@tangle-network/agent-eval 1.0.3 → 1.0.4

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "1.0.3",
+    "version": "1.0.4",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.1.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/{proposal-findings-duV5flPv.js → proposal-findings-BclX0PFh.js}

@@ -1,4 +1,4 @@
-import { S as ProposalFindingSchema } from "./external-optimizer-subprocess-wzMaA0oN.js";
+import { S as ProposalFindingSchema } from "./external-optimizer-subprocess-CeHuAaMJ.js";
 // src/analyst/proposal-findings.ts
 /** True when a finding names a source candidate generation may learn from. */
 function isProposalFinding(finding) {
@@ -22,4 +22,4 @@ function findingLabel(finding, index) {
 }
 export { assertProposalFindings as t };
 
-//# sourceMappingURL=proposal-findings-duV5flPv.js.map
+//# sourceMappingURL=proposal-findings-BclX0PFh.js.map

package/dist/{proposal-findings-duV5flPv.js.map → proposal-findings-BclX0PFh.js.map}

@@ -1 +1 @@
-{"version":3,"file":"proposal-findings-duV5flPv.js","names":[],"sources":["../src/analyst/proposal-findings.ts"],"sourcesContent":["import { ProposalFindingSchema } from './finding-scope'\nimport type { ProposalFinding } from './types'\n\n/** True when a finding names a source candidate generation may learn from. */\nfunction isProposalFinding(finding: unknown): finding is ProposalFinding {\n  return ProposalFindingSchema.safeParse(finding).success\n}\n\n/**\n * Reject findings whose source has not been explicitly admitted for candidate\n * generation. Search feedback and observed production behavior are allowed;\n * final evaluation data has no allowed origin.\n */\nexport function assertProposalFindings(\n  findings: unknown,\n  context = 'proposal findings',\n): ReadonlyArray<ProposalFinding> {\n  if (!Array.isArray(findings)) {\n    throw new TypeError(`${context}: expected an array`)\n  }\n  const rejected = findings.flatMap((finding, index) =>\n    isProposalFinding(finding) ? [] : [findingLabel(finding, index)],\n  )\n  if (rejected.length > 0) {\n    throw new Error(\n      `${context}: every finding must match AnalystFinding and declare ` +\n        `proposal_origin as search or production. ` +\n        `Rejected findings: [${rejected.join(', ')}].`,\n    )\n  }\n  return findings as ReadonlyArray<ProposalFinding>\n}\n\nfunction findingLabel(finding: unknown, index: number): string {\n  if (typeof finding !== 'object' || finding === null) return `index ${index}`\n  const id = (finding as { finding_id?: unknown }).finding_id\n  return typeof id === 'string' && id.length > 0 ? id : `index ${index}`\n}\n"],"mappings":";;;AAIA,SAAS,kBAAkB,SAA8C;CACvE,OAAO,sBAAsB,UAAU,OAAO,CAAC,CAAC;AAClD;;;;;;AAOA,SAAgB,uBACd,UACA,UAAU,qBACsB;CAChC,IAAI,CAAC,MAAM,QAAQ,QAAQ,GACzB,MAAM,IAAI,UAAU,GAAG,QAAQ,oBAAoB;CAErD,MAAM,WAAW,SAAS,SAAS,SAAS,UAC1C,kBAAkB,OAAO,IAAI,CAAC,IAAI,CAAC,aAAa,SAAS,KAAK,CAAC,CACjE;CACA,IAAI,SAAS,SAAS,GACpB,MAAM,IAAI,MACR,GAAG,QAAQ,qHAEc,SAAS,KAAK,IAAI,EAAE,GAC/C;CAEF,OAAO;AACT;AAEA,SAAS,aAAa,SAAkB,OAAuB;CAC7D,IAAI,OAAO,YAAY,YAAY,YAAY,MAAM,OAAO,SAAS;CACrE,MAAM,KAAM,QAAqC;CACjD,OAAO,OAAO,OAAO,YAAY,GAAG,SAAS,IAAI,KAAK,SAAS;AACjE"}
+{"version":3,"file":"proposal-findings-BclX0PFh.js","names":[],"sources":["../src/analyst/proposal-findings.ts"],"sourcesContent":["import { ProposalFindingSchema } from './finding-scope'\nimport type { ProposalFinding } from './types'\n\n/** True when a finding names a source candidate generation may learn from. */\nfunction isProposalFinding(finding: unknown): finding is ProposalFinding {\n  return ProposalFindingSchema.safeParse(finding).success\n}\n\n/**\n * Reject findings whose source has not been explicitly admitted for candidate\n * generation. Search feedback and observed production behavior are allowed;\n * final evaluation data has no allowed origin.\n */\nexport function assertProposalFindings(\n  findings: unknown,\n  context = 'proposal findings',\n): ReadonlyArray<ProposalFinding> {\n  if (!Array.isArray(findings)) {\n    throw new TypeError(`${context}: expected an array`)\n  }\n  const rejected = findings.flatMap((finding, index) =>\n    isProposalFinding(finding) ? [] : [findingLabel(finding, index)],\n  )\n  if (rejected.length > 0) {\n    throw new Error(\n      `${context}: every finding must match AnalystFinding and declare ` +\n        `proposal_origin as search or production. ` +\n        `Rejected findings: [${rejected.join(', ')}].`,\n    )\n  }\n  return findings as ReadonlyArray<ProposalFinding>\n}\n\nfunction findingLabel(finding: unknown, index: number): string {\n  if (typeof finding !== 'object' || finding === null) return `index ${index}`\n  const id = (finding as { finding_id?: unknown }).finding_id\n  return typeof id === 'string' && id.length > 0 ? id : `index ${index}`\n}\n"],"mappings":";;;AAIA,SAAS,kBAAkB,SAA8C;CACvE,OAAO,sBAAsB,UAAU,OAAO,CAAC,CAAC;AAClD;;;;;;AAOA,SAAgB,uBACd,UACA,UAAU,qBACsB;CAChC,IAAI,CAAC,MAAM,QAAQ,QAAQ,GACzB,MAAM,IAAI,UAAU,GAAG,QAAQ,oBAAoB;CAErD,MAAM,WAAW,SAAS,SAAS,SAAS,UAC1C,kBAAkB,OAAO,IAAI,CAAC,IAAI,CAAC,aAAa,SAAS,KAAK,CAAC,CACjE;CACA,IAAI,SAAS,SAAS,GACpB,MAAM,IAAI,MACR,GAAG,QAAQ,qHAEc,SAAS,KAAK,IAAI,EAAE,GAC/C;CAEF,OAAO;AACT;AAEA,SAAS,aAAa,SAAkB,OAAuB;CAC7D,IAAI,OAAO,YAAY,YAAY,YAAY,MAAM,OAAO,SAAS;CACrE,MAAM,KAAM,QAAqC;CACjD,OAAO,OAAO,OAAO,YAAY,GAAG,SAAS,IAAI,KAAK,SAAS;AACjE"}

