrC,eAAS,EAAE;AACX,UAAI,SAAS,UAAW,WAAU,EAAE,SAAS,MAAM;AAAA,IACrD,CAAC;AACD,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AACrC,gBAAU,EAAE,SAAS,MAAM;AAAA,IAC7B,CAAC;AACD,UAAM,GAAG,SAAS,CAAC,QAAQ,OAAO,GAAG,CAAC;AACtC,UAAM,GAAG,SAAS,CAAC,SAAS;AAC1B,UAAI,SAAS,EAAG,CAAAA,SAAQ,MAAM;AAAA,UACzB,QAAO,IAAI,OAAO,UAAU,UAAU,QAAQ,IAAI,IAAI,MAAM,GAAG,IAAI,CAAC,CAAC;AAAA,IAC5E,CAAC;AACD,UAAM,MAAM,IAAI,KAAK;AAAA,EACvB,CAAC;AACH;AAsCO,IAAM,mBAAN,cAA+B,MAAM;AAAA,EACjC;AAAA,EAET,YAAY,SAAiB,gBAAuC,SAAwB;AAC1F,UAAM,SAAS,OAAO;AACtB,SAAK,OAAO;AACZ,SAAK,iBAAiB;AAAA,EACxB;AACF;AAEA,SAAS,OAAO,OAAuC;AACrD,SAAO,UAAU,WAAW,QAAQ,EAAE,OAAO,KAAK,EAAE,OAAO,KAAK,CAAC;AACnE;AAEA,SAAS,aAAa,MAAsB;AAC1C,SAAO,KAAK,MAAM,GAAG,EAAE,KAAK,GAAG;AACjC;AAEA,SAAS,YAAY,MAAc,OAAuB;AACxD,SAAO,OAAO,QAAQ,KAAK,OAAO,QAAQ,IAAI;AAChD;AAEA,eAAe,qBACb,MACA,SACqC;AACrC,QAAM,WAAW,KAAK,MAAM,OAAO;AACnC,QAAM,UAAU,MAAM,QAAQ,UAAU,EAAE,eAAe,KAAK,CAAC;AAC/D,QAAM,QAAoC,CAAC;AAC3C,aAAW,SAAS,QAAQ,KAAK,CAAC,MAAM,UAAU,YAAY,KAAK,MAAM,MAAM,IAAI,CAAC,GAAG;AACrF,UAAM,eAAe,KAAK,SAAS,MAAM,IAAI;AAC7C,UAAM,OAAO,KAAK,MAAM,YAAY;AACpC,QAAI,MAAM,YAAY,GAAG;AACvB,YAAM,KAAK,GAAG,MAAM,qBAAqB,MAAM,YAAY,CAAC;AAC5D;AAAA,IACF;AACA,QAAI,MAAM,OAAO,GAAG;AAClB,YAAM,QAAQ,MAAM,SAAS,IAAI;AACjC,YAAM,KAAK;AAAA,QACT,MAAM,aAAa,YAAY;AAAA,QAC/B,YAAY,MAAM;AAAA,QAClB,QAAQ,OAAO,KAAK;AAAA,QACpB,MAAM;AAAA,MACR,CAAC;AACD;AAAA,IACF;AACA,QAAI,MAAM,eAAe,GAAG;AAC1B,YAAM,cAAc,MAAM,SAAS,MAAM,EAAE,UAAU,SAAS,CAAC;AAC/D,YAAM,KAAK;AAAA,QACT,MAAM,aAAa,YAAY;AAAA,QAC/B,YAAY,YAAY;AAAA,QACxB,QAAQ,OAAO,WAAW;AAAA,QAC1B,MAAM;AAAA,MACR,CAAC;AACD;AAAA,IACF;AACA,UAAM,IAAI,MAAM,yCAAyC,YAAY,EAAE;AAAA,EACzE;AACA,SAAO;AACT;AAEA,SAAS,SAAS,QAAgB,WAA4B;AAC5D,QAAM,OAAO,SAAS,QAAQ,SAAS;AACvC,SAAO,SAAS,MAAO,CAAC,KAAK,WAAW,KAAK,GAAG,EAAE,KAAK,SAAS,QAAQ,CAAC,WAAW,IAAI;AAC1F;AAEA,eAAe,wBAAwB,aAAoC;AACzE,MAAI;AACF,UAAM,MAAM,WAAW;AAAA,EACzB,SAAS,OAAO;AACd,QAAK,MAAgC,SAAS,SAAU;AACxD,UAAM;AAAA,EACR;AACA,QAAM,IAAI,MAAM,oDAAoD,WAAW,EAAE;AACnF;AAEA,eAAe,iBACb,iBACA,MACA,eACA,oBAC+B;AAC/B,QAAM,SAAS,QAAQ,eAAe;AACtC,QAAM,cAAc,QAAQ,KAAK,WAAW;AAC5C,MAAI,SAAS,QAAQ,WAAW,GAAG;AACjC,UAAM,IAAI,MAAM,0EAA0E;AAAA,EAC5F;AACA,QAAM,MAAM,QAAQ,WAAW,GAAG,EAAE,WAAW,KAAK,CAAC;AACrD,QAAM,wBAAwB,WAAW;AACzC,QAAM,UAAU,MAAM,QAAQ,KAAK,QAAQ,WAAW,GAAG,IAAI,SAAS,WAAW,CAAC,WAAW,CAAC;AAC9F,MAAI;AACF,UAAM,GAAG,QAAQ,KAAK,SAAS,WAAW,GAAG;AAAA,MAC3C,WAAW;AAAA,MACX,cAAc;AAAA,MACd,OAAO;AAAA,MACP,oBAAoB;AAAA,MACpB,kBAAkB;AAAA,IACpB,CAAC;AACD,UAAM,MAAM,KAAK,SAAS,SAAS,CAAC;AACpC,UAAM,UAAU,KAAK,SAAS,WAAW,YAAY,GAAG,cAAc,MAAM;AAC5E,UAAM,UAAU,KAAK,SAAS,WAAW,YAAY,GAAG,cAAc,MAAM;AAE5E,UAAM,QAAQ;AAAA,MACZ,GAAG,MAAM,qBAAqB,SAAS,WAAW;AAAA,MAClD,GAAG,MAAM,qBAAqB,SAAS,SAAS;AAAA,IAClD,EAAE,KAAK,CAAC,MAAM,UAAU,YAAY,KAAK,MAAM,MAAM,IAAI,CAAC;AAC1D,UAAM,aAAa,MAAM,OAAO,CAAC,OAAO,SAAS,QAAQ,KAAK,YAAY,CAAC;AAC3E,UAAM,YAAY,OAAO;AAAA,MACvB,MACG,IAAI,CAAC,SAAS