@tangle-network/agent-eval 0.23.1 → 0.24.0

package/dist/chunk-OWLAAMME.js

@@ -0,0 +1,250 @@
+import {
+  TraceEmitter
+} from "./chunk-TVVP3ZZQ.js";
+import {
+  ConfigError
+} from "./chunk-NG236HPC.js";
+
+// src/sandbox-harness.ts
+var vitestTestParser = {
+  id: "vitest",
+  parse(stdout) {
+    const m = stdout.match(/Tests\s+(\d+)\s+(passed|failed)(?:\s*\|\s*(\d+)\s+(passed|failed))?/i);
+    if (!m) return void 0;
+    let passed = 0;
+    let failed = 0;
+    const a = parseInt(m[1], 10);
+    const aLabel = m[2].toLowerCase();
+    if (aLabel === "passed") passed += a;
+    else failed += a;
+    if (m[3] && m[4]) {
+      const b = parseInt(m[3], 10);
+      if (m[4].toLowerCase() === "passed") passed += b;
+      else failed += b;
+    }
+    return { testsTotal: passed + failed, testsPassed: passed };
+  }
+};
+var pytestTestParser = {
+  id: "pytest",
+  parse(stdout) {
+    const total = stdout.match(/collected\s+(\d+)\s+items?/i);
+    const passed = stdout.match(/(\d+)\s+passed/);
+    if (!total || !passed) return void 0;
+    return { testsTotal: parseInt(total[1], 10), testsPassed: parseInt(passed[1], 10) };
+  }
+};
+var jestTestParser = {
+  id: "jest",
+  parse(stdout) {
+    const m = stdout.match(/Tests:\s+(?:(\d+)\s+failed[^,]*,\s*)?(\d+)\s+passed,\s+(\d+)\s+total/i);
+    if (!m) return void 0;
+    return { testsTotal: parseInt(m[3], 10), testsPassed: parseInt(m[2], 10) };
+  }
+};
+function composeParsers(...parsers) {
+  return {
+    id: parsers.map((p) => p.id).join("|"),
+    parse(stdout, stderr, exitCode) {
+      for (const p of parsers) {
+        const res = p.parse(stdout, stderr, exitCode);
+        if (res) return res;
+      }
+      return void 0;
+    }
+  };
+}
+var SubprocessSandboxDriver = class {
+  id = "subprocess";
+  defaultCwd;
+  defaultEnv;
+  constructor(options = {}) {
+    this.defaultCwd = options.cwd;
+    this.defaultEnv = options.env;
+  }
+  async exec(phase, command, config) {
+    const { spawn } = await import("child_process");
+    const start = Date.now();
+    const effectiveCwd = config.cwd ?? this.defaultCwd;
+    const effectiveEnv = { ...process.env, ...this.defaultEnv ?? {}, ...config.env ?? {} };
+    return await new Promise((resolve) => {
+      const child = spawn(command, {
+        shell: true,
+        cwd: effectiveCwd,
+        env: effectiveEnv
+      });
+      let stdout = "";
+      let stderr = "";
+      child.stdout?.on("data", (d) => {
+        stdout += String(d);
+      });
+      child.stderr?.on("data", (d) => {
+        stderr += String(d);
+      });
+      const timeout = setTimeout(
+        () => {
+          try {
+            child.kill("SIGKILL");
+          } catch {
+          }
+        },
+        config.timeoutMs ?? 10 * 6e4
+      );
+      child.on("close", (code) => {
+        clearTimeout(timeout);
+        const wallMs = Date.now() - start;
+        const parsed = phase === "test" && config.testParser ? config.testParser.parse(stdout, stderr, code ?? 1) : void 0;
+        resolve({
+          phase,
+          exitCode: code ?? 1,
+          stdout,
+          stderr,
+          wallMs,
+          testsTotal: parsed?.testsTotal,
+          testsPassed: parsed?.testsPassed
+        });
+      });
+      child.on("error", (err) => {
+        clearTimeout(timeout);
+        const wallMs = Date.now() - start;
+        resolve({ phase, exitCode: 127, stdout, stderr: stderr + String(err), wallMs });
+      });
+    });
+  }
+};
+var DockerSandboxDriver = class {
+  id = "docker";
+  async exec(phase, command, config) {
+    if (!config.image) throw new ConfigError("DockerSandboxDriver requires config.image");
+    const sub = new SubprocessSandboxDriver();
+    const envArgs = Object.entries(config.env ?? {}).map(([k, v]) => `-e ${shellQuote(k)}=${shellQuote(v)}`).join(" ");
+    const wrapped = `docker run --rm ${envArgs} ${shellQuote(config.image)} sh -c ${shellQuote(command)}`;
+    return sub.exec(phase, wrapped, { ...config, env: void 0 });
+  }
+};
+function shellQuote(v) {
+  if (/^[A-Za-z0-9_\-/.@:=]+$/.test(v)) return v;
+  return `'${v.replace(/'/g, `'\\''`)}'`;
+}
+var SandboxHarness = class {
+  driver;
+  constructor(driver = new SubprocessSandboxDriver()) {
+    this.driver = driver;
+  }
+  async run(config, emitter) {
+    const handle = await emitter.sandbox({
+      name: `sandbox(${this.driver.id})`,
+      image: config.image,
+      command: [config.setupCommand, config.runCommand, config.testCommand].filter(Boolean).join(" && ")
+    });
+    const result = { passed: false, totalWallMs: 0, score: 0 };
+    try {
+      if (config.setupCommand) {
+        result.setup = await this.driver.exec("setup", config.setupCommand, config);
+        result.totalWallMs += result.setup.wallMs;
+        if (result.setup.exitCode !== 0) {
+          await handle.fail(`setup failed (exit ${result.setup.exitCode})`, {
+            exitCode: result.setup.exitCode,
+            wallMs: result.totalWallMs
+          });
+          return result;
+        }
+      }
+      if (config.runCommand) {
+        result.run = await this.driver.exec("run", config.runCommand, config);
+        result.totalWallMs += result.run.wallMs;
+        if (result.run.exitCode !== 0) {
+          await handle.fail(`run failed (exit ${result.run.exitCode})`, {
+            exitCode: result.run.exitCode,
+            wallMs: result.totalWallMs
+          });
+          return result;
+        }
+      }
+      if (config.testCommand) {
+        result.test = await this.driver.exec("test", config.testCommand, config);
+        result.totalWallMs += result.test.wallMs;
+        const passed = result.test.exitCode === 0;