package/dist/{sandbox-harness-BuacGaVn.js → sandbox-harness-BDr1GRm0.js}

@@ -8,16 +8,15 @@ var SubprocessSandboxDriver = class {
 		this.options = options;
 	}
 	async exec(phase, command, config) {
-		const effectiveEnv = {
-			...process.env,
-			...this.options.env ?? {},
-			...config.env ?? {}
-		};
 		const result = await runBoundedProcess({
 			command,
 			shell: true,
 			cwd: config.cwd ?? this.options.cwd,
-			env: effectiveEnv,
+			env: {
+				...process.env,
+				...this.options.env,
+				...config.env
+			},
 			envMode: "replace",
 			timeoutMs: config.timeoutMs ?? 6e5,
 			maxOutputBytes: config.maxOutputBytes ?? 16777216
@@ -80,12 +79,10 @@ var SandboxHarness = class {
 			result.passed = !test || !test.killedByTimeout && test.exitCode === 0;
 			result.score = test?.testsTotal !== void 0 && test.testsTotal > 0 ? (test.testsPassed ?? 0) / test.testsTotal : result.passed ? 1 : 0;
 			await handle.end({
-				...test ? {
-					exitCode: test.exitCode,
-					testsTotal: test.testsTotal,
-					testsPassed: test.testsPassed,
-					status: result.passed ? "ok" : "error"
-				} : {},
+				exitCode: test?.exitCode,
+				testsTotal: test?.testsTotal,
+				testsPassed: test?.testsPassed,
+				status: result.passed ? "ok" : "error",
 				wallMs: result.totalWallMs
 			});
 		} catch (err) {
@@ -104,15 +101,12 @@ async function runSandboxScenario(spec, options = {}) {
 		assertions: [],
 		timings: []
 	};
+	let measurementFailures = 0;
 	const retirements = [];
 	const scope = {
 		signal,
-		check: (...assertions) => {
-			report.assertions.push(...assertions);
-		},
-		defer: (retire) => {
-			retirements.push(retire);
-		},
+		check: (...assertions) => report.assertions.push(...assertions),
+		defer: (retire) => retirements.push(retire),
 		async measure(name, run) {
 			const start = performance.now();
 			let measured = {
@@ -135,9 +129,24 @@ async function runSandboxScenario(spec, options = {}) {
 					error: error instanceof Error ? error.message : String(error)
 				};
 			}
+			if (!measured.success) measurementFailures++;
 			report.timings.push(measured);
 			return measured;
 		},
+		async resource(input) {
+			let resource;
+			const timing = await scope.measure(input.name, async () => {
+				const owned = await input.create(signal);
+				resource = owned;
+				scope.defer(() => input.retire(owned));
+				input.onCreate?.(owned);
+				await input.ready?.(owned, signal);
+			});
+			return {
+				resource,
+				timing
+			};
+		},
 		async poll(read, ready, budget) {
 			if (!(budget.timeoutMs > 0 && budget.intervalMs > 0)) throw new Error("Polling timeout and interval must be positive");
 			const deadline = performance.now() + budget.timeoutMs;
@@ -152,10 +161,12 @@ async function runSandboxScenario(spec, options = {}) {
 	try {
 		signal.throwIfAborted();
 		const state = await spec.setup(scope);
-		for (const step of spec.steps) {
+		for (const step of spec.steps ?? []) {
 			signal.throwIfAborted();
 			if (step.when && !step.when(state)) continue;
+			const failuresBefore = measurementFailures;
 			const timing = await scope.measure(step.name, () => step.run(state, scope));
+			if (step.advisory) measurementFailures = failuresBefore;
 			scope.check(...step.expect?.(state, timing) ?? []);
 		}
 		signal.throwIfAborted();
@@ -173,9 +184,9 @@ async function runSandboxScenario(spec, options = {}) {
 			});
 		}
 	}
-	report.pass = !report.error && report.timings.every((t) => t.success) && report.assertions.every((a) => a.soft || a.status === "pass");
+	report.pass = !report.error && measurementFailures === 0 && report.assertions.every((a) => a.soft || a.status === "pass");
 	return report;
 }
 export { runSandboxScenario as n, SandboxHarness as t };
 
