@jinn-network/client 0.1.2 → 0.1.3-canary.24b69ea4

package/dist/harnesses/impls/swe-rebench-v2-evaluator/harness.js

@@ -0,0 +1,288 @@
+/**
+ * SWE-rebench v2 Evaluator Harness — wraps the {@link SweRebenchV2Evaluator}
+ * grading library as a first-class {@link Harness} so the daemon dispatches
+ * `swe-rebench-v2.v1` evaluation tasks to it.
+ *
+ * Operator setup is automated via {@link SweRebenchV2EvaluatorHarness.onEnable}:
+ *   `jinn harnesses enable swe-rebench-v2-evaluator`
+ *   - Validates Docker + Python availability.
+ *   - Clones the upstream `SWE-rebench/SWE-rebench-V2` repo into
+ *     `<implStateDir>/upstream` (one-time; idempotent on rerun).
+ *   - Writes a `state.json` marker with `enabled: true`.
+ *
+ * `isReady()` reports `ready: false` (with a `nextStep` pointing at the
+ * enable command) until the marker file is written.
+ *
+ * Spec: docs/superpowers/specs/2026-05-06-agent-harness-solvernet-design.md §3.4
+ */
+import { existsSync, readFileSync } from 'node:fs';
+import { mkdir, writeFile } from 'node:fs/promises';
+import { spawn } from 'node:child_process';
+import { join } from 'node:path';
+import { SweRebenchV2TaskSchema, SweRebenchV2SolutionPayloadSchema, } from '../../../vendor/@jinn-network/sdk/dist/solvernets/swe-rebench-v2.js';
+import { REQUIRES_LIVE_DAEMON_READINESS } from '../../types.js';
+import { SignedEnvelopeSchema } from '../../../types/envelope.js';
+import { uploadToIpfs } from '../../../adapters/mech/ipfs.js';
+import { SweRebenchV2Evaluator } from './index.js';
+import { PythonEvalRunner } from './eval-runner.js';
+import { HttpHfFetcher } from './hf-fetcher.js';
+const DEFAULT_IPFS_REGISTRY_URL = 'https://registry.autonolas.tech';
+const UPSTREAM_REPO_URL = 'https://github.com/SWE-rebench/SWE-rebench-V2.git';
+const STATE_FILE = 'state.json';
+const ENABLE_CLI = 'jinn harnesses enable swe-rebench-v2-evaluator';
+export class SweRebenchV2EvaluatorHarness {
+    name = 'swe-rebench-v2-evaluator';
+    version = '1.0.0';
+    stub;
+    implStateDir;
+    ipfsRegistryUrl;
+    deps;
+    constructor(opts = {}) {
+        this.stub = opts.stub ?? false;
+        this.implStateDir = opts.implStateDir;
+        this.ipfsRegistryUrl = opts.ipfsRegistryUrl ?? DEFAULT_IPFS_REGISTRY_URL;
+        this.deps = opts._testDeps ?? {};
+    }
+    supports(ctx) {
+        return ctx.solverType === 'swe-rebench-v2.v1' && ctx.role === 'evaluation';
+    }
+    async canAttempt(task) {
+        if (task.solverType !== 'swe-rebench-v2.v1') {
+            return { ok: false, reason: 'solverType is not swe-rebench-v2.v1' };
+        }
+        if (task.role !== 'evaluation') {
+            return { ok: false, reason: 'role is not evaluation' };
+        }
+        if (typeof task.context?.['restorationResult'] !== 'string') {
+            return { ok: false, reason: 'context.restorationResult required' };
+        }
+        return { ok: true };
+    }
+    async isReady() {
+        if (this.stub)
+            return { ...REQUIRES_LIVE_DAEMON_READINESS };
+        if (!this.implStateDir) {
+            return {
+                ready: false,
+                reason: 'implStateDir not configured',
+                nextStep: {
+                    description: 'Configure JINN_ENGINE_IMPL_STATE_DIR_ROOT (or `engine.implStateDirRoot` in config) and restart the daemon.',
+                },
+            };
+        }
+        const state = readEnabledState(this.implStateDir);
+        if (!state) {
+            return {
+                ready: false,
+                reason: 'swe-rebench-v2 evaluator not enabled',
+                nextStep: {
+                    description: `Run \`${ENABLE_CLI}\` to clone the upstream eval harness and validate Docker + Python.`,
+                    cli: ENABLE_CLI,
+                },
+            };
+        }
+        if (!existsSync(state.upstreamRepoDir)) {
+            return {
+                ready: false,
+                reason: `upstream repo missing at ${state.upstreamRepoDir}`,
+                nextStep: {
+                    description: `Re-run \`${ENABLE_CLI}\` to re-clone the upstream eval harness.`,
+                    cli: ENABLE_CLI,
+                },
+            };
+        }
+        return { ready: true };
+    }
+    enableMetadata() {
+        return {
+            description: 'swe-rebench-v2.v1 — code-issue benchmark from the SWE-rebench/SWE-rebench-V2 dataset. ' +
+                'Requires Docker (per-instance images) and Python 3 (upstream eval.py harness). ' +
+                'Enable will clone https://github.com/SWE-rebench/SWE-rebench-V2 into the impl state directory.',
+            externalResources: [
+                { name: 'SWE-rebench-V2 (upstream)', url: 'https://github.com/SWE-rebench/SWE-rebench-V2' },
+                { name: 'Docker install', url: 'https://docs.docker.com/get-docker/' },
+            ],
+        };
+    }
+    async onEnable(_ctx) {
+        if (!this.implStateDir) {
+            return {
+                status: 'error',
+                message: 'implStateDir is not configured. Set JINN_ENGINE_IMPL_STATE_DIR_ROOT (or engine.implStateDirRoot in config) and restart the daemon, then re-run.',
+            };
+        }
+        const upstreamRepoDir = join(this.implStateDir, 'upstream');
+        // Idempotent: if already enabled and the repo is present, we're done.
+        const existing = readEnabledState(this.implStateDir);
+        if (existing && existsSync(existing.upstreamRepoDir)) {
+            return { status: 'ready', details: { upstreamRepoDir: existing.upstreamRepoDir } };
+        }
+        const run = this.deps.runCommand ?? runCommand;
+        // Validate Docker (daemon reachable) before doing any disk work.
+        const dockerCheck = await run('docker', ['info']);
+        if (dockerCheck.exitCode !== 0) {
+            return {
+                status: 'waiting_for_external_action',
+                action: {
+                    description: `Docker daemon is not reachable. Install Docker Desktop (or start the daemon), then re-run \`${ENABLE_CLI}\`.`,
+                    url: 'https://docs.docker.com/get-docker/',
+                },
+                nextInvocation: {
+                    cli: ENABLE_CLI,
+                    purpose: 'Re-validate Docker availability and continue setup.',
+                },
+            };
+        }
+        // Validate Python 3.