,GAAG,KAAK,IAAI,KAAK,KAAK,IAAI,KAAK,KAAK,UAAU,KAAK,KAAK,MAAM;AAAA,CAAI,EAChF,KAAK,EAAE;AAAA,MACV;AAAA,IACF;AACA,UAAM,UAAgC;AAAA,MACpC,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,oBAAoB,KAAK,aAAa,WAAW;AAAA,MACjD,cAAc,KAAK,aAAa,cAAc;AAAA,MAC9C;AAAA,MACA;AAAA,MACA,WAAW,MAAM;AAAA,MACjB;AAAA,MACA,YAAY,OAAO,SAAS;AAAA,IAC9B;AACA,UAAM,UAAU,KAAK,SAAS,cAAc,GAAG,GAAG,KAAK,UAAU,SAAS,MAAM,CAAC,CAAC;AAAA,CAAI;AACtF,UAAM,OAAO,SAAS,WAAW;AACjC,WAAO;AAAA,EACT,SAAS,OAAO;AACd,UAAM,GAAG,SAAS,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AAClE,UAAM;AAAA,EACR;AACF;AAEA,SAAS,aAAa,OAAwB;AAC5C,MAAI,OAAO,SAAS,KAAK,EAAG,QAAO;AACnC,MAAI,UAAU,UAAa,UAAU,KAAM,QAAO,OAAO,MAAM,CAAC;AAChE,SAAO,OAAO,KAAK,OAAO,KAAK,GAAG,MAAM;AAC1C;AAOA,eAAsB,eAAe,MAA0C;AAC7E,QAAM,MAAM,MAAM,QAAQ,KAAK,OAAO,GAAG,KAAK,SAAS,CAAC;AACxD,MAAI,SAAkC,OAAO,MAAM,CAAC;AACpD,MAAI,SAAkC,OAAO,MAAM,CAAC;AACpD,MAAI,qBAAqB;AACzB,MAAI;AACJ,MAAI;AACJ,MAAI;AACF,QAAI;AACF,YAAM,KAAK,MAAM,GAAG;AACpB,YAAM,MAAM,KAAK,OAAO;AACxB,YAAM,OAAO,KAAK,KAAK,GAAG;AAC1B,UAAI;AACF,cAAM,SAAS,MAAM,cAAc,KAAK,MAAM;AAAA,UAC5C,KAAK,KAAK,MAAM,KAAK,IAAI,GAAG,IAAI;AAAA,UAChC,UAAU;AAAA,UACV,WAAW;AAAA,UACX,GAAI,KAAK,YAAY,EAAE,SAAS,KAAK,UAAU,IAAI,CAAC;AAAA,QACtD,CAAC;AACD,iBAAS,aAAa,OAAO,MAAM;AACnC,iBAAS,aAAa,OAAO,MAAM;AACnC,6BAAqB;AAAA,MACvB,SAAS,KAAK;AACZ,cAAM,IAAI;AACV,iBAAS,aAAa,EAAE,MAAM;AAC9B,iBAAS,aAAa,EAAE,MAAM;AAC9B,cAAM,SAAS,aAAa,EAAE,UAAU,EAAE,UAAU,EAAE,WAAW,OAAO,GAAG,CAAC,EACzE,SAAS,MAAM,EACf,MAAM,GAAG,GAAI;AAChB,cAAM,IAAI,MAAM,GAAG,KAAK,UAAU,QAAQ,MAAM,EAAE,CAAC,sBAAsB,GAAG,IAAI,KAAK,KAAK,GAAG,CAAC;AAAA,EAAO,MAAM,EAAE;AAAA,MAC/G;AACA,cAAQ,MAAM,KAAK,YAAY,GAAG;AAAA,IACpC,SAAS,KAAK;AACZ,gBAAU;AAAA,IACZ;AAAA,EACF,UAAE;AACA,QAAI;AACJ,QAAI,KAAK,SAAS;AAChB,UAAI;AACF,yBAAiB,MAAM;AAAA,UACrB;AAAA,UACA,KAAK;AAAA,UACL,EAAE,QAAQ,OAAO;AAAA,UACjB;AAAA,QACF;AAAA,MACF,SAAS,cAAc;AACrB,kBAAU,IAAI;AAAA,UACZ,uDAAuD,wBAAwB,QAAQ,aAAa,UAAU,YAAY;AAAA,UAC1H,EAAE,OAAO,WAAW,aAAa;AAAA,QACnC;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,KAAK,QAAS,OAAM,GAAG,KAAK,EAAE,WAAW,MAAM,OAAO,KAAK,CAAC,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AACjF,QAAI,SAAS;AACX,YAAM,IAAI;AAAA,QACR,mBAAmB,QAAQ,QAAQ,UAAU,OAAO,OAAO;AAAA,QAC3D;AAAA,QACA,EAAE,OAAO,QAAQ;AAAA,MACnB;AAAA,IACF;AACA,QAAI,CAAC,MAAO,OAAM,IAAI,iBAAiB,0CAA0C,cAAc;AAC/F,WAAO,iBAAiB,EAAE,GAAG,OAAO,eAAe,IAAI;AAAA,EACzD;AACF;AAGA,eAAsB,eAAkB,MAA0B;AAChE,MAAI;AACJ,MAAI;AACF,UAAM,MAAM,SAAS,MAAM,MAAM;AAAA,EACnC,SAAS,KAAK;AACZ,UAAM,IAAI,MAAM,gCAAgC,IAAI,KAAK,eAAe,QAAQ,IAAI,UAAU,GAAG,GAAG;AAAA,EACtG;AACA,MAAI;AACF,WAAO,KAAK,MAAM,GAAG;AAAA,EACvB,SAAS,KAAK;AACZ,UAAM,IAAI,MAAM,4BAA4B,IAAI,KAAK,eAAe,QAAQ,IAAI,UAAU,GAAG,EAAE;AAAA,EACjG;AACF;AAGA,eAAsB,UAAU,MAAc,SAAgC;AAC5E,QAAM,UAAU,MAAM,OAAO;AAC/B;AAGO,SAAS,UAAU,QAAgB,IAAoB;AAC5D,SAAO,GAAG,MAAM,IAAI,EAAE,GAAG,QAAQ,oBAAoB,GAAG;AAC1D;","names":["resolve"]}