-//# sourceMappingURL=sandbox-harness-BuacGaVn.js.map
+//# sourceMappingURL=sandbox-harness-BDr1GRm0.js.map

package/dist/sandbox-harness-BDr1GRm0.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox-harness-BDr1GRm0.js","names":["delay"],"sources":["../src/sandbox-harness.ts"],"sourcesContent":["import { setTimeout as delay } from 'node:timers/promises'\nimport { runBoundedProcess } from './bounded-process'\n\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  cwd?: string\n  /** Max wall-clock per phase in ms. Default 10 minutes. */\n  timeoutMs?: number\n  /** Captured stdout + stderr cap, default 16 MiB; excess is dropped. */\n  maxOutputBytes?: number\n  image?: string\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  /** A timed-out phase always fails, even with exitCode 0. Absent/false means natural exit. */\n  killedByTimeout?: boolean\n  /** Captured output exceeded maxOutputBytes. */\n  outputTruncated?: boolean\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\ninterface SubprocessSandboxDriverOptions {\n  cwd?: string\n  env?: Record<string, string>\n}\n\nexport class SubprocessSandboxDriver implements SandboxDriver {\n  id = 'subprocess'\n  constructor(private options: SubprocessSandboxDriverOptions = {}) {}\n\n  async exec(\n    phase: SandboxResult['phase'],\n    command: string,\n    config: HarnessConfig,\n  ): Promise<SandboxResult> {\n    const result = await runBoundedProcess({\n      command,\n      shell: true,\n      cwd: config.cwd ?? this.options.cwd,\n      env: { ...process.env, ...this.options.env, ...config.env },\n      envMode: 'replace',\n      timeoutMs: config.timeoutMs ?? 10 * 60_000,\n      maxOutputBytes: config.maxOutputBytes ?? 16 * 1024 * 1024,\n    })\n    // A timed-out phase never reaches the parser: the output of a killed\n    // process is a fragment, and a fragment that happens to parse would\n    // report a pass for a run that never finished.\n    const parsed =\n      !result.killedByTimeout && phase === 'test' && config.testParser\n        ? config.testParser.parse(result.stdout, result.stderr, result.exitCode)\n        : undefined\n    return {\n      phase,\n      exitCode: result.exitCode,\n      stdout: result.stdout,\n      stderr: result.runnerError ? result.stderr + result.runnerError : result.stderr,\n      wallMs: result.wallMs,\n      testsTotal: parsed?.testsTotal,\n      testsPassed: parsed?.testsPassed,\n      killedByTimeout: result.killedByTimeout || undefined,\n      outputTruncated: result.outputTruncated || undefined,\n    }\n  }\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  constructor(private driver: SandboxDriver = new SubprocessSandboxDriver()) {}\n\n  async run(config: HarnessConfig, emitter: TraceEmitter): Promise<SandboxHarnessResult> {\n    const handle = await emitter.sandbox({\n      name: `sandbox(${this.driver.id})`,\n      image: config.image,\n      command: [config.setupCommand, config.runCommand, config.testCommand]\n        .filter(Boolean)\n        .join(' && '),\n    })\n    const result: SandboxHarnessResult = { passed: false, totalWallMs: 0, score: 0 }\n    try {\n      for (const phase of ['setup', 'run', 'test'] as const) {\n        const command = config[`${phase}Command`]\n        if (!command) continue\n        const measured = await this.driver.exec(phase, command, config)\n        result[phase] = measured\n        result.totalWallMs += measured.wallMs\n        const passed = !measured.killedByTimeout && measured.exitCode === 0\n        if (phase !