+        const pythonCheck = await run('python3', ['--version']);
+        if (pythonCheck.exitCode !== 0) {
+            return {
+                status: 'waiting_for_external_action',
+                action: {
+                    description: 'Python 3 is required to run the upstream eval.py harness. Install python3 and ensure it is on PATH, then re-run.',
+                },
+                nextInvocation: {
+                    cli: ENABLE_CLI,
+                    purpose: 'Re-validate Python availability and continue setup.',
+                },
+            };
+        }
+        // Clone upstream repo.
+        await mkdir(this.implStateDir, { recursive: true, mode: 0o755 });
+        if (!existsSync(upstreamRepoDir)) {
+            const clone = await run('git', [
+                'clone',
+                '--depth=1',
+                UPSTREAM_REPO_URL,
+                upstreamRepoDir,
+            ]);
+            if (clone.exitCode !== 0) {
+                return {
+                    status: 'error',
+                    message: `git clone failed (exit ${clone.exitCode}): ${clone.stderr.slice(-500)}`,
+                };
+            }
+        }
+        // Persist marker.
+        const state = {
+            schemaVersion: 'swe-rebench-v2-evaluator-state.v1',
+            enabled: true,
+            enabledAt: new Date().toISOString(),
+            upstreamRepoDir,
+        };
+        await writeFile(join(this.implStateDir, STATE_FILE), JSON.stringify(state, null, 2));
+        return {
+            status: 'ready',
+            details: { upstreamRepoDir },
+        };
+    }
+    async run(ctx) {
+        if (this.stub) {
+            throw new Error('swe-rebench-v2-evaluator: stub registry cannot run evaluation (requires live daemon)');
+        }
+        if (!this.implStateDir) {
+            throw new Error('swe-rebench-v2-evaluator: implStateDir not configured. Set engine.implStateDirRoot.');
+        }
+        const state = readEnabledState(this.implStateDir);
+        if (!state) {
+            throw new Error(`swe-rebench-v2-evaluator: not enabled. Run \`${ENABLE_CLI}\`.`);
+        }
+        const task = SweRebenchV2TaskSchema.parse(ctx.task.spec);
+        // Parse the solver's restoration envelope and pull out the patch.
+        const manifestJson = ctx.task.context['restorationResult'];
+        const envelope = SignedEnvelopeSchema.parse(JSON.parse(manifestJson));
+        if (envelope.solverType !== 'swe-rebench-v2.v1' || envelope.role !== 'restoration') {
+            throw new Error(`swe-rebench-v2-evaluator: expected swe-rebench-v2.v1/restoration envelope, got ${envelope.solverType}/${envelope.role}`);
+        }
+        const solutionPayload = SweRebenchV2SolutionPayloadSchema.parse(envelope.payload);
+        const fetcher = this.deps.fetcher ?? new HttpHfFetcher();
+        const runner = this.deps.runner ?? new PythonEvalRunner({ upstreamRepoDir: state.upstreamRepoDir });
+        const evaluator = new SweRebenchV2Evaluator({ fetcher, runner });
+        const graded = await evaluator.grade({ task, solutionPayload });
+        // Pin the test log to IPFS so anyone (evaluator dispute, audit, model
+        // training) can fetch it anonymously by CID. The CID is surfaced via the
+        // verdict-artifact's metadata, not the typed Verdict payload — the
+        // schema does not couple Verdicts to daemon-derived IPFS provenance.
+        const upload = this.deps.uploadToIpfs ?? uploadToIpfs;
+        const test_log_cid = await upload(this.ipfsRegistryUrl, {
+            kind: 'swe-rebench-v2-test-log.v1',
+            instance_id: task.instance_id,
+            log: graded.test_log,
+            gradedAt: new Date().toISOString(),
+        });
+        const verdictPayload = {
+            schemaVersion: 'swe-rebench-v2-verdict.v1',
+            score: graded.score,
+            passed_match: graded.passed_match,
+            evaluator_cost_usd: 0,
+        };
+        const verdictArtifactPayload = {
+            schemaVersion: 'swe-rebench-v2-verdict-artifact.v1',
+            verdict: verdictPayload,
+            informational: {
+                instance_id: task.instance_id,
+                round_month: task.round_month,
+                test_log_cid,
+            },
+        };
+        await writeFile(join(ctx.workingDir, 'swe-rebench-v2-verdict.json'), `${JSON.stringify(verdictArtifactPayload, null, 2)}\n`, 'utf8');
+        return {
+            venueRef: { name: 'swe-rebench-v2' },
+            gating: {
+                score: verdictPayload.score,
+                passed_match: verdictPayload.passed_match,
+            },
+            informational: {
+                instance_id: task.instance_id,
+                round_month: task.round_month,
+                test_log_cid,
+            },
+            verdictPayload: verdictPayload,
+            artifacts: [
+                {
+                    path: 'swe-rebench-v2-verdict.json',
+                    artifactType: 'swe-rebench_v2_verdict',
+                    metadata: {
+                        score: verdictPayload.score,
+                        passed_match: verdictPayload.passed_match,
+                        test_log_cid,
+                    },
+                    access: { priceUsdc: '0' },
+                },
+            ],
+        };
+    }
+}
+async function runCommand(bin, args, opts = {}) {
+    return new Promise((resolve, reject) => {
+        const child = spawn(bin, args, { ...opts, stdio: ['ignore', 'pipe', 'pipe'] });
+        let stdout = '';
+        let stderr = '';
+        child.stdout?.on('data', (d) => {
+            stdout += d.toString();
+        });
+        child.stderr?.on('data', (d) => {
+            stderr += d.toString();
+        });
+        child.on('error', reject);
+        child.on('close', (code) => {
+            resolve({ exitCode: code ?? 1, stdout, stderr });
+        });
+    });
+}
+function readEnabledState(implStateDir) {
+    const path = join(implStateDir, STATE_FILE);
+    if (!existsSync(path))
+        return null;
+    try {
+        const raw = JSON.parse(readFileSync(path, 'utf8'));
+        if (raw.enabled !== true || typeof raw.upstreamRepoDir !== 'string')
+            return null;
+        return raw;
+    }
+    catch {
+        return null;
+    }
+}
+export { runCommand };
+//# sourceMappingURL=harness.js.map