+        result.passed = passed;
+        if (result.test.testsTotal !== void 0 && result.test.testsTotal > 0) {
+          result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal;
+        } else {
+          result.score = passed ? 1 : 0;
+        }
+        await handle.end({
+          exitCode: result.test.exitCode,
+          testsTotal: result.test.testsTotal,
+          testsPassed: result.test.testsPassed,
+          wallMs: result.totalWallMs,
+          status: passed ? "ok" : "error"
+        });
+      } else {
+        result.passed = true;
+        result.score = 1;
+        await handle.end({ wallMs: result.totalWallMs });
+      }
+    } catch (err) {
+      await handle.fail(err instanceof Error ? err : String(err));
+      throw err;
+    }
+    return result;
+  }
+};
+
+// src/test-graded-scenario.ts
+async function runTestGradedScenario(scenario, store, options = {}) {
+  const emitter = new TraceEmitter(store);
+  await emitter.startRun({
+    scenarioId: scenario.id,
+    variantId: options.variantId,
+    datasetVersion: scenario.datasetVersion,
+    tags: scenario.tags,
+    ...options.provenance
+  });
+  const harness = new SandboxHarness(options.driver);
+  const result = await harness.run(scenario.harness, emitter);
+  const threshold = scenario.passThreshold ?? 1;
+  const pass = result.passed && result.score >= threshold;
+  const setupFailed = result.setup !== void 0 && result.setup.exitCode !== 0;
+  const runFailed = result.run !== void 0 && result.run.exitCode !== 0;
+  const testFailed = result.test !== void 0 && result.test.exitCode !== 0;
+  const failureClass = pass ? "success" : setupFailed || runFailed ? "sandbox_failure" : testFailed ? "format_drift" : "unknown";
+  await emitter.endRun({
+    pass,
+    score: result.score,
+    failureClass,
+    notes: pass ? void 0 : reasonForFailure(result)
+  });
+  return {
+    runId: emitter.runId,
+    scenario,
+    harness: result,
+    pass,
+    score: result.score,
+    failureClass
+  };
+}
+function reasonForFailure(result) {
+  if (result.setup && result.setup.exitCode !== 0)
+    return `setup failed: exit ${result.setup.exitCode}`;
+  if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`;
+  if (result.test) {
+    if (result.test.testsTotal !== void 0) {
+      return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`;
+    }
+    return `test exit ${result.test.exitCode}`;
+  }
+  return "no test command";
+}
+
+export {
+  vitestTestParser,
+  pytestTestParser,
+  jestTestParser,
+  composeParsers,
+  SubprocessSandboxDriver,
+  DockerSandboxDriver,
+  SandboxHarness,
+  runTestGradedScenario
+};
+//# sourceMappingURL=chunk-OWLAAMME.js.map

package/dist/chunk-OWLAAMME.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/sandbox-harness.ts","../src/test-graded-scenario.ts"],"sourcesContent":["/**\n * SandboxHarness — executes a scenario in an isolated environment and\n * emits a rich SandboxSpan into the trace.\n *\n * Two built-in drivers:\n *   - `SubprocessSandboxDriver` — spawn in a local cwd with env vars.\n *     Fast, no dependencies, fine for unit tests and most CI gates.\n *   - `DockerSandboxDriver` — lifted from tangle-router's sandbox path;\n *     shells out to `docker run`. Stronger isolation, slower startup.\n *\n * Consumers implement `SandboxDriver` for custom backends (Firecracker,\n * Cloudflare sandbox product, etc.). The harness doesn't care which.\n */\n\nimport { ConfigError } from './errors'\nimport type { TraceEmitter } from './trace/emitter'\nimport type { SandboxSpan } from './trace/schema'\n\nexport interface HarnessConfig {\n  /** Setup command (e.g. \"pnpm install\"). Non-zero exit fails the run. */\n  setupCommand?: string\n  /** Run command (e.g. \"pnpm build\"). */\n  runCommand?: string\n  /** Test command (e.g. \"pnpm test --run\"). Drives the test count + pass count. */\n  testCommand?: string\n  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */\n  cwd?: string\n  /** Max wall-clock per phase in ms. Default 10 minutes. */\n  timeoutMs?: number\n  /** Image for the docker driver. */\n  image?: string\n  /** Extra env vars (validated; shell-escaped). */\n  env?: Record<string, string>\n  /** Parser for the test output — maps stdout/stderr/exit code → pass count. */\n  testParser?: TestOutputParser\n}\n\nexport interface TestOutputParser {\n  id: string\n  parse(\n    stdout: string,\n    stderr: string,\n    exitCode: number,\n  ): { testsTotal: number; testsPassed: number } | undefined\n}\n\nexport interface SandboxResult {\n  phase: 'setup' | 'run' | 'test'\n  exitCode: number\n  stdout: string\n  stderr: string\n  wallMs: number\n  testsTotal?: number\n  testsPassed?: number\n}\n\nexport interface SandboxDriver {\n  id: string\n  exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult>\n}\n\n// ── Parsers ──────────────────────────────────────────────────────────\n\n/** Vitest default summary line: \"Tests  X passed | Y failed\". */\nexport const vitestTestParser: TestOutputParser = {\n  id: 'vitest',\n  parse(stdout) {\n    const m = stdout.match(/Tests\\s+(\\d+)\\s+(passed|failed)(?:\\s*\\|\\s*(\\d+)\\s+(passed|failed))?/i)\n    if (!m) return undefined\n    let passed = 0\n    let failed = 0\n    const a = parseInt(m[1]!, 10)\n    const aLabel = m[2]!.toLowerCase()\n    if (aLabel === 'passed') passed += a\n    else failed += a\n    if (m[3] && m[4]) {\n      const b = parseInt(m[3], 10)\n      if (m[4].toLowerCase() === 'passed') passed += b\n      else failed += b\n    }\n    return { testsTotal: passed + failed, testsPassed: passed }\n  },\n}\n\n/** Pytest default: \"collected N items\" + \" X passed, Y failed\". */\nexport const pytestTestParser: TestOutputParser = {\n  id: 'pytest',\n  parse(stdout) {\n    const total = stdout.match(/collected\\s+(\\d+)\\s+items?