== 'test' && !passed) {\n          await handle.fail(\n            measured.killedByTimeout\n              ? `${phase} timed out after ${measured.wallMs}ms`\n              : `${phase} failed (exit ${measured.exitCode})`,\n            {\n              exitCode: measured.exitCode,\n              wallMs: result.totalWallMs,\n            } as Partial<SandboxSpan>,\n          )\n          return result\n        }\n      }\n      const test = result.test\n      result.passed = !test || (!test.killedByTimeout && test.exitCode === 0)\n      result.score =\n        test?.testsTotal !== undefined && test.testsTotal > 0\n          ? (test.testsPassed ?? 0) / test.testsTotal\n          : result.passed\n            ? 1\n            : 0\n      await handle.end({\n        exitCode: test?.exitCode,\n        testsTotal: test?.testsTotal,\n        testsPassed: test?.testsPassed,\n        status: result.passed ? 'ok' : 'error',\n        wallMs: result.totalWallMs,\n      } as Partial<SandboxSpan>)\n    } catch (err) {\n      await handle.fail(err instanceof Error ? err : String(err))\n      throw err\n    }\n    return result\n  }\n}\n\n/** Portable integration report; products may retain extra timing and assertion fields. */\nexport interface HarnessAssertion {\n  name: string\n  status: 'pass' | 'fail' | 'skip'\n  expected?: string\n  actual?: string\n  message?: string\n  soft?: boolean\n}\nexport interface HarnessTiming {\n  name: string\n  durationMs: number\n  success: boolean\n  error?: string\n}\nexport interface HarnessReport {\n  pass: boolean\n  assertions: HarnessAssertion[]\n  timings: HarnessTiming[]\n  error?: string\n  metadata?: Record<string, unknown>\n}\nexport interface HarnessResource<T> {\n  name: string\n  create(signal: AbortSignal): Promise<T>\n  retire(resource: T): Promise<unknown>\n  ready?: (resource: T, signal: AbortSignal) => Promise<unknown>\n  onCreate?: (resource: T) => void\n}\nexport interface HarnessScope {\n  signal: AbortSignal\n  resource<T>(input: HarnessResource<T>): Promise<{ resource?: T; timing: HarnessTiming }>\n  check(...assertions: HarnessAssertion[]): void\n  measure(name: string, run: () => Promise<unknown>): Promise<HarnessTiming>\n  /** Register retirement immediately after allocation; cleanup ignores workload cancellation. */\n  defer(retire: () => Promise<unknown>): void\n  poll<T>(\n    read: () => Promise<T>,\n    ready: (value: T) => boolean,\n    budget: { timeoutMs: number; intervalMs: number },\n  ): Promise<T | undefined>\n}\nexport interface HarnessScenario<T> {\n  setup(scope: HarnessScope): Promise<T>\n  steps?: ReadonlyArray<{\n    name: string\n    /** Diagnostic measurements retain failures without gating the scenario. */\n    advisory?: boolean\n    when?: (state: T) => boolean\n    run(state: T, scope: HarnessScope): Promise<unknown>\n    expect?: (state: T, timing: HarnessTiming) => HarnessAssertion[]\n  }>\n  expectations?: (state: T) => HarnessAssertion[]\n  metadata?: (state: T) => Record<string, unknown>\n  /** Includes setup and steps. Transports must honor scope.signal. */\n  budgetMs: number\n}\n\n/** Campaigns own whole-scenario retries; polling only repeats the declared read. */\nexport async function runSandboxScenario<T>(\n  spec: HarnessScenario<T>,\n  options: {\n    signal?: AbortSignal\n    measure?: HarnessScope['measure']\n  } = {},\n): Promise<HarnessReport> {\n  if (!(spec.budgetMs > 0 && Number.isSafeInteger(spec.budgetMs)))\n    throw new Error('Scenario budgetMs must be positive and finite')\n  const signal = AbortSignal.any([\n    AbortSignal.timeout(spec.budgetMs),\n    ...