package/dist/harnesses/impls/swe-rebench-v2-evaluator/harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/swe-rebench-v2-evaluator/harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,KAAK,EAAqB,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EACL,sBAAsB,EACtB,iCAAiC,GAElC,MAAM,6CAA6C,CAAC;AAUrD,OAAO,EAAE,8BAA8B,EAAE,MAAM,gBAAgB,CAAC;AAEhE,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,YAAY,EAAE,MAAM,gCAAgC,CAAC;AAC9D,OAAO,EAAE,qBAAqB,EAAmC,MAAM,YAAY,CAAC;AACpF,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAEhD,MAAM,yBAAyB,GAAG,iCAAiC,CAAC;AACpE,MAAM,iBAAiB,GAAG,mDAAmD,CAAC;AAC9E,MAAM,UAAU,GAAG,YAAY,CAAC;AAChC,MAAM,UAAU,GAAG,gDAAgD,CAAC;AAgCpE,MAAM,OAAO,4BAA4B;IAC9B,IAAI,GAAG,0BAA0B,CAAC;IAClC,OAAO,GAAG,OAAO,CAAC;IAEV,IAAI,CAAU;IACd,YAAY,CAAqB;IACjC,eAAe,CAAS;IACxB,IAAI,CAAgE;IAErF,YAAY,OAA4C,EAAE;QACxD,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,KAAK,CAAC;QAC/B,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC;QACtC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,eAAe,IAAI,yBAAyB,CAAC;QACzE,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC;IACnC,CAAC;IAED,QAAQ,CAAC,GAAgE;QACvE,OAAO,GAAG,CAAC,UAAU,KAAK,mBAAmB,IAAI,GAAG,CAAC,IAAI,KAAK,YAAY,CAAC;IAC7E,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,IAAU;QACzB,IAAI,IAAI,CAAC,UAAU,KAAK,mBAAmB,EAAE,CAAC;YAC5C,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,qCAAqC,EAAE,CAAC;QACtE,CAAC;QACD,IAAI,IAAI,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;YAC/B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,wBAAwB,EAAE,CAAC;QACzD,CAAC;QACD,IAAI,OAAO,IAAI,CAAC,OAAO,EAAE,CAAC,mBAAmB,CAAC,KAAK,QAAQ,EAAE,CAAC;YAC5D,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,oCAAoC,EAAE,CAAC;QACrE,CAAC;QACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,OAAO;QACX,IAAI,IAAI,CAAC,IAAI;YAAE,OAAO,EAAE,GAAG,8BAA8B,EAAE,CAAC;QAC5D,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,6BAA6B;gBACrC,QAAQ,EAAE;oBACR,WAAW,EACT,4GAA4G;iBAC/G;aACF,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,GAAG,gBAAgB,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAClD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,sCAAsC;gBAC9C,QAAQ,EAAE;oBACR,WAAW,EACT,SAAS,UAAU,qEAAqE;oBAC1F,GAAG,EAAE,UAAU;iBAChB;aACF,CAAC;QACJ,CAAC;QACD,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,eAAe,CAAC,EAAE,CAAC;YACvC,OAAO;gBACL,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,4BAA4B,KAAK,CAAC,eAAe,EAAE;gBAC3D,QAAQ,EAAE;oBACR,WAAW,EACT,YAAY,UAAU,2CAA2C;oBACnE,GAAG,EAAE,UAAU;iBAChB;aACF,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IACzB,CAAC;IAED,cAAc;QACZ,OAAO;YACL,WAAW,EACT,wFAAwF;gBACxF,iFAAiF;gBACjF,gGAAgG;YAClG,iBAAiB,EAAE;gBACjB,EAAE,IAAI,EAAE,2BAA2B,EAAE,GAAG,EAAE,+CAA+C,EAAE;gBAC3F,EAAE,IAAI,EAAE,gBAAgB,EAAE,GAAG,EAAE,qCAAqC,EAAE;aACvE;SACF,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,IAA0B;QACvC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,OAAO;gBACL,MAAM,EAAE,OAAO;gBACf,OAAO,EACL,iJAAiJ;aACpJ,CAAC;QACJ,CAAC;QACD,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC;QAE5D,sEAAsE;QACtE,MAAM,QAAQ,GAAG,gBAAgB,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QACrD,IAAI,QAAQ,IAAI,UAAU,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE,CAAC;YACrD,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,eAAe,EAAE,QAAQ,CAAC,eAAe,EAAE,EAAE,CAAC;QACrF,CAAC;QAED,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,IAAI,UAAU,CAAC;QAE/C,iEAAiE;QACjE,MAAM,WAAW,GAAG,MAAM,GAAG,CAAC,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;QAClD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YAC/B,OAAO;gBACL,MAAM,EAAE,6BAA6B;gBACrC,MAAM,EAAE;oBACN,WAAW,EACT,+FAA+F,UAAU,KAAK;oBAChH,GAAG,EAAE,qCAAqC;iBAC3C;gBACD,cAAc,EAAE;oBACd,GAAG,EAAE,UAAU;oBACf,OAAO,EAAE,qDAAqD;iBAC/D;aACF,CAAC;QACJ,CAAC;QAED,qBAAqB;QACrB,MAAM,WAAW,GAAG,MAAM,GAAG,CAAC,SAAS,EAAE,CAAC,WAAW,CAAC,CAAC,CAAC;QACxD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YAC/B,OAAO;gBACL,MAAM,EAAE,6BAA6B;gBACrC,MAAM,EAAE;oBACN,WAAW,EACT,kHAAkH;iBACrH;gBACD,cAAc,EAAE;oBACd,GAAG,EAAE,UAAU;oBACf,OAAO,EAAE,qDAAqD;iBAC/D;aACF,CAAC;QACJ,CAAC;QAED,uBAAuB;QACvB,MAAM,KAAK,CAAC,IAAI,CAAC,YAAY,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;QACjE,IAAI,CAAC,UAAU,CAAC,eAAe,CAAC,EAAE,CAAC;YACjC,MAAM,KAAK,GAAG,MAAM,GAAG,CAAC,KAAK,EAAE;gBAC7B,OAAO;gBACP,WAAW;gBACX,iBAAiB;gBACjB,eAAe;aAChB,CAAC,CAAC;YACH,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;gBACzB,OAAO;oBACL,MAAM,EAAE,OAAO;