/i)\n    const passed = stdout.match(/(\\d+)\\s+passed/)\n    if (!total || !passed) return undefined\n    return { testsTotal: parseInt(total[1]!, 10), testsPassed: parseInt(passed[1]!, 10) }\n  },\n}\n\n/** Jest: \"Tests: X passed, Y total\" (and optional failed). */\nexport const jestTestParser: TestOutputParser = {\n  id: 'jest',\n  parse(stdout) {\n    const m = stdout.match(/Tests:\\s+(?:(\\d+)\\s+failed[^,]*,\\s*)?(\\d+)\\s+passed,\\s+(\\d+)\\s+total/i)\n    if (!m) return undefined\n    return { testsTotal: parseInt(m[3]!, 10), testsPassed: parseInt(m[2]!, 10) }\n  },\n}\n\n/** Composite parser — tries a list of parsers in order. */\nexport function composeParsers(...parsers: TestOutputParser[]): TestOutputParser {\n  return {\n    id: parsers.map((p) => p.id).join('|'),\n    parse(stdout, stderr, exitCode) {\n      for (const p of parsers) {\n        const res = p.parse(stdout, stderr, exitCode)\n        if (res) return res\n      }\n      return undefined\n    },\n  }\n}\n\n// ── Drivers ──────────────────────────────────────────────────────────\n\nexport interface SubprocessSandboxDriverOptions {\n  /**\n   * Default cwd for all `exec` calls. Used when the per-call `HarnessConfig`\n   * does not set its own `cwd`. Lets callers bind the driver to a working\n   * directory once instead of spreading cwd into every harness config —\n   * useful when the harness config is constructed far from the call site\n   * (e.g. starter-foundry's promoter passes a static HarnessConfig per\n   * family taxonomy but needs a per-run composed-scaffold cwd).\n   */\n  cwd?: string\n  /**\n   * Default env merged into every `exec` call's env (per-call `HarnessConfig.env`\n   * still wins on key collision). Same ergonomic rationale as `cwd` above.\n   */\n  env?: Record<string, string>\n}\n\nexport class SubprocessSandboxDriver implements SandboxDriver {\n  id = 'subprocess'\n  private defaultCwd?: string\n  private defaultEnv?: Record<string, string>\n\n  constructor(options: SubprocessSandboxDriverOptions = {}) {\n    this.defaultCwd = options.cwd\n    this.defaultEnv = options.env\n  }\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    const { spawn } = await import('node:child_process')\n    const start = Date.now()\n    // Per-call config wins; fall back to constructor defaults. Historically\n    // `config.cwd` was the only path, which silently dropped the constructor\n    // arg when callers passed `new SubprocessSandboxDriver({ cwd })` — the\n    // subprocess then inherited Node's cwd and e.g. ran `tsc --noEmit`\n    // against the wrong repo. Honoring the constructor `cwd` restores the\n    // invariant implied by the constructor shape.\n    const effectiveCwd = config.cwd ?? this.defaultCwd\n    const effectiveEnv = { ...process.env, ...(this.defaultEnv ?? {}), ...(config.env ?? {}) }\n    return await new Promise<SandboxResult>((resolve) => {\n      const child = spawn(command, {\n        shell: true,\n        cwd: effectiveCwd,\n        env: effectiveEnv,\n      })\n      let stdout = ''\n      let stderr = ''\n      child.stdout?.on('data', (d) => {\n        stdout += String(d)\n      })\n      child.stderr?.on('data', (d) => {\n        stderr += String(d)\n      })\n      const timeout = setTimeout(\n        () => {\n          try {\n            child.kill('SIGKILL')\n          } catch {}\n        },\n        config.timeoutMs ?? 10 * 60_000,\n      )\n      child.on('close', (code) => {\n        clearTimeout(timeout)\n        const wallMs = Date.now() - start\n        const parsed =\n          phase === 'test' && config.testParser\n            ? config.testParser.parse(stdout, stderr, code ?? 1)\n            : undefined\n        resolve({\n          phase,\n          exitCode: code ?? 1,\n          stdout,\n          stderr,\n          wallMs,\n          testsTotal: parsed?.testsTotal,\n          testsPassed: parsed?.testsPassed,\n        })\n      })\n      child.on('error', (err) => {\n        clearTimeout(timeout)\n        const wallMs = Date.now() - start\n        resolve({ phase, exitCode: 127, stdout, stderr: stderr + String(err), wallMs })\n      })\n    })\n  }\n}\n\nexport class DockerSandboxDriver implements SandboxDriver {\n  id = 'docker'\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    if (!config.image) throw new ConfigError('DockerSandboxDriver requires config.image')\n    const sub = new SubprocessSandboxDriver()\n    const envArgs = Object.entries(config.env ?? {})\n      .map(([k, v]) => `-e ${shellQuote(k)}=${shellQuote(v)}`)\n      .join(' ')\n    const wrapped = `docker run --rm ${envArgs} ${shellQuote(config.image)} sh -c ${shellQuote(command)}`\n    return sub.exec(phase, wrapped, { ...config, env: undefined })\n  }\n}\n\nfunction shellQuote(v: string): string {\n  if (/^[A-Za-z0-9_\\-/.@:=]+$/.test(v)) return v\n  return `'${v.replace(/'/g, `'\\\\''`)}'`\n}\n\n// ── Harness orchestration ────────────────────────────────────────────\n\nexport interface SandboxHarnessResult {\n  passed: boolean\n  setup?: SandboxResult\n  run?: SandboxResult\n  test?: SandboxResult\n  totalWallMs: number\n  /** Final score — 0 when no tests; otherwise testsPassed/testsTotal. */\n  score: number\n}\n\nexport class SandboxHarness {\n  private driver: SandboxDriver\n  constructor(driver: SandboxDriver = new SubprocessSandboxDriver()) {\n    this.driver = driver\n  }\n\n  async run(config: HarnessConfig, emitter: TraceEmitter): Promise<SandboxHarnessResult> {\n    const handle = await emitter.sandbox({\n      name: `sandbox(${this.driver.id})`,\n      image: config.image,\n      command: [config.setupCommand, config.runCommand, config.testCommand]\n        .filter(Boolean)\n        .join(' && '),\n    })\n    const result: SandboxHarnessResult = { passed: false, totalWallMs: 0, score: 0 }\n    try {\n      if (config.setupCommand) {\n        result.setup = await this.driver.exec('setup', config.setupCommand, config)\n        result.totalWallMs += result.setup.wallMs\n        if (result.setup.exitCode !