package/dist/{chunk-2PVVP7GN.js → chunk-XKEFIFIC.js}

@@ -1,6 +1,6 @@
 import {
   benchRoot
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/toollm.ts
 import { readFile } from "fs/promises";
@@ -194,4 +194,4 @@ export {
   toollmOutput,
   createToolLlmAdapter
 };
-//# sourceMappingURL=chunk-2PVVP7GN.js.map
+//# sourceMappingURL=chunk-XKEFIFIC.js.map

package/dist/{chunk-JRWWGMK7.js → chunk-XYA4XSNU.js}

@@ -7,7 +7,7 @@ import {
   safeRunId,
   stageFile,
   venvPythonAt
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/terminal-bench.ts
 import { join } from "path";
@@ -145,4 +145,4 @@ function createTerminalBenchAdapter() {
 export {
   createTerminalBenchAdapter
 };
-//# sourceMappingURL=chunk-JRWWGMK7.js.map
+//# sourceMappingURL=chunk-XYA4XSNU.js.map

package/dist/{chunk-X5YKXC6V.js → chunk-YSMEKBTD.js}

@@ -1,7 +1,7 @@
 import {
   benchRoot,
   runVenvScriptStdin
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/enterpriseops-gym.ts
 import { mkdir, readFile, rm, writeFile } from "fs/promises";
@@ -208,4 +208,4 @@ export {
   enterpriseOpsTranscriptOutput,
   createEnterpriseOpsGymAdapter
 };
-//# sourceMappingURL=chunk-X5YKXC6V.js.map
+//# sourceMappingURL=chunk-YSMEKBTD.js.map

package/dist/{chunk-2XU6OGEN.js → chunk-Z4TZ76N7.js}

@@ -3,7 +3,7 @@ import {
   preflightVenvImports,
   runVenvScriptStdin,
   venvPythonAt
-} from "./chunk-LRRD7NAG.js";
+} from "./chunk-WSKWVEQB.js";
 
 // src/benchmarks/commit0.ts
 import { join } from "path";
@@ -167,4 +167,4 @@ export {
   commit0DiffOutput,
   createCommit0Adapter
 };
-//# sourceMappingURL=chunk-2XU6OGEN.js.map
+//# sourceMappingURL=chunk-Z4TZ76N7.js.map