oBACf,OAAO,EAAE,0BAA0B,KAAK,CAAC,QAAQ,MAAM,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,EAAE;iBAClF,CAAC;YACJ,CAAC;QACH,CAAC;QAED,kBAAkB;QAClB,MAAM,KAAK,GAAiB;YAC1B,aAAa,EAAE,mCAAmC;YAClD,OAAO,EAAE,IAAI;YACb,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,eAAe;SAChB,CAAC;QACF,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,UAAU,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAErF,OAAO;YACL,MAAM,EAAE,OAAO;YACf,OAAO,EAAE,EAAE,eAAe,EAAE;SAC7B,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,GAAmB;QAC3B,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CACb,sFAAsF,CACvF,CAAC;QACJ,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvB,MAAM,IAAI,KAAK,CACb,qFAAqF,CACtF,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,GAAG,gBAAgB,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAClD,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CACb,gDAAgD,UAAU,KAAK,CAChE,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,sBAAsB,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEzD,kEAAkE;QAClE,MAAM,YAAY,GAAG,GAAG,CAAC,IAAI,CAAC,OAAQ,CAAC,mBAAmB,CAAW,CAAC;QACtE,MAAM,QAAQ,GAAG,oBAAoB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC;QACtE,IAAI,QAAQ,CAAC,UAAU,KAAK,mBAAmB,IAAI,QAAQ,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;YACnF,MAAM,IAAI,KAAK,CACb,kFAAkF,QAAQ,CAAC,UAAU,IAAI,QAAQ,CAAC,IAAI,EAAE,CACzH,CAAC;QACJ,CAAC;QACD,MAAM,eAAe,GAAG,iCAAiC,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAElF,MAAM,OAAO,GAAc,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,IAAI,aAAa,EAAE,CAAC;QACpE,MAAM,MAAM,GACV,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,IAAI,gBAAgB,CAAC,EAAE,eAAe,EAAE,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC;QACvF,MAAM,SAAS,GAAG,IAAI,qBAAqB,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEjE,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC,CAAC;QAEhE,sEAAsE;QACtE,yEAAyE;QACzE,mEAAmE;QACnE,qEAAqE;QACrE,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,IAAI,YAAY,CAAC;QACtD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,eAAe,EAAE;YACtD,IAAI,EAAE,4BAA4B;YAClC,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,GAAG,EAAE,MAAM,CAAC,QAAQ;YACpB,QAAQ,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SACnC,CAAC,CAAC;QAEH,MAAM,cAAc,GAA+B;YACjD,aAAa,EAAE,2BAA2B;YAC1C,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,kBAAkB,EAAE,CAAC;SACtB,CAAC;QACF,MAAM,sBAAsB,GAAG;YAC7B,aAAa,EAAE,oCAAoC;YACnD,OAAO,EAAE,cAAc;YACvB,aAAa,EAAE;gBACb,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,YAAY;aACb;SACF,CAAC;QACF,MAAM,SAAS,CACb,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,6BAA6B,CAAC,EACnD,GAAG,IAAI,CAAC,SAAS,CAAC,sBAAsB,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EACtD,MAAM,CACP,CAAC;QAEF,OAAO;YACL,QAAQ,EAAE,EAAE,IAAI,EAAE,gBAAgB,EAAE;YACpC,MAAM,EAAE;gBACN,KAAK,EAAE,cAAc,CAAC,KAAK;gBAC3B,YAAY,EAAE,cAAc,CAAC,YAAY;aAC1C;YACD,aAAa,EAAE;gBACb,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,YAAY;aACb;YACD,cAAc,EAAE,cAAoD;YACpE,SAAS,EAAE;gBACT;oBACE,IAAI,EAAE,6BAA6B;oBACnC,YAAY,EAAE,wBAAwB;oBACtC,QAAQ,EAAE;wBACR,KAAK,EAAE,cAAc,CAAC,KAAK;wBAC3B,YAAY,EAAE,cAAc,CAAC,YAAY;wBACzC,YAAY;qBACb;oBACD,MAAM,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE;iBAC3B;aACF;SACF,CAAC;IACJ,CAAC;CACF;AAUD,KAAK,UAAU,UAAU,CACvB,GAAW,EACX,IAAc,EACd,OAAqB,EAAE;IAEvB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,GAAG,IAAI,EAAE,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QAC/E,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE;YAC7B,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC;QACzB,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE;YAC7B,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC;QACzB,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC1B,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE;YACzB,OAAO,CAAC,EAAE,QAAQ,EAAE,IAAI,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;QACnD,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,gBAAgB,CAAC,YAAoB;IAC5C,MAAM,IAAI,GAAG,IAAI,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC;IAC5C,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IACnC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAA0B,CAAC;QAC5E,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,IAAI,OAAO,GAAG,CAAC,eAAe,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC;QACjF,OAAO,GAAmB,CAAC;IAC7B,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,OAAO,EAAE,UAAU,EAAE,CAAC"}