(options.signal ? [options.signal] : []),\n  ])\n  const report: HarnessReport = { pass: false, assertions: [], timings: [] }\n  let measurementFailures = 0\n  const retirements: Array<() => Promise<unknown>> = []\n  const scope: HarnessScope = {\n    signal,\n    check: (...assertions) => report.assertions.push(...assertions),\n    defer: (retire) => retirements.push(retire),\n    async measure(name, run) {\n      const start = performance.now()\n      let measured: HarnessTiming = { name, durationMs: 0, success: true }\n      try {\n        if (options.measure) measured = await options.measure(name, run)\n        else {\n          const result = (await run()) as { success?: boolean; error?: string } | undefined\n          if (result?.success === false)\n            throw new Error(result.error ?? 'measured operation returned success=false')\n          measured.durationMs = Math.round(performance.now() - start)\n        }\n      } catch (error) {\n        measured = {\n          name,\n          durationMs: Math.round(performance.now() - start),\n          success: false,\n          error: error instanceof Error ? error.message : String(error),\n        }\n      }\n      if (!measured.success) measurementFailures++\n      report.timings.push(measured)\n      return measured\n    },\n    async resource<T>(input: HarnessResource<T>) {\n      let resource: T | undefined\n      const timing = await scope.measure(input.name, async () => {\n        const owned = await input.create(signal)\n        resource = owned\n        scope.defer(() => input.retire(owned))\n        input.onCreate?.(owned)\n        await input.ready?.(owned, signal)\n      })\n      return { resource, timing }\n    },\n    async poll(read, ready, budget) {\n      if (!(budget.timeoutMs > 0 && budget.intervalMs > 0))\n        throw new Error('Polling timeout and interval must be positive')\n      const deadline = performance.now() + budget.timeoutMs\n      while (performance.now() < deadline) {\n        signal.throwIfAborted()\n        const value = await read()\n        if (ready(value)) return value\n        await delay(\n          Math.min(budget.intervalMs, Math.max(0, deadline - performance.now())),\n          undefined,\n          { signal },\n        )\n      }\n      return undefined\n    },\n  }\n  try {\n    signal.throwIfAborted()\n    const state = await spec.setup(scope)\n    for (const step of spec.steps ?? []) {\n      signal.throwIfAborted()\n      if (step.when && !step.when(state)) continue\n      const failuresBefore = measurementFailures\n      const timing = await scope.measure(step.name, () => step.run(state, scope))\n      if (step.advisory) measurementFailures = failuresBefore\n      scope.check(...(step.expect?.(state, timing) ?? []))\n    }\n    signal.throwIfAborted()\n    scope.check(...(spec.expectations?.(state) ?? []))\n    report.metadata = spec.metadata?.(state)\n  } catch (error) {\n    report.error = error instanceof Error ? error.message : String(error)\n  } finally {\n    for (const retire of retirements.reverse()) {\n      const measured = await scope.measure('cleanup', retire)\n      if (!measured.success)\n        scope.check({ name: 'cleanup', status: 'fail', message: measured.error })\n    }\n  }\n  report.pass =\n    !report.error &&\n    measurementFailures === 0 &&\n    report.assertions.every((a) => a.soft || a.status === 'pass')\n  return