Fix #121: Fix WasmAgent/fresharena#115 ([milestone Milestone 4 ] open-agent-audit integration: evaluation records exported as AE

packages/aep-export/package.json

@@ -0,0 +1,33 @@
+{
+  "name": "@fresharena/aep-export",
+  "version": "0.1.0",
+  "description": "Export FreshArena FAEP evaluation records as AEP evidence bundles for open-agent-audit integration",
+  "license": "Apache-2.0",
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": "./dist/index.js",
+      "types": "./dist/index.d.ts"
+    }
+  },
+  "files": ["dist"],
+  "scripts": {
+    "build": "tsc -p tsconfig.json",
+    "typecheck": "tsc -p tsconfig.json --noEmit",
+    "test": "bun test ./src || (ls src 2>/dev/null | grep -q \"\\.test\\.ts\\|\\.spec\\.ts\" && exit 1 || exit 0)",
+    "clean": "rm -rf dist .turbo"
+  },
+  "dependencies": {
+    "@fresharena/faep-schema": "workspace:*"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/WasmAgent/fresharena.git",
+    "directory": "packages/aep-export"
+  }
+}

packages/aep-export/src/index.test.ts

@@ -0,0 +1,140 @@
+import { expect, test } from 'bun:test';
+import type { FaepRecord } from '@fresharena/faep-schema';
+import type { AepEvidenceBundle } from './index.js';
+import {
+  exportAsAepBundle,
+  exportAsAepBundles,
+  serializeBundle,
+  serializeBundles,
+} from './index.js';
+
+function makeFaepRecord(overrides?: Partial<FaepRecord>): FaepRecord {
+  const zeros = '0'.repeat(64);
+  const ones = '1'.repeat(64);
+  const twos = '2'.repeat(64);
+  const threes = '3'.repeat(64);
+  const fours = '4'.repeat(64);
+  const fives = '5'.repeat(64);
+  const sixes = '6'.repeat(64);
+  const sevens = '7'.repeat(64);
+
+  return {
+    schema_version: '0.1.0',
+    run_id: 'run-test-1',
+    task: {
+      id: 'json_transform.normalize.v0|seed-1',
+      family: 'json_transform.normalize.v0',
+      family_version: '0.1.0',
+      seed_hash: zeros,
+      spec_hash: ones,
+    },
+    solver: {
+      id: 'reference',
+      track: 'non_llm',
+      model_metadata_hash: twos,
+      workflow_hash: threes,
+      artifact_hash: fours,
+    },
+    generator: {
+      id: 'random-baseline',
+      version: '0.1.0',
+      seed_hash: zeros,
+    },
+    tester: {
+      id: 'closed-semantics',
+      version: '0.1.0',
+      tests_hash: fives,
+    },
+    verifier: {
+      package: 'json_transform_verifier',
+      version: '0.1.0',
+      result_hash: sixes,
+    },
+    environment: {
+      os: 'linux',
+      runtime: 'bun-1.3.14',
+      container_hash: sevens,
+    },
+    score: {
+      canonical_pass: true,
+      hidden_pass: true,
+      adversarial_pass: true,
+      immunity_pass: true,
+      cost: {},
+      score_vector: {},
+    },
+    replay: {
+      command: 'bun run fresharena replay --run-id run-test-1',
+      log_hash: zeros,
+    },
+    ...overrides,
+  } as FaepRecord;
+}
+
+test('exportAsAepBundle produces a valid AEP evidence envelope', () => {
+  const record = makeFaepRecord();
+  const bundle = exportAsAepBundle(record);
+
+  expect(bundle.aep_version).toBe('0.1.0');
+  expect(bundle.source.system).toBe('fresharena');
+  expect(bundle.source.component).toBe('aep-export');
+  expect(bundle.source.version).toBe('0.1.0');
+  expect(bundle.evidence).toEqual(record);
+  expect(bundle.produced_at).toBeTruthy();
+  expect(bundle.evidence_hash).toBeTruthy();
+  expect(bundle.evidence_hash).toHaveLength(64);
+});
+
+test('exportAsAepBundle accepts a custom version string', () => {
+  const record = makeFaepRecord();
+  const bundle = exportAsAepBundle(record, '2.0.0');
+
+  expect(bundle.source.version).toBe('2.0.0');
+});
+
+test('exportAsAepBundles converts multiple records', () => {
+  const records = [makeFaepRecord(), makeFaepRecord({ run_id: 'run-test-2' })];
+  const bundles = exportAsAepBundles(records);
+
+  expect(bundles).toHaveLength(2);
+  expect(bundles[0].evidence.run_id).toBe('run-test-1');
+  expect(bundles[1].evidence.run_id).toBe('run-test-2');
+  // Each bundle gets a unique evidence_hash based on its content.
+  expect(bundles[0].evidence_hash).not.toBe(bundles[1].evidence_hash);
+});
+
+test('serializeBundle produces valid JSON', () => {
+  const record = makeFaepRecord();
+  const bundle = exportAsAepBundle(record);
+  const json = serializeBundle(bundle);
+
+  const parsed = JSON.parse(json) as AepEvidenceBundle;
+  expect(parsed.aep_version).toBe('0.1.0');
+  expect(parsed.source.system).toBe('fresharena');
+  expect(parsed.evidence.run_id).toBe('run-test-1');
+});
+
+test('serializeBundles produces JSONL (one JSON object per line)', () => {
+  const records = [makeFaepRecord(), makeFaepRecord({ run_id: 'run-test-2' })];
+  const bundles = exportAsAepBundles(records);
+  const jsonl = serializeBundles(bundles);
+
+  const lines = jsonl.split('\n');
+  expect(lines).toHaveLength(2);
+
+  for (const line of lines) {
+    const parsed = JSON.parse(line) as AepEvidenceBundle;
+    expect(parsed.aep_version).toBe('0.1.0');
+  }
+});
+
+test('exportAsAepBundle preserves failure_diff annotation when present', () => {
+  const record = makeFaepRecord({
+    failure_diff: { added: true, removed: false },
+    failure_diff_hash: 'a'.repeat(64),
+  });
+  const bundle = exportAsAepBundle(record);
+
+  expect(bundle.evidence.failure_diff).toEqual({ added: true, removed: false });
+  expect(bundle.evidence.failure_diff_hash).toBe('a'.repeat(64));
+});