package/dist/harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.d.ts

@@ -0,0 +1,33 @@
+/**
+ * HfFetcher implementation that resolves a single HF dataset row by
+ * `(dataset, split, instance_id)` via the public HF datasets-server.
+ *
+ * The datasets-server `/rows` endpoint paginates with `offset` + `length`
+ * (max length is 100 per request). Splits are typically <250 rows, so we
+ * walk pages until either the `instance_id` is found or we hit a hard
+ * cap (default 1000 rows = 10 pages). The harness fails loud rather than
+ * silently grading against a missing instance.
+ */
+import type { HfFetcher, HfRow } from './index.js';
+export interface HttpHfFetcherOptions {
+    /** Override the default datasets-server base URL. */
+    baseUrl?: string;
+    /** Page size (max 100 per HF API). Defaults to 100. */
+    pageSize?: number;
+    /** Hard cap on rows scanned per lookup. Defaults to 1000. */
+    maxRows?: number;
+    /** Fetch implementation. Defaults to global fetch (allows test injection). */
+    fetchImpl?: typeof fetch;
+}
+export declare class HttpHfFetcher implements HfFetcher {
+    private readonly baseUrl;
+    private readonly pageSize;
+    private readonly maxRows;
+    private readonly fetchImpl;
+    constructor(opts?: HttpHfFetcherOptions);
+    fetchTaskRow(args: {
+        hf_dataset: string;
+        hf_split: string;
+        instance_id: string;
+    }): Promise<HfRow>;
+}

package/dist/harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.js

@@ -0,0 +1,96 @@
+/**
+ * HfFetcher implementation that resolves a single HF dataset row by
+ * `(dataset, split, instance_id)` via the public HF datasets-server.
+ *
+ * The datasets-server `/rows` endpoint paginates with `offset` + `length`
+ * (max length is 100 per request). Splits are typically <250 rows, so we
+ * walk pages until either the `instance_id` is found or we hit a hard
+ * cap (default 1000 rows = 10 pages). The harness fails loud rather than
+ * silently grading against a missing instance.
+ */
+const DEFAULT_BASE_URL = 'https://datasets-server.huggingface.co/rows';
+export class HttpHfFetcher {
+    baseUrl;
+    pageSize;
+    maxRows;
+    fetchImpl;
+    constructor(opts = {}) {
+        this.baseUrl = opts.baseUrl ?? DEFAULT_BASE_URL;
+        this.pageSize = Math.min(opts.pageSize ?? 100, 100);
+        this.maxRows = opts.maxRows ?? 1000;
+        this.fetchImpl = opts.fetchImpl ?? fetch.bind(globalThis);
+    }
+    async fetchTaskRow(args) {
+        let offset = 0;
+        while (offset < this.maxRows) {
+            const url = new URL(this.baseUrl);
+            url.searchParams.set('dataset', args.hf_dataset);
+            url.searchParams.set('config', 'default');
+            url.searchParams.set('split', args.hf_split);
+            url.searchParams.set('offset', String(offset));
+            url.searchParams.set('length', String(this.pageSize));
+            const res = await this.fetchImpl(url.toString());
+            if (!res.ok) {
+                throw new Error(`HF datasets-server returned ${res.status} for ${args.hf_dataset}/${args.hf_split}`);
+            }
+            const body = (await res.json());
+            const rows = (body.rows ?? []).map((r) => r.row ?? {});
+            if (rows.length === 0)
+                break;
+            for (const row of rows) {
+                if (row['instance_id'] === args.instance_id) {
+                    return rowToHfRow(row, args.instance_id);
+                }
+            }
+            if (rows.length < this.pageSize)
+                break; // exhausted
+            offset += rows.length;
+        }
+        throw new Error(`instance_id ${args.instance_id} not found in ${args.hf_dataset}/${args.hf_split} ` +
+            `(scanned up to ${this.maxRows} rows)`);
+    }
+}
+function rowToHfRow(row, instance_id) {
+    const image_name = stringField(row['image_name']);
+    if (!image_name) {
+        throw new Error(`HF row for ${instance_id} missing image_name`);
+    }
+    const repo = stringField(row['repo']);
+    if (!repo) {
+        throw new Error(`HF row for ${instance_id} missing repo`);
+    }
+    const installConfig = (row['install_config'] ?? {});
+    const install = installConfig['install'];
+    const test_cmd = installConfig['test_cmd'];
+    const log_parser = installConfig['log_parser'];
+    return {
+        instance_id,
+        repo,
+        image_name,
+        FAIL_TO_PASS: arrayOfStrings(row['FAIL_TO_PASS']),
+        PASS_TO_PASS: arrayOfStrings(row['PASS_TO_PASS']),
+        test_patch: stringField(row['test_patch']) ?? '',
+        install_config: {
+            install: typeof install === 'string'
+                ? install
+                : Array.isArray(install)
+                    ? install
+                    : undefined,
+            test_cmd: typeof test_cmd === 'string'
+                ? test_cmd
+                : Array.isArray(test_cmd)
+                    ? test_cmd
+                    : '',
+            log_parser: typeof log_parser === 'string' ? log_parser : '',
+        },
+    };
+}
+function stringField(v) {
+    return typeof v === 'string' ? v : undefined;
+}
+function arrayOfStrings(v) {
+    if (!Array.isArray(v))
+        return [];
+    return v.filter((s) => typeof s === 'string');
+}
+//# sourceMappingURL=hf-fetcher.js.map