== 0) {\n          await handle.fail(`setup failed (exit ${result.setup.exitCode})`, {\n            exitCode: result.setup.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.runCommand) {\n        result.run = await this.driver.exec('run', config.runCommand, config)\n        result.totalWallMs += result.run.wallMs\n        if (result.run.exitCode !== 0) {\n          await handle.fail(`run failed (exit ${result.run.exitCode})`, {\n            exitCode: result.run.exitCode,\n            wallMs: result.totalWallMs,\n          } as Partial<SandboxSpan>)\n          return result\n        }\n      }\n      if (config.testCommand) {\n        result.test = await this.driver.exec('test', config.testCommand, config)\n        result.totalWallMs += result.test.wallMs\n        const passed = result.test.exitCode === 0\n        result.passed = passed\n        if (result.test.testsTotal !== undefined && result.test.testsTotal > 0) {\n          result.score = (result.test.testsPassed ?? 0) / result.test.testsTotal\n        } else {\n          result.score = passed ? 1 : 0\n        }\n        await handle.end({\n          exitCode: result.test.exitCode,\n          testsTotal: result.test.testsTotal,\n          testsPassed: result.test.testsPassed,\n          wallMs: result.totalWallMs,\n          status: passed ? 'ok' : 'error',\n        } as Partial<SandboxSpan>)\n      } else {\n        result.passed = true\n        result.score = 1\n        await handle.end({ wallMs: result.totalWallMs } as Partial<SandboxSpan>)\n      }\n    } catch (err) {\n      await handle.fail(err instanceof Error ? err : String(err))\n      throw err\n    }\n    return result\n  }\n}\n","/**\n * TestGradedScenario — a scenario whose score comes from a test suite.\n *\n * This is the SWE-bench pattern generalized. The scenario ships:\n *   - fixture data (setup instructions)\n *   - a test command the harness runs\n *   - optional assertion overrides\n *\n * The runner emits a run, delegates to SandboxHarness, records the\n * outcome, and returns a structured verdict. Consumers bind their own\n * agent execution to this contract.\n */\n\nimport type { HarnessConfig, SandboxDriver, SandboxHarnessResult } from './sandbox-harness'\nimport { SandboxHarness } from './sandbox-harness'\nimport { TraceEmitter } from './trace/emitter'\nimport type { FailureClass, Run } from './trace/schema'\nimport type { TraceStore } from './trace/store'\n\nexport interface TestGradedScenario {\n  id: string\n  description?: string\n  harness: HarnessConfig\n  /** Optional pass threshold in 0..1 (default 1.0 = all tests must pass). */\n  passThreshold?: number\n  /** Provenance for dataset tracking. */\n  datasetVersion?: string\n  /** Free-form tags (difficulty, category, etc.). */\n  tags?: Record<string, string>\n}\n\nexport interface TestGradedRunOptions {\n  variantId?: string\n  driver?: SandboxDriver\n  /** Metadata recorded on the Run (codeSha, promptSha, modelFingerprint, seed). */\n  provenance?: Pick<Run, 'codeSha' | 'promptSha' | 'modelFingerprint' | 'seed' | 'envFingerprint'>\n}\n\nexport interface TestGradedRunResult {\n  runId: string\n  scenario: TestGradedScenario\n  harness: SandboxHarnessResult\n  pass: boolean\n  score: number\n  failureClass?: FailureClass\n}\n\nexport async function runTestGradedScenario(\n  scenario: TestGradedScenario,\n  store: TraceStore,\n  options: TestGradedRunOptions = {},\n): Promise<TestGradedRunResult> {\n  const emitter = new TraceEmitter(store)\n  await emitter.startRun({\n    scenarioId: scenario.id,\n    variantId: options.variantId,\n    datasetVersion: scenario.datasetVersion,\n    tags: scenario.tags,\n    ...options.provenance,\n  })\n  const harness = new SandboxHarness(options.driver)\n  const result = await harness.run(scenario.harness, emitter)\n  const threshold = scenario.passThreshold ?? 1.0\n  const pass = result.passed && result.score >= threshold\n  const setupFailed = result.setup !== undefined && result.setup.exitCode !== 0\n  const runFailed = result.run !== undefined && result.run.exitCode !== 0\n  const testFailed = result.test !== undefined && result.test.exitCode !== 0\n  const failureClass: FailureClass | undefined = pass\n    ? 'success'\n    : setupFailed || runFailed\n      ? 'sandbox_failure'\n      : testFailed\n        ? 'format_drift'\n        : 'unknown'\n  await emitter.endRun({\n    pass,\n    score: result.score,\n    failureClass,\n    notes: pass ? undefined : reasonForFailure(result),\n  })\n  return {\n    runId: emitter.runId,\n    scenario,\n    harness: result,\n    pass,\n    score: result.score,\n    failureClass,\n  }\n}\n\nfunction reasonForFailure(result: SandboxHarnessResult): string {\n  if (result.setup && result.setup.exitCode !== 0)\n    return `setup failed: exit ${result.setup.exitCode}`\n  if (result.run && result.run.exitCode !== 0) return `run failed: exit ${result.run.exitCode}`\n  if (result.test) {\n    if (result.test.testsTotal !== undefined) {\n      return `tests: ${result.test.testsPassed ?? 