report\n}\n"],"mappings":";;;AA6DA,IAAa,0BAAb,MAA8D;CAExC;CADpB,KAAK;CACL,YAAY,UAAkD,CAAC,GAAG;EAA9C,KAAA,UAAA;CAA+C;CAEnE,MAAM,KACJ,OACA,SACA,QACwB;EACxB,MAAM,SAAS,MAAM,kBAAkB;GACrC;GACA,OAAO;GACP,KAAK,OAAO,OAAO,KAAK,QAAQ;GAChC,KAAK;IAAE,GAAG,QAAQ;IAAK,GAAG,KAAK,QAAQ;IAAK,GAAG,OAAO;GAAI;GAC1D,SAAS;GACT,WAAW,OAAO,aAAa;GAC/B,gBAAgB,OAAO,kBAAkB;EAC3C,CAAC;EAID,MAAM,SACJ,CAAC,OAAO,mBAAmB,UAAU,UAAU,OAAO,aAClD,OAAO,WAAW,MAAM,OAAO,QAAQ,OAAO,QAAQ,OAAO,QAAQ,IACrE,KAAA;EACN,OAAO;GACL;GACA,UAAU,OAAO;GACjB,QAAQ,OAAO;GACf,QAAQ,OAAO,cAAc,OAAO,SAAS,OAAO,cAAc,OAAO;GACzE,QAAQ,OAAO;GACf,YAAY,QAAQ;GACpB,aAAa,QAAQ;GACrB,iBAAiB,OAAO,mBAAmB,KAAA;GAC3C,iBAAiB,OAAO,mBAAmB,KAAA;EAC7C;CACF;AACF;AAcA,IAAa,iBAAb,MAA4B;CACN;CAApB,YAAY,SAAgC,IAAI,wBAAwB,GAAG;EAAvD,KAAA,SAAA;CAAwD;CAE5E,MAAM,IAAI,QAAuB,SAAsD;EACrF,MAAM,SAAS,MAAM,QAAQ,QAAQ;GACnC,MAAM,WAAW,KAAK,OAAO,GAAG;GAChC,OAAO,OAAO;GACd,SAAS;IAAC,OAAO;IAAc,OAAO;IAAY,OAAO;GAAW,CAAC,CAClE,OAAO,OAAO,CAAC,CACf,KAAK,MAAM;EAChB,CAAC;EACD,MAAM,SAA+B;GAAE,QAAQ;GAAO,aAAa;GAAG,OAAO;EAAE;EAC/E,IAAI;GACF,KAAK,MAAM,SAAS;IAAC;IAAS;IAAO;GAAM,GAAY;IACrD,MAAM,UAAU,OAAO,GAAG,MAAM;IAChC,IAAI,CAAC,SAAS;IACd,MAAM,WAAW,MAAM,KAAK,OAAO,KAAK,OAAO,SAAS,MAAM;IAC9D,OAAO,SAAS;IAChB,OAAO,eAAe,SAAS;IAC/B,MAAM,SAAS,CAAC,SAAS,mBAAmB,SAAS,aAAa;IAClE,IAAI,UAAU,UAAU,CAAC,QAAQ;KAC/B,MAAM,OAAO,KACX,SAAS,kBACL,GAAG,MAAM,mBAAmB,SAAS,OAAO,MAC5C,GAAG,MAAM,gBAAgB,SAAS,SAAS,IAC/C;MACE,UAAU,SAAS;MACnB,QAAQ,OAAO;KACjB,CACF;KACA,OAAO;IACT;GACF;GACA,MAAM,OAAO,OAAO;GACpB,OAAO,SAAS,CAAC,QAAS,CAAC,KAAK,mBAAmB,KAAK,aAAa;GACrE,OAAO,QACL,MAAM,eAAe,KAAA,KAAa,KAAK,aAAa,KAC/C,KAAK,eAAe,KAAK,KAAK,aAC/B,OAAO,SACL,IACA;GACR,MAAM,OAAO,IAAI;IACf,UAAU,MAAM;IAChB,YAAY,MAAM;IAClB,aAAa,MAAM;IACnB,QAAQ,OAAO,SAAS,OAAO;IAC/B,QAAQ,OAAO;GACjB,CAAyB;EAC3B,SAAS,KAAK;GACZ,MAAM,OAAO,KAAK,eAAe,QAAQ,MAAM,OAAO,GAAG,CAAC;GAC1D,MAAM;EACR;EACA,OAAO;CACT;AACF;;AA6DA,eAAsB,mBACpB,MACA,UAGI,CAAC,GACmB;CACxB,IAAI,EAAE,KAAK,WAAW,KAAK,OAAO,cAAc,KAAK,QAAQ,IAC3D,MAAM,IAAI,MAAM,+CAA+C;CACjE,MAAM,SAAS,YAAY,IAAI,CAC7B,YAAY,QAAQ,KAAK,QAAQ,GACjC,GAAI,QAAQ,SAAS,CAAC,QAAQ,MAAM,IAAI,CAAC,CAC3C,CAAC;CACD,MAAM,SAAwB;EAAE,MAAM;EAAO,YAAY,CAAC;EAAG,SAAS,CAAC;CAAE;CACzE,IAAI,sBAAsB;CAC1B,MAAM,cAA6C,CAAC;CACpD,MAAM,QAAsB;EAC1B;EACA,QAAQ,GAAG,eAAe,OAAO,WAAW,KAAK,GAAG,UAAU;EAC9D,QAAQ,WAAW,YAAY,KAAK,MAAM;EAC1C,MAAM,QAAQ,MAAM,KAAK;GACvB,MAAM,QAAQ,YAAY,IAAI;GAC9B,IAAI,WAA0B;IAAE;IAAM,YAAY;IAAG,SAAS;GAAK;GACnE,IAAI;IACF,IAAI,QAAQ,SAAS,WAAW,MAAM,QAAQ,QAAQ,MAAM,GAAG;SAC1D;KACH,MAAM,SAAU,MAAM,IAAI;KAC1B,IAAI,QAAQ,YAAY,OACtB,MAAM,IAAI,MAAM,OAAO,SAAS,2CAA2C;KAC7E,SAAS,aAAa,KAAK,MAAM,YAAY,IAAI,IAAI,KAAK;IAC5D;GACF,SAAS,OAAO;IACd,WAAW;KACT;KACA,YAAY,KAAK,MAAM,YAAY,IAAI,IAAI,KAAK;KAChD,SAAS;KACT,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;IAC9D;GACF;GACA,IAAI,CAAC,SAAS,SAAS;GACvB,OAAO,QAAQ,KAAK,QAAQ;GAC5B,OAAO;EACT;EACA,MAAM,SAAY,OAA2B;GAC3C,IAAI;GACJ,MAAM,SAAS,MAAM,MAAM,QAAQ,MAAM,MAAM,YAAY;IACzD,MAAM,QAAQ,MAAM,MAAM,OAAO,MAAM;IACvC,WAAW;IACX,MAAM,YAAY,MAAM,OAAO,KAAK,CAAC;IACrC,MAAM,WAAW,KAAK;IACtB,MAAM,MAAM,QAAQ,OAAO,MAAM;GACnC,CAAC;GACD,OAAO;IAAE;IAAU;GAAO;EAC5B;EACA,MAAM,KAAK,MAAM,OAAO,QAAQ;GAC9B,IAAI,EAAE,OAAO,YAAY,KAAK,OAAO,aAAa,IAChD,MAAM,IAAI,MAAM,+CAA+C;GACjE,MAAM,WAAW,YAAY,IAAI,IAAI,OAAO;GAC5C,OAAO,YAAY,IAAI,IAAI,UAAU;IACnC,OAAO,eAAe;IACtB,MAAM,QAAQ,MAAM,KAAK;IACzB,IAAI,MAAM,KAAK,GAAG,OAAO;IACzB,MAAMA,WACJ,KAAK,IAAI,OAAO,YAAY,KAAK,IAAI,GAAG,WAAW,YAAY,IAAI,CAAC,CAAC,GACrE,KAAA,GACA,EAAE,OAAO,CACX;GACF;EAEF;CACF;CACA,IAAI;EACF,OAAO,eAAe;EACtB,MAAM,QAAQ,MAAM,KAAK,MAAM,KAAK;EACpC,KAAK,MAAM,QAAQ,KAAK,SAAS,CAAC,GAAG;GACnC,OAAO,eAAe;GACtB,IAAI,KAAK,QAAQ,CAAC,KAAK,KAAK,KAAK,GAAG;GACpC,MAAM,iBAAiB;GACvB,MAAM,SAAS,MAAM,MAAM,QAAQ,KAAK,YAAY,KAAK,IAAI,OAAO,KAAK,CAAC;GAC1E,IAAI,KAAK,UAAU,sBAAsB;GACzC,MAAM,MAAM,GAAI,KAAK,SAAS,OAAO,MAAM,KAAK,CAAC,CAAE;EACrD;EACA,OAAO,eAAe;EACtB,MAAM,MAAM,GAAI,KAAK,eAAe,KAAK,KAAK,CAAC,CAAE;EACjD,OAAO,WAAW,KAAK,WAAW,KAAK;CACzC,SAAS,OAAO;EACd,OAAO,QAAQ,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;CACtE,UAAU;EACR,KAAK,MAAM,UAAU,YAAY,QAAQ,GAAG;GAC1C,MAAM,WAAW,MAAM,MAAM,QAAQ,WAAW,MAAM;GACtD,IAAI,CAAC,SAAS,SACZ,MAAM,MAAM;IAAE,MAAM;IAAW,QAAQ;IAAQ,SAAS,SAAS;GAAM,CAAC;EAC5E;CACF;CACA,OAAO,OACL,CAAC,OAAO,SACR,wBAAwB,KACxB,OAAO,WAAW,OAAO,MAAM,EAAE,QAAQ,EAAE,WAAW,MAAM;CAC9D,OAAO;AACT"}