package/dist/harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"hf-fetcher.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/swe-rebench-v2-evaluator/hf-fetcher.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAeH,MAAM,gBAAgB,GAAG,6CAA6C,CAAC;AAEvE,MAAM,OAAO,aAAa;IACP,OAAO,CAAS;IAChB,QAAQ,CAAS;IACjB,OAAO,CAAS;IAChB,SAAS,CAAe;IAEzC,YAAY,OAA6B,EAAE;QACzC,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,gBAAgB,CAAC;QAChD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,IAAI,GAAG,EAAE,GAAG,CAAC,CAAC;QACpD,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC;QACpC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAC5D,CAAC;IAED,KAAK,CAAC,YAAY,CAAC,IAIlB;QACC,IAAI,MAAM,GAAG,CAAC,CAAC;QACf,OAAO,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;YAC7B,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAClC,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;YACjD,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;YAC1C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC7C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;YAC/C,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC;YAEtD,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,EAAE,CAAC,CAAC;YACjD,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;gBACZ,MAAM,IAAI,KAAK,CACb,+BAA+B,GAAG,CAAC,MAAM,QAAQ,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,QAAQ,EAAE,CACpF,CAAC;YACJ,CAAC;YACD,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAwD,CAAC;YACvF,MAAM,IAAI,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC;YACvD,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;gBAAE,MAAM;YAE7B,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;gBACvB,IAAI,GAAG,CAAC,aAAa,CAAC,KAAK,IAAI,CAAC,WAAW,EAAE,CAAC;oBAC5C,OAAO,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;gBAC3C,CAAC;YACH,CAAC;YAED,IAAI,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,QAAQ;gBAAE,MAAM,CAAC,YAAY;YACpD,MAAM,IAAI,IAAI,CAAC,MAAM,CAAC;QACxB,CAAC;QACD,MAAM,IAAI,KAAK,CACb,eAAe,IAAI,CAAC,WAAW,iBAAiB,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,QAAQ,GAAG;YACjF,kBAAkB,IAAI,CAAC,OAAO,QAAQ,CACzC,CAAC;IACJ,CAAC;CACF;AAED,SAAS,UAAU,CAAC,GAA4B,EAAE,WAAmB;IACnE,MAAM,UAAU,GAAG,WAAW,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC;IAClD,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,MAAM,IAAI,KAAK,CAAC,cAAc,WAAW,qBAAqB,CAAC,CAAC;IAClE,CAAC;IACD,MAAM,IAAI,GAAG,WAAW,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC;IACtC,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,MAAM,IAAI,KAAK,CAAC,cAAc,WAAW,eAAe,CAAC,CAAC;IAC5D,CAAC;IACD,MAAM,aAAa,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,IAAI,EAAE,CAA4B,CAAC;IAC/E,MAAM,OAAO,GAAG,aAAa,CAAC,SAAS,CAAC,CAAC;IACzC,MAAM,QAAQ,GAAG,aAAa,CAAC,UAAU,CAAC,CAAC;IAC3C,MAAM,UAAU,GAAG,aAAa,CAAC,YAAY,CAAC,CAAC;IAC/C,OAAO;QACL,WAAW;QACX,IAAI;QACJ,UAAU;QACV,YAAY,EAAE,cAAc,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QACjD,YAAY,EAAE,cAAc,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;QACjD,UAAU,EAAE,WAAW,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,IAAI,EAAE;QAChD,cAAc,EAAE;YACd,OAAO,EACL,OAAO,OAAO,KAAK,QAAQ;gBACzB,CAAC,CAAC,OAAO;gBACT,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;oBACtB,CAAC,CAAE,OAAoB;oBACvB,CAAC,CAAC,SAAS;YACjB,QAAQ,EACN,OAAO,QAAQ,KAAK,QAAQ;gBAC1B,CAAC,CAAC,QAAQ;gBACV,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC;oBACvB,CAAC,CAAE,QAAqB;oBACxB,CAAC,CAAC,EAAE;YACV,UAAU,EAAE,OAAO,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE;SAC7D;KACF,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,CAAU;IAC7B,OAAO,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/C,CAAC;AAED,SAAS,cAAc,CAAC,CAAU;IAChC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;QAAE,OAAO,EAAE,CAAC;IACjC,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,CAAC,CAAC;AAC7D,CAAC"}