0}/${result.test.testsTotal}`\n    }\n    return `test exit ${result.test.exitCode}`\n  }\n  return 'no test command'\n}\n"],"mappings":";;;;;;;;AAoEO,IAAM,mBAAqC;AAAA,EAChD,IAAI;AAAA,EACJ,MAAM,QAAQ;AACZ,UAAM,IAAI,OAAO,MAAM,sEAAsE;AAC7F,QAAI,CAAC,EAAG,QAAO;AACf,QAAI,SAAS;AACb,QAAI,SAAS;AACb,UAAM,IAAI,SAAS,EAAE,CAAC,GAAI,EAAE;AAC5B,UAAM,SAAS,EAAE,CAAC,EAAG,YAAY;AACjC,QAAI,WAAW,SAAU,WAAU;AAAA,QAC9B,WAAU;AACf,QAAI,EAAE,CAAC,KAAK,EAAE,CAAC,GAAG;AAChB,YAAM,IAAI,SAAS,EAAE,CAAC,GAAG,EAAE;AAC3B,UAAI,EAAE,CAAC,EAAE,YAAY,MAAM,SAAU,WAAU;AAAA,UAC1C,WAAU;AAAA,IACjB;AACA,WAAO,EAAE,YAAY,SAAS,QAAQ,aAAa,OAAO;AAAA,EAC5D;AACF;AAGO,IAAM,mBAAqC;AAAA,EAChD,IAAI;AAAA,EACJ,MAAM,QAAQ;AACZ,UAAM,QAAQ,OAAO,MAAM,6BAA6B;AACxD,UAAM,SAAS,OAAO,MAAM,gBAAgB;AAC5C,QAAI,CAAC,SAAS,CAAC,OAAQ,QAAO;AAC9B,WAAO,EAAE,YAAY,SAAS,MAAM,CAAC,GAAI,EAAE,GAAG,aAAa,SAAS,OAAO,CAAC,GAAI,EAAE,EAAE;AAAA,EACtF;AACF;AAGO,IAAM,iBAAmC;AAAA,EAC9C,IAAI;AAAA,EACJ,MAAM,QAAQ;AACZ,UAAM,IAAI,OAAO,MAAM,uEAAuE;AAC9F,QAAI,CAAC,EAAG,QAAO;AACf,WAAO,EAAE,YAAY,SAAS,EAAE,CAAC,GAAI,EAAE,GAAG,aAAa,SAAS,EAAE,CAAC,GAAI,EAAE,EAAE;AAAA,EAC7E;AACF;AAGO,SAAS,kBAAkB,SAA+C;AAC/E,SAAO;AAAA,IACL,IAAI,QAAQ,IAAI,CAAC,MAAM,EAAE,EAAE,EAAE,KAAK,GAAG;AAAA,IACrC,MAAM,QAAQ,QAAQ,UAAU;AAC9B,iBAAW,KAAK,SAAS;AACvB,cAAM,MAAM,EAAE,MAAM,QAAQ,QAAQ,QAAQ;AAC5C,YAAI,IAAK,QAAO;AAAA,MAClB;AACA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AAqBO,IAAM,0BAAN,MAAuD;AAAA,EAC5D,KAAK;AAAA,EACG;AAAA,EACA;AAAA,EAER,YAAY,UAA0C,CAAC,GAAG;AACxD,SAAK,aAAa,QAAQ;AAC1B,SAAK,aAAa,QAAQ;AAAA,EAC5B;AAAA,EAEA,MAAM,KACJ,OACA,SACA,QACwB;AACxB,UAAM,EAAE,MAAM,IAAI,MAAM,OAAO,eAAoB;AACnD,UAAM,QAAQ,KAAK,IAAI;AAOvB,UAAM,eAAe,OAAO,OAAO,KAAK;AACxC,UAAM,eAAe,EAAE,GAAG,QAAQ,KAAK,GAAI,KAAK,cAAc,CAAC,GAAI,GAAI,OAAO,OAAO,CAAC,EAAG;AACzF,WAAO,MAAM,IAAI,QAAuB,CAAC,YAAY;AACnD,YAAM,QAAQ,MAAM,SAAS;AAAA,QAC3B,OAAO;AAAA,QACP,KAAK;AAAA,QACL,KAAK;AAAA,MACP,CAAC;AACD,UAAI,SAAS;AACb,UAAI,SAAS;AACb,YAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM;AAC9B,kBAAU,OAAO,CAAC;AAAA,MACpB,CAAC;AACD,YAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM;AAC9B,kBAAU,OAAO,CAAC;AAAA,MACpB,CAAC;AACD,YAAM,UAAU;AAAA,QACd,MAAM;AACJ,cAAI;AACF,kBAAM,KAAK,SAAS;AAAA,UACtB,QAAQ;AAAA,UAAC;AAAA,QACX;AAAA,QACA,OAAO,aAAa,KAAK;AAAA,MAC3B;AACA,YAAM,GAAG,SAAS,CAAC,SAAS;AAC1B,qBAAa,OAAO;AACpB,cAAM,SAAS,KAAK,IAAI,IAAI;AAC5B,cAAM,SACJ,UAAU,UAAU,OAAO,aACvB,OAAO,WAAW,MAAM,QAAQ,QAAQ,QAAQ,CAAC,IACjD;AACN,gBAAQ;AAAA,UACN;AAAA,UACA,UAAU,QAAQ;AAAA,UAClB;AAAA,UACA;AAAA,UACA;AAAA,UACA,YAAY,QAAQ;AAAA,UACpB,aAAa,QAAQ;AAAA,QACvB,CAAC;AAAA,MACH,CAAC;AACD,YAAM,GAAG,SAAS,CAAC,QAAQ;AACzB,qBAAa,OAAO;AACpB,cAAM,SAAS,KAAK,IAAI,IAAI;AAC5B,gBAAQ,EAAE,OAAO,UAAU,KAAK,QAAQ,QAAQ,SAAS,OAAO,GAAG,GAAG,OAAO,CAAC;AAAA,MAChF,CAAC;AAAA,IACH,CAAC;AAAA,EACH;AACF;AAEO,IAAM,sBAAN,MAAmD;AAAA,EACxD,KAAK;AAAA,EAEL,MAAM,KACJ,OACA,SACA,QACwB;AACxB,QAAI,CAAC,OAAO,MAAO,OAAM,IAAI,YAAY,2CAA2C;AACpF,UAAM,MAAM,IAAI,wBAAwB;AACxC,UAAM,UAAU,OAAO,QAAQ,OAAO,OAAO,CAAC,CAAC,EAC5C,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,MAAM,WAAW,CAAC,CAAC,IAAI,WAAW,CAAC,CAAC,EAAE,EACtD,KAAK,GAAG;AACX,UAAM,UAAU,mBAAmB,OAAO,IAAI,WAAW,OAAO,KAAK,CAAC,UAAU,WAAW,OAAO,CAAC;AACnG,WAAO,IAAI,KAAK,OAAO,SAAS,EAAE,GAAG,QAAQ,KAAK,OAAU,CAAC;AAAA,EAC/D;AACF;AAEA,SAAS,WAAW,GAAmB;AACrC,MAAI,yBAAyB,KAAK,CAAC,EAAG,QAAO;AAC7C,SAAO,IAAI,EAAE,QAAQ,MAAM,OAAO,CAAC;AACrC;AAcO,IAAM,iBAAN,MAAqB;AAAA,EAClB;AAAA,EACR,YAAY,SAAwB,IAAI,wBAAwB,GAAG;AACjE,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,MAAM,IAAI,QAAuB,SAAsD;AACrF,UAAM,SAAS,MAAM,QAAQ,QAAQ;AAAA,MACnC,MAAM,WAAW,KAAK,OAAO,EAAE;AAAA,MAC/B,OAAO,OAAO;AAAA,MACd,SAAS,CAAC,OAAO,cAAc,OAAO,YAAY,OAAO,WAAW,EACjE,OAAO,OAAO,EACd,KAAK,MAAM;AAAA,IAChB,CAAC;AACD,UAAM,SAA+B,EAAE,QAAQ,OAAO,aAAa,GAAG,OAAO,EAAE;AAC/E,QAAI;AACF,UAAI,OAAO,cAAc;AACvB,eAAO,QAAQ,MAAM,KAAK,OAAO,KAAK,SAAS,OAAO,cAAc,MAAM;AAC1E,eAAO,eAAe,OAAO,MAAM;AACnC,YAAI,OAAO,MAAM,aAAa,GAAG;AAC/B,gBAAM,OAAO,KAAK,sBAAsB,OAAO,MAAM,QAAQ,KAAK;AAAA,YAChE,UAAU,OAAO,MAAM;AAAA,YACvB,QAAQ,OAAO;AAAA,UACjB,CAAyB;AACzB,iBAAO;AAAA,QACT;AAAA,MACF;AACA,UAAI,OAAO,YAAY;AACrB,eAAO,MAAM,MAAM,KAAK,OAAO,KAAK,OAAO,OAAO,YAAY,MAAM;AACpE,eAAO,eAAe,OAAO,IAAI;AACjC,YAAI,OAAO,IAAI,aAAa,GAAG;AAC7B,gBAAM,OAAO,KAAK,oBAAoB,OAAO,IAAI,QAAQ,KAAK;AAAA,YAC5D,UAAU,OAAO,IAAI;AAAA,YACrB,QAAQ,OAAO;AAAA,UACjB,CAAyB;AACzB,iBAAO;AAAA,QACT;AAAA,MACF;AACA,UAAI,OAAO,aAAa;AACtB,eAAO,OAAO,MAAM,KAAK,OAAO,KAAK,QAAQ,OAAO,aAAa,MAAM;AACvE,eAAO,eAAe,OAAO,KAAK;AAClC,cAAM,SAAS,OAAO,KAAK,aAAa;AACxC,eAAO,SAAS;AAChB,YAAI,OAAO,KAAK,eAAe,UAAa,OAAO,KAAK,aAAa,GAAG;AACtE,iBAAO,SAAS,OAAO,KAAK,eAAe,KAAK,OAAO,KAAK;AAAA,QAC9D,OAAO;AACL,iBAAO,QAAQ,SAAS,IAAI;AAAA,QAC9B;AACA,cAAM,OAAO,IAAI;AAAA,UACf,UAAU,OAAO,KAAK;AAAA,UACtB,YAAY,OAAO,KAAK;AAAA,UACxB,aAAa,OAAO,KAAK;AAAA,UACzB,QAAQ,OAAO;AAAA,UACf,QAAQ,SAAS,OAAO;AAAA,QAC1B,CAAyB;AAAA,MAC3B,OAAO;AACL,eAAO,SAAS;AAChB,eAAO,QAAQ;AACf,cAAM,OAAO,IAAI,EAAE,QAAQ,OAAO,YAAY,CAAyB;AAAA,MACzE;AAAA,IACF,SAAS,KAAK;AACZ,YAAM,OAAO,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;AAC1D,YAAM;AAAA,IACR;AACA,WAAO;AAAA,EACT;AACF;;;AC7QA,eAAsB,sBACpB,UACA,OACA,UAAgC,CAAC,GACH;AAC9B,QAAM,UAAU,IAAI,aAAa,KAAK;AACtC,QAAM,QAAQ,SAAS;AAAA,IACrB,YAAY,SAAS;AAAA,IACrB,WAAW,QAAQ;AAAA,IACnB,gBAAgB,SAAS;AAAA,IACzB,MAAM,SAAS;AAAA,IACf,GAAG,QAAQ;AAAA,EACb,CAAC;AACD,QAAM,UAAU,IAAI,eAAe,QAAQ,MAAM;AACjD,QAAM,SAAS,MAAM,QAAQ,IAAI,SAAS,SAAS,OAAO;AAC1D,QAAM,YAAY,SAAS,iBAAiB;AAC5C,QAAM,OAAO,OAAO,UAAU,OAAO,SAAS;AAC9C,QAAM,cAAc,OAAO,UAAU,UAAa,OAAO,MAAM,aAAa;AAC5E,QAAM,YAAY,OAAO,QAAQ,UAAa,OAAO,IAAI,aAAa;AACtE,QAAM,aAAa,OAAO,SAAS,UAAa,OAAO,KAAK,aAAa;AACzE,QAAM,eAAyC,OAC3C,YACA,eAAe,YACb,oBACA,aACE,iBACA;AACR,QAAM,QAAQ,OAAO;AAAA,IACnB;AAAA,IACA,OAAO,OAAO;AAAA,IACd;AAAA,IACA,OAAO,OAAO,SAAY,iBAAiB,MAAM;AAAA,EACnD,CAAC;AACD,SAAO;AAAA,IACL,OAAO,QAAQ;AAAA,IACf;AAAA,IACA,SAAS;AAAA,IACT;AAAA,IACA,OAAO,OAAO;AAAA,IACd;AAAA,EACF;AACF;AAEA,SAAS,iBAAiB,QAAsC;AAC9D,MAAI,OAAO,SAAS,OAAO,MAAM,aAAa;AAC5C,WAAO,sBAAsB,OAAO,MAAM,QAAQ;AACpD,MAAI,OAAO,OAAO,OAAO,IAAI,aAAa,EAAG,QAAO,oBAAoB,OAAO,IAAI,QAAQ;AAC3F,MAAI,OAAO,MAAM;AACf,QAAI,OAAO,KAAK,eAAe,QAAW;AACxC,aAAO,UAAU,OAAO,KAAK,eAAe,CAAC,IAAI,OAAO,KAAK,UAAU;AAAA,IACzE;AACA,WAAO,aAAa,OAAO,KAAK,QAAQ;AAAA,EAC1C;AACA,SAAO;AACT;","names":[]}