package/dist/{sandbox-harness--VUvMji7.d.ts → sandbox-harness-CdO35T2g.d.ts}

@@ -5,15 +5,12 @@ interface HarnessConfig {
   runCommand?: string;
   /** Test command (e.g. "pnpm test --run"). Drives the test count + pass count. */
   testCommand?: string;
-  /** Absolute cwd for the subprocess driver. Ignored by docker driver. */
   cwd?: string;
   /** Max wall-clock per phase in ms. Default 10 minutes. */
   timeoutMs?: number;
   /** Captured stdout + stderr cap, default 16 MiB; excess is dropped. */
   maxOutputBytes?: number;
-  /** Image for the docker driver. */
   image?: string;
-  /** Extra env vars (validated; shell-escaped). */
   env?: Record<string, string>;
   /** Parser for the test output — maps stdout/stderr/exit code → pass count. */
   testParser?: TestOutputParser;
@@ -73,8 +70,19 @@ interface HarnessReport {
   error?: string;
   metadata?: Record<string, unknown>;
 }
+interface HarnessResource<T> {
+  name: string;
+  create(signal: AbortSignal): Promise<T>;
+  retire(resource: T): Promise<unknown>;
+  ready?: (resource: T, signal: AbortSignal) => Promise<unknown>;
+  onCreate?: (resource: T) => void;
+}
 interface HarnessScope {
   signal: AbortSignal;
+  resource<T>(input: HarnessResource<T>): Promise<{
+    resource?: T;
+    timing: HarnessTiming;
+  }>;
   check(...assertions: HarnessAssertion[]): void;
   measure(name: string, run: () => Promise<unknown>): Promise<HarnessTiming>;
   /** Register retirement immediately after allocation; cleanup ignores workload cancellation. */
@@ -86,8 +94,10 @@ interface HarnessScope {
 }
 interface HarnessScenario<T> {
   setup(scope: HarnessScope): Promise<T>;
-  steps: ReadonlyArray<{
+  steps?: ReadonlyArray<{
     name: string;
+    /** Diagnostic measurements retain failures without gating the scenario. */
+    advisory?: boolean;
     when?: (state: T) => boolean;
     run(state: T, scope: HarnessScope): Promise<unknown>;
     expect?: (state: T, timing: HarnessTiming) => HarnessAssertion[];
@@ -103,4 +113,4 @@ declare function runSandboxScenario<T>(spec: HarnessScenario<T>, options?: {
   measure?: HarnessScope['measure'];
 }): Promise<HarnessReport>;
 export { HarnessScope as a, SandboxHarnessResult as c, HarnessScenario as i, runSandboxScenario as l, HarnessConfig as n, HarnessTiming as o, HarnessReport as r, SandboxDriver as s, HarnessAssertion as t };
-//# sourceMappingURL=sandbox-harness--VUvMji7.d.ts.map
+//# sourceMappingURL=sandbox-harness-CdO35T2g.d.ts.map

package/dist/sandbox-harness-CdO35T2g.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sandbox-harness-CdO35T2g.d.ts","names":[],"sources":["../src/sandbox-harness.ts"],"mappings":"A,UAMiB;;EAEf;;EAEA;;EAEA;EACA;;EAEA;;EAEA;EACA;EACA,MAAM;;EAEN,aAAa;;UAGE;EACf;EACA,MACE,gBACA,gBACA;IACG;IAAoB;;;UAGV;EACf;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;UAGe;EACf;EACA,KACE,OAAO,wBACP,iBACA,QAAQ,gBACP,QAAQ;;UAiDI;EACf;EACA,QAAQ;EACR,MAAM;EACN,OAAO;EACP;;EAEA;;;UA4De;EACf;EACA;EACA;EACA;EACA;EACA;;UAEe;EACf;EACA;EACA;EACA;;UAEe;EACf;EACA,YAAY;EACZ,SAAS;EACT;EACA,WAAW;;UAEI,gBAAgB;EAC/B;EACA,OAAO,QAAQ,cAAc,QAAQ;EACrC,OAAO,UAAU,IAAI;EACrB,SAAS,UAAU,GAAG,QAAQ,gBAAgB;EAC9C,YAAY,UAAU;;UAEP;EACf,QAAQ;EACR,SAAS,GAAG,OAAO,gBAAgB,KAAK;IAAU,WAAW;IAAG,QAAQ;;EACxE,SAAS,YAAY;EACrB,QAAQ,cAAc,WAAW,mBAAmB,QAAQ;;EAE5D,MAAM,cAAc;EACpB,KAAK,GACH,YAAY,QAAQ,IACpB,QAAQ,OAAO,eACf;IAAU;IAAmB;MAC5B,QAAQ;;UAEI,gBAAgB;EAC/B,MAAM,OAAO,eAAe,QAAQ;EACpC,QAAQ;IACN;;IAEA;IACA,QAAQ,OAAO;IACf,IAAI,OAAO,GAAG,OAAO,eAAe;IACpC,UAAU,OAAO,GAAG,QAAQ,kBAAkB;;EAEhD,gBAAgB,OAAO,MAAM;EAC7B,YAAY,OAAO,MAAM;;EAEzB;;;iBAIoB,mBAAmB,GACvC,MAAM,gBAAgB,IACtB;EACE,SAAS;EACT,UAAU;IAEX,QAAQ"}

package/dist/{skillopt-optimization-method-DmfZz-2d.js → skillopt-optimization-method-DpasRfXS.js}