package/dist/harnesses/impls/swe-rebench-v2-evaluator/index.d.ts

@@ -0,0 +1,61 @@
+/**
+ * SWE-rebench v2 evaluator. Wraps the upstream `scripts/eval.py` (MIT) to
+ * grade Solver patches against the per-instance Docker test suite.
+ *
+ * Spec: docs/superpowers/specs/2026-05-06-agent-harness-solvernet-design.md §3.4
+ */
+import type { SweRebenchV2Task, SweRebenchV2SolutionPayload, SweRebenchV2VerdictPayload } from '../../../vendor/@jinn-network/sdk/dist/solvernets/swe-rebench-v2.js';
+export interface HfRow {
+    instance_id: string;
+    repo: string;
+    image_name: string;
+    FAIL_TO_PASS: string[];
+    PASS_TO_PASS: string[];
+    test_patch: string;
+    install_config: {
+        install?: string | string[];
+        test_cmd: string | string[];
+        log_parser: string;
+    };
+}
+export interface HfFetcher {
+    fetchTaskRow(args: {
+        hf_dataset: string;
+        hf_split: string;
+        instance_id: string;
+    }): Promise<HfRow>;
+}
+export interface EvalRunner {
+    runEval(args: {
+        instance_id: string;
+        repo: string;
+        image: string;
+        patch: string;
+        test_patch: string;
+        install?: string | string[];
+        test_cmd: string | string[];
+        log_parser: string;
+        fail_to_pass: string[];
+        pass_to_pass: string[];
+    }): Promise<{
+        passed_match: boolean;
+        passed: string[];
+        failed: string[];
+        log: string;
+        exitCode: number;
+    }>;
+}
+export interface GradeArgs {
+    task: SweRebenchV2Task;
+    solutionPayload: SweRebenchV2SolutionPayload;
+}
+export declare class SweRebenchV2Evaluator {
+    private readonly deps;
+    constructor(deps: {
+        fetcher: HfFetcher;
+        runner: EvalRunner;
+    });
+    grade(args: GradeArgs): Promise<SweRebenchV2VerdictPayload & {
+        test_log: string;
+    }>;
+}

package/dist/harnesses/impls/swe-rebench-v2-evaluator/index.js

@@ -0,0 +1,42 @@
+/**
+ * SWE-rebench v2 evaluator. Wraps the upstream `scripts/eval.py` (MIT) to
+ * grade Solver patches against the per-instance Docker test suite.
+ *
+ * Spec: docs/superpowers/specs/2026-05-06-agent-harness-solvernet-design.md §3.4
+ */
+export class SweRebenchV2Evaluator {
+    deps;
+    constructor(deps) {
+        this.deps = deps;
+    }
+    async grade(args) {
+        const row = await this.deps.fetcher.fetchTaskRow({
+            hf_dataset: args.task.hf_dataset,
+            hf_split: args.task.hf_split,
+            instance_id: args.task.instance_id,
+        });
+        if (!row.image_name) {
+            throw new Error(`HF row for ${args.task.instance_id} missing image_name`);
+        }
+        const result = await this.deps.runner.runEval({
+            instance_id: row.instance_id,
+            repo: row.repo,
+            image: row.image_name,
+            patch: args.solutionPayload.patch,
+            test_patch: row.test_patch,
+            install: row.install_config.install,
+            test_cmd: row.install_config.test_cmd,
+            log_parser: row.install_config.log_parser,
+            fail_to_pass: row.FAIL_TO_PASS,
+            pass_to_pass: row.PASS_TO_PASS,
+        });
+        return {
+            schemaVersion: 'swe-rebench-v2-verdict.v1',
+            score: result.passed_match ? 1 : 0,
+            passed_match: result.passed_match,
+            evaluator_cost_usd: 0, // populated by caller from runtime metrics
+            test_log: result.log,
+        };
+    }
+}
+//# sourceMappingURL=index.js.map

package/dist/harnesses/impls/swe-rebench-v2-evaluator/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/swe-rebench-v2-evaluator/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAgDH,MAAM,OAAO,qBAAqB;IAEb;IADnB,YACmB,IAAgD;QAAhD,SAAI,GAAJ,IAAI,CAA4C;IAChE,CAAC;IAEJ,KAAK,CAAC,KAAK,CAAC,IAAe;QACzB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YAC/C,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,UAAU;YAChC,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ;YAC5B,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW;SACnC,CAAC,CAAC;QACH,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,CAAC;YACpB,MAAM,IAAI,KAAK,CAAC,cAAc,IAAI,CAAC,IAAI,CAAC,WAAW,qBAAqB,CAAC,CAAC;QAC5E,CAAC;QACD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC;YAC5C,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,IAAI,EAAE,GAAG,CAAC,IAAI;YACd,KAAK,EAAE,GAAG,CAAC,UAAU;YACrB,KAAK,EAAE,IAAI,CAAC,eAAe,CAAC,KAAK;YACjC,UAAU,EAAE,GAAG,CAAC,UAAU;YAC1B,OAAO,EAAE,GAAG,CAAC,cAAc,CAAC,OAAO;YACnC,QAAQ,EAAE,GAAG,CAAC,cAAc,CAAC,QAAQ;YACrC,UAAU,EAAE,GAAG,CAAC,cAAc,CAAC,UAAU;YACzC,YAAY,EAAE,GAAG,CAAC,YAAY;YAC9B,YAAY,EAAE,GAAG,CAAC,YAAY;SAC/B,CAAC,CAAC;QACH,OAAO;YACL,aAAa,EAAE,2BAA2B;YAC1C,KAAK,EAAE,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAClC,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,kBAAkB,EAAE,CAAC,EAAG,2CAA2C;YACnE,QAAQ,EAAE,MAAM,CAAC,GAAG;SACrB,CAAC;IACJ,CAAC;CACF"}