package/dist/{chunk-SQQLHODJ.js → chunk-PC4UYEBM.js}

@@ -32,7 +32,8 @@ function redactHeaders(headers, prefix, redactedFields) {
 }
 function redactBody(value, pathStr, redactedFields) {
   if (value == null) return value;
-  if (Array.isArray(value)) return value.map((v, i) => redactBody(v, `${pathStr}[${i}]`, redactedFields));
+  if (Array.isArray(value))
+    return value.map((v, i) => redactBody(v, `${pathStr}[${i}]`, redactedFields));
   if (typeof value === "object") {
     const out = {};
     for (const [k, v] of Object.entries(value)) {
@@ -104,7 +105,8 @@ var FileSystemRawProviderSink = class {
   async record(event) {
     await this.ensureInit();
     const redacted = this.redactor({ ...event, redactedFields: event.redactedFields ?? [] });
-    const line = JSON.stringify(redacted) + "\n";
+    const line = `${JSON.stringify(redacted)}
+`;
     if (this.bytesWritten + line.length > this.rollAtBytes && this.bytesWritten > 0) {
       this.rollIndex += 1;
       this.bytesWritten = 0;
@@ -130,7 +132,8 @@ var FileSystemRawProviderSink = class {
         if (filter.runId !== void 0 && event.runId !== filter.runId) continue;
         if (filter.spanId !== void 0 && event.spanId !== filter.spanId) continue;
         if (filter.direction !== void 0 && event.direction !== filter.direction) continue;
-        if (filter.attemptIndex !== void 0 && event.attemptIndex !== filter.attemptIndex) continue;
+        if (filter.attemptIndex !== void 0 && event.attemptIndex !== filter.attemptIndex)
+          continue;
         out.push(event);
       }
     }
@@ -160,4 +163,4 @@ export {
   FileSystemRawProviderSink,
   providerFromBaseUrl
 };
-//# sourceMappingURL=chunk-SQQLHODJ.js.map
+//# sourceMappingURL=chunk-PC4UYEBM.js.map

package/dist/chunk-PC4UYEBM.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/trace/raw-provider-sink.ts"],"sourcesContent":["/**\n * RawProviderSink — first-class persistence for the actual HTTP-level\n * request/response bodies of every LLM provider call.\n *\n * Why this is a separate sink from the structured `LlmSpan`:\n *\n *   - `LlmSpan` records the *intent* — model name, messages, output text,\n *     usage. It's what dashboards read; it's NOT enough for forensics.\n *   - When a downstream consumer reports \"the verifier used the wrong route\"\n *     or \"tokens look right but reasoning was missing,\" the only way to\n *     answer is the raw HTTP body. Span fields can lie (a proxy can echo\n *     a different `model` value than what actually answered); the raw\n *     response is ground truth.\n *\n * Default behaviour: opt-in. Pass `rawSink` to `LlmClientOptions` (or the\n * matrix runner / BuilderSession sets it up automatically) and every\n * request, response, and error is recorded — including retries, with the\n * attempt index attached so a flaky call's full event chain is recoverable.\n *\n * Redaction is enforced at sink time. The default redactor strips\n * `Authorization`, `X-Api-Key`, `X-Auth-Token`, `Cookie` headers and any\n * payload field whose key matches `apiKey | api_key | bearer | password |\n * secret | token` (case-insensitive). Override via the sink constructor or\n * the per-call `redactor`. The `redactedFields` array on the persisted\n * event lets a reviewer see what was stripped without exposing the values.\n */\n\nimport { promises as fs } from 'node:fs'\nimport * as path from 'node:path'\n\nexport type RawProviderDirection = 'request' | 'response' | 'error'\n\nexport interface RawProviderEvent {\n  /** Stable id. Generated by the sink if omitted. */\n  eventId: string\n  /** Trace context populated by `LlmClient` when the call is wrapped in a span. */\n  runId?: string\n  spanId?: string\n  /**\n   * Logical provider name. Free-form so callers can use whatever id matches\n   * their topology (`'openai'`, `'anthropic'`, `'tangle-router'`, …). When\n   * omitted, derived from `baseUrl` in `LlmClientOptions`.\n   */\n  provider: string\n  model: string\n  /** Endpoint path, e.g. `'/v1/chat/completions'`. */\n  endpoint: string\n  /** Base URL used for the call (already-normalised — no trailing slash). */\n  baseUrl: string\n  /** 0-indexed retry attempt. The first attempt is 0; a retried call gets 1, 2, … */\n  attemptIndex: number\n  direction: RawProviderDirection\n  /** Unix ms. */\n  timestamp: number\n  /** Wall-clock duration of the call leg. Set on `response` and `error` events; null on `request`. */\n  durationMs?: number\n  statusCode?: number\n  requestHeaders?: Record<string, string>\n  requestBody?: unknown\n  responseHeaders?: Record<string, string>\n  responseBody?: unknown\n  /** Set on `direction: 'error'` events. */\n  errorMessage?: string\n  /** Field paths the redactor stripped from this event ('header:Authorization', 'body.apiKey', …). */\n  redactedFields: string[]\n}\n\nexport interface RawProviderSinkFilter {\n  runId?: string\n  spanId?: string\n  direction?: RawProviderDirection\n  attemptIndex?: number\n}\n\nexport interface RawProviderSink {\n  record(event: RawProviderEvent): Promise<void>\n  /** Optional listing — implementations that durably persist (file, db) should support this. */\n  list?(filter?: RawProviderSinkFilter): Promise<RawProviderEvent[]>\n  /** Optional teardown for backed implementations. */\n  close?(): Promise<void>\n}\n\nexport type ProviderRedactor = (event: RawProviderEvent) => RawProviderEvent\n\nconst REDACTED_HEADER_NAMES = new Set([\n  'authorization',\n  'x-api-key',\n  'x-auth-token',\n  'cookie',\n  'set-cookie',\n  'proxy-authorization',\n])\n\nconst REDACTED_BODY_KEY =\n  /^(api[_-]?key|bearer|password|secret|token|access[_-]?token|refresh[_-]?token)$/i\n\n/**\n * Default redactor — strips well-known auth headers and any body field whose\n * key matches the credential pattern. Records every redacted path on\n * `event.redactedFields` so a downstream reviewer can see what was removed.