@@ -4,17 +4,17 @@ import { t as decidePairedPromotion } from "./paired-promotion-decision-5Pl_qJLX
 import { a as heldoutSignificance, o as pairHoldout } from "./statistical-heldout-DYF5f10K.js";
 import { d as mapConcurrent } from "./journal-C1uLJi0w.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-BVH-fl18.js";
-import { $ as resolveRunDir, C as surfaceDiff, G as assertIndependentEvaluationSplit, J as exposeFinalEvidence, K as captureFinalEvidencePolicy, R as captureJudge, T as assertSearchHistoryAdmissionOptions, Y as reserveFinalEvidence, Z as runCampaign, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, d as optimizationTokenUsageFromSummary, l as combineComparisonCosts, o as executeOptimizationMethod, q as evaluationUnitMap, s as assertGepaCandidatePopulationSummary, u as costFromLedgerSummary, w as surfaceNode, x as developmentClaim } from "./llm-judge-BYBLxREr.js";
+import { $ as resolveRunDir, C as surfaceDiff, G as assertIndependentEvaluationSplit, J as exposeFinalEvidence, K as captureFinalEvidencePolicy, R as captureJudge, T as assertSearchHistoryAdmissionOptions, Y as reserveFinalEvidence, Z as runCampaign, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, d as optimizationTokenUsageFromSummary, l as combineComparisonCosts, o as executeOptimizationMethod, q as evaluationUnitMap, s as assertGepaCandidatePopulationSummary, u as costFromLedgerSummary, w as surfaceNode, x as developmentClaim } from "./llm-judge-BiUljZDR.js";
 import { D as assertCompleteCampaign, T as assertCampaignDesign, f as campaignBreakdown, t as createCampaignEvidenceReceipt } from "./campaign-evidence-BICB3a9Z.js";
 import { t as deepFreezeCanonicalJson } from "./deep-freeze-CK8-2W7p.js";
-import { M as fsCampaignStorage, _ as resolveExternalOptimizerCallbackLimits, d as assertExternalOptimizerModelBudget, f as assertJsonValue, g as removeCredentialEnvironment, h as isExternalTextCandidate, i as startExternalOptimizerModelProxy, j as createRunCostLedger, m as isCandidateText, n as closeExternalOptimizerResources, p as assertNoCredentialValues, r as runWithCleanup, t as runExternalOptimizerProcess, v as resolveExternalOptimizerProcessLimits, y as safePathComponent } from "./external-optimizer-subprocess-wzMaA0oN.js";
+import { M as fsCampaignStorage, _ as resolveExternalOptimizerCallbackLimits, d as assertExternalOptimizerModelBudget, f as assertJsonValue, g as removeCredentialEnvironment, h as isExternalTextCandidate, i as startExternalOptimizerModelProxy, j as createRunCostLedger, m as isCandidateText, n as closeExternalOptimizerResources, p as assertNoCredentialValues, r as runWithCleanup, t as runExternalOptimizerProcess, v as resolveExternalOptimizerProcessLimits, y as safePathComponent } from "./external-optimizer-subprocess-CeHuAaMJ.js";
 import { s as surfaceContentHash } from "./json-recovery-DK4MDm5r.js";
 import { r as summarizeEvaluationUnits, t as defineEvaluationClaim } from "./claim-Dod_mjk3.js";
 import { n as openSearchLedger } from "./search-ledger-BsVi-Ngv.js";
 import { t as acquireSingleRunLock } from "./single-run-lock-BGlW3OSk.js";
 import { n as isRecord } from "./json-DtvZNAU9.js";
 import { c as combineAbortSignals, f as assertServedModelPolicy } from "./llm-client-BdSiZ4qz.js";
-import { t as startExternalOptimizerCallback } from "./external-optimizer-process-DKgrrX0J.js";
+import { t as startExternalOptimizerCallback } from "./external-optimizer-process-BgitfZ-Q.js";
 import { createHash, randomBytes } from "node:crypto";
 import { mkdir } from "node:fs/promises";
 import { isDeepStrictEqual } from "node:util";
@@ -3223,4 +3223,4 @@ async function finishSkillOptRun(context, result, outputDir, modelProxy) {
 }
 export { importExternalEvaluations as a, composeGate as c, readExternalOptimizerObservationArtifact as d, externalSurface as i, externalTextOptimizationMethod as l, compareOptimizationMethods as n, importGepaPopulation as o, gepaOptimizationMethod as r, heldOutGate as s, skillOptOptimizationMethod as t, decodeExternalTextCandidate as u };
 
-//# sourceMappingURL=skillopt-optimization-method-DmfZz-2d.js.map
+//# sourceMappingURL=skillopt-optimization-method-DpasRfXS.js.map