package/dist/harnesses/manifest/index.d.ts

@@ -0,0 +1,3 @@
+export { loadManifest } from './load.js';
+export { canonicaliseManifest, verifyManifestSignature } from './verify.js';
+export type { JinnManifest, SignerTrustEntry, CapabilityAllowEntry, ManifestRpcAllow, TypedDataAllowEntry, } from './types.js';

package/dist/harnesses/manifest/index.js

@@ -0,0 +1,3 @@
+export { loadManifest } from './load.js';
+export { canonicaliseManifest, verifyManifestSignature } from './verify.js';
+//# sourceMappingURL=index.js.map

package/dist/harnesses/manifest/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/harnesses/manifest/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AACzC,OAAO,EAAE,oBAAoB,EAAE,uBAAuB,EAAE,MAAM,aAAa,CAAC"}

package/dist/harnesses/manifest/load.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Load a `jinn.manifest.json` from disk and validate it against the
+ * Jinn manifest JSON Schema. Pure (no signature verification — that is
+ * `verifyManifestSignature` in `verify.ts`).
+ */
+import type { JinnManifest } from './types.js';
+export declare function loadManifest(absPath: string): Promise<JinnManifest>;

package/dist/harnesses/manifest/load.js

@@ -0,0 +1,49 @@
+/**
+ * Load a `jinn.manifest.json` from disk and validate it against the
+ * Jinn manifest JSON Schema. Pure (no signature verification — that is
+ * `verifyManifestSignature` in `verify.ts`).
+ */
+import { readFile } from 'node:fs/promises';
+import { fileURLToPath } from 'node:url';
+import { resolve } from 'node:path';
+import { Ajv2020 } from 'ajv/dist/2020.js';
+import * as addFormatsModule from 'ajv-formats';
+// `ajv-formats` is published as CJS; under Node16 module resolution
+// the callable plugin lives on `.default`.
+const addFormats = addFormatsModule
+    .default;
+const SCHEMA_PATH = fileURLToPath(new URL('../../../schemas/jinn-manifest-v1.json', import.meta.url));
+let validatorPromise = null;
+let lastValidator;
+async function getValidator() {
+    if (validatorPromise)
+        return validatorPromise;
+    validatorPromise = (async () => {
+        const ajv = new Ajv2020({ allErrors: true, strict: false });
+        addFormats(ajv);
+        const schema = JSON.parse(await readFile(SCHEMA_PATH, 'utf8'));
+        const compiled = ajv.compile(schema);
+        lastValidator = compiled;
+        return compiled;
+    })();
+    return validatorPromise;
+}
+export async function loadManifest(absPath) {
+    const text = await readFile(resolve(absPath), 'utf8');
+    let parsed;
+    try {
+        parsed = JSON.parse(text);
+    }
+    catch (err) {
+        throw new Error(`manifest at ${absPath} is not valid JSON: ${err.message}`);
+    }
+    const validate = await getValidator();
+    if (!validate(parsed)) {
+        const errs = (lastValidator?.errors ?? [])
+            .map((e) => `${e.instancePath} ${e.message}`)
+            .join('; ');
+        throw new Error(`manifest at ${absPath} failed schema validation: ${errs}`);
+    }
+    return parsed;
+}
+//# sourceMappingURL=load.js.map

package/dist/harnesses/manifest/load.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"load.js","sourceRoot":"","sources":["../../../src/harnesses/manifest/load.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,KAAK,gBAAgB,MAAM,aAAa,CAAC;AAEhD,oEAAoE;AACpE,2CAA2C;AAC3C,MAAM,UAAU,GAAI,gBAA4D;KAC7E,OAAO,CAAC;AAGX,MAAM,WAAW,GAAG,aAAa,CAC/B,IAAI,GAAG,CAAC,wCAAwC,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CACnE,CAAC;AAEF,IAAI,gBAAgB,GAA4C,IAAI,CAAC;AACrE,IAAI,aAEH,CAAC;AAEF,KAAK,UAAU,YAAY;IACzB,IAAI,gBAAgB;QAAE,OAAO,gBAAgB,CAAC;IAC9C,gBAAgB,GAAG,CAAC,KAAK,IAAI,EAAE;QAC7B,MAAM,GAAG,GAAG,IAAI,OAAO,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,CAAC;QAC5D,UAAU,CAAC,GAAG,CAAC,CAAC;QAChB,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC,CAAC;QAC/D,MAAM,QAAQ,GAAG,GAAG,CAAC,OAAO,CAAC,MAAM,CAElC,CAAC;QACF,aAAa,GAAG,QAAQ,CAAC;QACzB,OAAO,QAAQ,CAAC;IAClB,CAAC,CAAC,EAAE,CAAC;IACL,OAAO,gBAAgB,CAAC;AAC1B,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,OAAe;IAChD,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC,CAAC;IACtD,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CACb,eAAe,OAAO,uBAAwB,GAAa,CAAC,OAAO,EAAE,CACtE,CAAC;IACJ,CAAC;IACD,MAAM,QAAQ,GAAG,MAAM,YAAY,EAAE,CAAC;IACtC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,GAAG,CAAC,aAAa,EAAE,MAAM,IAAI,EAAE,CAAC;aACvC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC,OAAO,EAAE,CAAC;aAC5C,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,MAAM,IAAI,KAAK,CACb,eAAe,OAAO,8BAA8B,IAAI,EAAE,CAC3D,CAAC;IACJ,CAAC;IACD,OAAO,MAAsB,CAAC;AAChC,CAAC"}