\n */\nexport function defaultProviderRedactor(event: RawProviderEvent): RawProviderEvent {\n  const redactedFields: string[] = [...(event.redactedFields ?? [])]\n  const requestHeaders = redactHeaders(event.requestHeaders, 'request', redactedFields)\n  const responseHeaders = redactHeaders(event.responseHeaders, 'response', redactedFields)\n  const requestBody = redactBody(event.requestBody, 'requestBody', redactedFields)\n  const responseBody = redactBody(event.responseBody, 'responseBody', redactedFields)\n  return { ...event, requestHeaders, responseHeaders, requestBody, responseBody, redactedFields }\n}\n\nfunction redactHeaders(\n  headers: Record<string, string> | undefined,\n  prefix: 'request' | 'response',\n  redactedFields: string[],\n): Record<string, string> | undefined {\n  if (!headers) return headers\n  const out: Record<string, string> = {}\n  for (const [k, v] of Object.entries(headers)) {\n    if (REDACTED_HEADER_NAMES.has(k.toLowerCase())) {\n      redactedFields.push(`${prefix}Headers.${k}`)\n      continue\n    }\n    out[k] = v\n  }\n  return out\n}\n\nfunction redactBody(value: unknown, pathStr: string, redactedFields: string[]): unknown {\n  if (value == null) return value\n  if (Array.isArray(value))\n    return value.map((v, i) => redactBody(v, `${pathStr}[${i}]`, redactedFields))\n  if (typeof value === 'object') {\n    const out: Record<string, unknown> = {}\n    for (const [k, v] of Object.entries(value as Record<string, unknown>)) {\n      if (REDACTED_BODY_KEY.test(k)) {\n        redactedFields.push(`${pathStr}.${k}`)\n        continue\n      }\n      out[k] = redactBody(v, `${pathStr}.${k}`, redactedFields)\n    }\n    return out\n  }\n  return value\n}\n\n// ── In-memory ────────────────────────────────────────────────────────────\n\nexport interface InMemoryRawProviderSinkOptions {\n  redactor?: ProviderRedactor\n}\n\nexport class InMemoryRawProviderSink implements RawProviderSink {\n  private events: RawProviderEvent[] = []\n  private redactor: ProviderRedactor\n\n  constructor(opts: InMemoryRawProviderSinkOptions = {}) {\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    this.events.push(this.redactor({ ...event, redactedFields: event.redactedFields ?? [] }))\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    return this.events.filter(\n      (e) =>\n        (filter.runId === undefined || e.runId === filter.runId) &&\n        (filter.spanId === undefined || e.spanId === filter.spanId) &&\n        (filter.direction === undefined || e.direction === filter.direction) &&\n        (filter.attemptIndex === undefined || e.attemptIndex === filter.attemptIndex),\n    )\n  }\n\n  size(): number {\n    return this.events.length\n  }\n}\n\nexport class NoopRawProviderSink implements RawProviderSink {\n  async record(): Promise<void> {\n    /* no-op */\n  }\n  /**\n   * Returns an empty array. Implemented so `assertRunCaptured` does not\n   * trip the `no_raw_sink` issue when a caller explicitly opts out of\n   * capture by passing this sink — opt-out is a deliberate choice, not a\n   * misconfiguration.\n   */\n  async list(): Promise<RawProviderEvent[]> {\n    return []\n  }\n}\n\n// ── Filesystem (NDJSON) ──────────────────────────────────────────────────\n\nexport interface FileSystemRawProviderSinkOptions {\n  /** Directory the NDJSON file is written into. Created if missing. */\n  dir: string\n  /** File name; default `'raw-provider-events.ndjson'`. */\n  fileName?: string\n  /** Bytes after which the writer rolls over to a new file (default 32 MiB). */\n  rollAtBytes?: number\n  redactor?: ProviderRedactor\n}\n\nexport class FileSystemRawProviderSink implements RawProviderSink {\n  private dir: string\n  private fileName: string\n  private rollAtBytes: number\n  private redactor: ProviderRedactor\n  private bytesWritten = 0\n  private rollIndex = 0\n  private initPromise: Promise<void> | null = null\n\n  constructor(opts: FileSystemRawProviderSinkOptions) {\n    this.dir = opts.dir\n    this.fileName = opts.fileName ?? 'raw-provider-events.ndjson'\n    this.rollAtBytes = opts.rollAtBytes ?? 32 * 1024 * 1024\n    this.redactor = opts.redactor ?? defaultProviderRedactor\n  }\n\n  private async ensureInit(): Promise<void> {\n    if (!this.initPromise) {\n      this.initPromise = fs.mkdir(this.dir, { recursive: true }).then(() => undefined)\n    }\n    await this.initPromise\n  }\n\n  private currentPath(): string {\n    if (this.rollIndex === 0) return path.join(this.dir, this.fileName)\n    return path.join(this.dir, `${this.fileName}.${this.rollIndex}`)\n  }\n\n  async record(event: RawProviderEvent): Promise<void> {\n    await this.ensureInit()\n    const redacted = this.redactor({ ...event, redactedFields: event.redactedFields ?? [] })\n    const line = `${JSON.stringify(redacted)}\\n`\n    if (this.bytesWritten + line.length > this.rollAtBytes && this.bytesWritten > 0) {\n      this.rollIndex += 1\n      this.bytesWritten = 0\n    }\n    await fs.appendFile(this.currentPath(), line, 'utf8')\n    this.bytesWritten += line.length\n  }\n\n  async list(filter: RawProviderSinkFilter = {}): Promise<RawProviderEvent[]> {\n    await this.ensureInit()\n    const out: RawProviderEvent[] = []\n    for (let i = 0; i <= this.rollIndex; i++) {\n      const file =\n        i === 0 ? path.join(this.dir, this.fileName) : path.join(this.dir, `${this.fileName}.