packages/aep-export/src/index.ts

@@ -0,0 +1,100 @@
+/**
+ * @fresharena/aep-export
+ *
+ * Exports FreshArena FAEP evaluation records as AEP (Audit Evidence Protocol)
+ * evidence bundles compatible with the open-agent-audit integration.
+ *
+ * The AEP envelope wraps each FAEP record with provenance metadata so that
+ * downstream audit pipelines (trace-pipeline, open-agent-audit) can ingest
+ * FreshArena evaluations without coupling to FAEP internals.
+ */
+
+import { createHash } from 'node:crypto';
+import type { FaepRecord } from '@fresharena/faep-schema';
+
+// ─── AEP envelope schema ────────────────────────────────────────────────────
+//
+// Minimal envelope that open-agent-audit expects. New fields are additive —
+// existing consumers must ignore unknown keys.
+
+export interface AepEvidenceBundle {
+  /** AEP schema version this envelope targets. */
+  readonly aep_version: '0.1.0';
+  /** Provenance: which system produced this evidence. */
+  readonly source: {
+    readonly system: 'fresharena';
+    readonly component: 'aep-export';
+    readonly version: string;
+  };
+  /** The original FAEP record, embedded verbatim. */
+  readonly evidence: FaepRecord;
+  /** ISO-8601 timestamp when this bundle was produced. */
+  readonly produced_at: string;
+  /** SHA-256 of the canonical JSON serialization of `evidence`. */
+  readonly evidence_hash: string;
+}
+
+// ─── Export function ──────────────────────────────────────────────────────────
+
+/**
+ * Convert a single FAEP record into an AEP evidence bundle.
+ *
+ * @param record  A validated FAEP v0.1 evaluation record.
+ * @param version Optional exporter version string (defaults to package version).
+ * @returns An `AepEvidenceBundle` ready for JSON serialization.
+ */
+export function exportAsAepBundle(record: FaepRecord, version = '0.1.0'): AepEvidenceBundle {
+  const evidenceJson = JSON.stringify(record);
+  const evidence_hash = sha256Hex(evidenceJson);
+
+  return {
+    aep_version: '0.1.0',
+    source: {
+      system: 'fresharena',
+      component: 'aep-export',
+      version,
+    },
+    evidence: record,
+    produced_at: new Date().toISOString(),
+    evidence_hash,
+  };
+}
+
+/**
+ * Convert an array of FAEP records into AEP evidence bundles.
+ *
+ * @param records Array of validated FAEP v0.1 evaluation records.
+ * @param version Optional exporter version string.
+ * @returns Array of `AepEvidenceBundle` objects.
+ */
+export function exportAsAepBundles(
+  records: readonly FaepRecord[],
+  version = '0.1.0',
+): AepEvidenceBundle[] {
+  return records.map((record) => exportAsAepBundle(record, version));
+}
+
+/**
+ * Serialize an AEP evidence bundle to a JSONL line (no trailing newline).
+ * Suitable for streaming writes to `.jsonl` audit logs.
+ */
+export function serializeBundle(bundle: AepEvidenceBundle): string {
+  return JSON.stringify(bundle);
+}
+
+/**
+ * Serialize multiple bundles as a JSONL string (one JSON object per line).
+ */
+export function serializeBundles(bundles: readonly AepEvidenceBundle[]): string {
+  return bundles.map(serializeBundle).join('\n');
+}
+
+// ─── SHA-256 helper ─────────────────────────────────────────────────────────
+
+function sha256Hex(input: string): string {
+  return createHash('sha256').update(input).digest('hex');
+}
+
+// ─── Re-exports for convenience ────────────────────────────────────────────────
+
+export type { FaepRecord } from '@fresharena/faep-schema';

packages/aep-export/tsconfig.json

@@ -0,0 +1,8 @@
+{
+  "extends": "../../tsconfig.base.json",
+  "compilerOptions": {
+    "rootDir": "src",
+    "outDir": "dist"
+  },
+  "include": ["src"]
+}