${i}`)\n      let body: string\n      try {\n        body = await fs.readFile(file, 'utf8')\n      } catch (err) {\n        if ((err as NodeJS.ErrnoException).code === 'ENOENT') continue\n        throw err\n      }\n      for (const line of body.split('\\n')) {\n        if (!line) continue\n        const event = JSON.parse(line) as RawProviderEvent\n        if (filter.runId !== undefined && event.runId !== filter.runId) continue\n        if (filter.spanId !== undefined && event.spanId !== filter.spanId) continue\n        if (filter.direction !== undefined && event.direction !== filter.direction) continue\n        if (filter.attemptIndex !== undefined && event.attemptIndex !== filter.attemptIndex)\n          continue\n        out.push(event)\n      }\n    }\n    return out\n  }\n}\n\n// ── Helpers ──────────────────────────────────────────────────────────────\n\n/**\n * Best-effort provider id from a base URL. Falls back to the URL host when\n * none of the well-known patterns match.\n */\nexport function providerFromBaseUrl(baseUrl: string): string {\n  const lower = baseUrl.toLowerCase()\n  if (lower.includes('api.openai.com')) return 'openai'\n  if (lower.includes('api.anthropic.com')) return 'anthropic'\n  if (lower.includes('generativelanguage.googleapis.com')) return 'google'\n  if (lower.includes('api.together.ai') || lower.includes('api.together.xyz')) return 'together'\n  if (lower.includes('api.deepseek.com')) return 'deepseek'\n  if (lower.includes('router.tangle.tools')) return 'tangle-router'\n  if (lower.includes('api.litellm') || lower.includes('litellm')) return 'litellm'\n  try {\n    return new URL(baseUrl).host\n  } catch {\n    return baseUrl\n  }\n}\n"],"mappings":";AA2BA,SAAS,YAAY,UAAU;AAC/B,YAAY,UAAU;AAwDtB,IAAM,wBAAwB,oBAAI,IAAI;AAAA,EACpC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,CAAC;AAED,IAAM,oBACJ;AAOK,SAAS,wBAAwB,OAA2C;AACjF,QAAM,iBAA2B,CAAC,GAAI,MAAM,kBAAkB,CAAC,CAAE;AACjE,QAAM,iBAAiB,cAAc,MAAM,gBAAgB,WAAW,cAAc;AACpF,QAAM,kBAAkB,cAAc,MAAM,iBAAiB,YAAY,cAAc;AACvF,QAAM,cAAc,WAAW,MAAM,aAAa,eAAe,cAAc;AAC/E,QAAM,eAAe,WAAW,MAAM,cAAc,gBAAgB,cAAc;AAClF,SAAO,EAAE,GAAG,OAAO,gBAAgB,iBAAiB,aAAa,cAAc,eAAe;AAChG;AAEA,SAAS,cACP,SACA,QACA,gBACoC;AACpC,MAAI,CAAC,QAAS,QAAO;AACrB,QAAM,MAA8B,CAAC;AACrC,aAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,OAAO,GAAG;AAC5C,QAAI,sBAAsB,IAAI,EAAE,YAAY,CAAC,GAAG;AAC9C,qBAAe,KAAK,GAAG,MAAM,WAAW,CAAC,EAAE;AAC3C;AAAA,IACF;AACA,QAAI,CAAC,IAAI;AAAA,EACX;AACA,SAAO;AACT;AAEA,SAAS,WAAW,OAAgB,SAAiB,gBAAmC;AACtF,MAAI,SAAS,KAAM,QAAO;AAC1B,MAAI,MAAM,QAAQ,KAAK;AACrB,WAAO,MAAM,IAAI,CAAC,GAAG,MAAM,WAAW,GAAG,GAAG,OAAO,IAAI,CAAC,KAAK,cAAc,CAAC;AAC9E,MAAI,OAAO,UAAU,UAAU;AAC7B,UAAM,MAA+B,CAAC;AACtC,eAAW,CAAC,GAAG,CAAC,KAAK,OAAO,QAAQ,KAAgC,GAAG;AACrE,UAAI,kBAAkB,KAAK,CAAC,GAAG;AAC7B,uBAAe,KAAK,GAAG,OAAO,IAAI,CAAC,EAAE;AACrC;AAAA,MACF;AACA,UAAI,CAAC,IAAI,WAAW,GAAG,GAAG,OAAO,IAAI,CAAC,IAAI,cAAc;AAAA,IAC1D;AACA,WAAO;AAAA,EACT;AACA,SAAO;AACT;AAQO,IAAM,0BAAN,MAAyD;AAAA,EACtD,SAA6B,CAAC;AAAA,EAC9B;AAAA,EAER,YAAY,OAAuC,CAAC,GAAG;AACrD,SAAK,WAAW,KAAK,YAAY;AAAA,EACnC;AAAA,EAEA,MAAM,OAAO,OAAwC;AACnD,SAAK,OAAO,KAAK,KAAK,SAAS,EAAE,GAAG,OAAO,gBAAgB,MAAM,kBAAkB,CAAC,EAAE,CAAC,CAAC;AAAA,EAC1F;AAAA,EAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;AAC1E,WAAO,KAAK,OAAO;AAAA,MACjB,CAAC,OACE,OAAO,UAAU,UAAa,EAAE,UAAU,OAAO,WACjD,OAAO,WAAW,UAAa,EAAE,WAAW,OAAO,YACnD,OAAO,cAAc,UAAa,EAAE,cAAc,OAAO,eACzD,OAAO,iBAAiB,UAAa,EAAE,iBAAiB,OAAO;AAAA,IACpE;AAAA,EACF;AAAA,EAEA,OAAe;AACb,WAAO,KAAK,OAAO;AAAA,EACrB;AACF;AAEO,IAAM,sBAAN,MAAqD;AAAA,EAC1D,MAAM,SAAwB;AAAA,EAE9B;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,MAAM,OAAoC;AACxC,WAAO,CAAC;AAAA,EACV;AACF;AAcO,IAAM,4BAAN,MAA2D;AAAA,EACxD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA,eAAe;AAAA,EACf,YAAY;AAAA,EACZ,cAAoC;AAAA,EAE5C,YAAY,MAAwC;AAClD,SAAK,MAAM,KAAK;AAChB,SAAK,WAAW,KAAK,YAAY;AACjC,SAAK,cAAc,KAAK,eAAe,KAAK,OAAO;AACnD,SAAK,WAAW,KAAK,YAAY;AAAA,EACnC;AAAA,EAEA,MAAc,aAA4B;AACxC,QAAI,CAAC,KAAK,aAAa;AACrB,WAAK,cAAc,GAAG,MAAM,KAAK,KAAK,EAAE,WAAW,KAAK,CAAC,EAAE,KAAK,MAAM,MAAS;AAAA,IACjF;AACA,UAAM,KAAK;AAAA,EACb;AAAA,EAEQ,cAAsB;AAC5B,QAAI,KAAK,cAAc,EAAG,QAAY,UAAK,KAAK,KAAK,KAAK,QAAQ;AAClE,WAAY,UAAK,KAAK,KAAK,GAAG,KAAK,QAAQ,IAAI,KAAK,SAAS,EAAE;AAAA,EACjE;AAAA,EAEA,MAAM,OAAO,OAAwC;AACnD,UAAM,KAAK,WAAW;AACtB,UAAM,WAAW,KAAK,SAAS,EAAE,GAAG,OAAO,gBAAgB,MAAM,kBAAkB,CAAC,EAAE,CAAC;AACvF,UAAM,OAAO,GAAG,KAAK,UAAU,QAAQ,CAAC;AAAA;AACxC,QAAI,KAAK,eAAe,KAAK,SAAS,KAAK,eAAe,KAAK,eAAe,GAAG;AAC/E,WAAK,aAAa;AAClB,WAAK,eAAe;AAAA,IACtB;AACA,UAAM,GAAG,WAAW,KAAK,YAAY,GAAG,MAAM,MAAM;AACpD,SAAK,gBAAgB,KAAK;AAAA,EAC5B;AAAA,EAEA,MAAM,KAAK,SAAgC,CAAC,GAAgC;AAC1E,UAAM,KAAK,WAAW;AACtB,UAAM,MAA0B,CAAC;AACjC,aAAS,IAAI,GAAG,KAAK,KAAK,WAAW,KAAK;AACxC,YAAM,OACJ,MAAM,IAAS,UAAK,KAAK,KAAK,KAAK,QAAQ,IAAS,UAAK,KAAK,KAAK,GAAG,KAAK,QAAQ,IAAI,CAAC,EAAE;AAC5F,UAAI;AACJ,UAAI;AACF,eAAO,MAAM,GAAG,SAAS,MAAM,MAAM;AAAA,MACvC,SAAS,KAAK;AACZ,YAAK,IAA8B,SAAS,SAAU;AACtD,cAAM;AAAA,MACR;AACA,iBAAW,QAAQ,KAAK,MAAM,IAAI,GAAG;AACnC,YAAI,CAAC,KAAM;AACX,cAAM,QAAQ,KAAK,MAAM,IAAI;AAC7B,YAAI,OAAO,UAAU,UAAa,MAAM,UAAU,OAAO,MAAO;AAChE,YAAI,OAAO,WAAW,UAAa,MAAM,WAAW,OAAO,OAAQ;AACnE,YAAI,OAAO,cAAc,UAAa,MAAM,cAAc,OAAO,UAAW;AAC5E,YAAI,OAAO,iBAAiB,UAAa,MAAM,iBAAiB,OAAO;AACrE;AACF,YAAI,KAAK,KAAK;AAAA,MAChB;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;AAQO,SAAS,oBAAoB,SAAyB;AAC3D,QAAM,QAAQ,QAAQ,YAAY;AAClC,MAAI,MAAM,SAAS,gBAAgB,EAAG,QAAO;AAC7C,MAAI,MAAM,SAAS,mBAAmB,EAAG,QAAO;AAChD,MAAI,MAAM,SAAS,mCAAmC,EAAG,QAAO;AAChE,MAAI,MAAM,SAAS,iBAAiB,KAAK,MAAM,SAAS,kBAAkB,EAAG,QAAO;AACpF,MAAI,MAAM,SAAS,kBAAkB,EAAG,QAAO;AAC/C,MAAI,MAAM,SAAS,qBAAqB,EAAG,QAAO;AAClD,MAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,SAAS,EAAG,QAAO;AACvE,MAAI;AACF,WAAO,IAAI,IAAI,OAAO,EAAE;AAAA,EAC1B,QAAQ;AACN,WAAO;AAAA,EACT;AACF;","names":[]}