@jinn-network/client 0.1.9 → 0.2.0-canary.0b79cd5d

package/dist/harnesses/impls/jinn-repo-evaluator/harness.js

@@ -0,0 +1,265 @@
+/**
+ * jinn-repo Evaluator Harness — wraps the {@link JinnRepoEvaluator} grading
+ * library as a first-class {@link Harness} so the daemon dispatches
+ * `jinn-repo.v1` evaluation tasks to it.
+ *
+ * Mirrors {@link SweRebenchV2EvaluatorHarness} but needs no Docker: grading is
+ * a repo-native git clone + `yarn install` + scoped test run (see
+ * `./eval-runner.ts`). Readiness is therefore just "not a stub" — git + the
+ * local pool are assumed present in any real daemon checkout.
+ *
+ * Two grading paths, keyed to the Task's `source` (issue #1891):
+ *   - merged-pr: gold FAIL_TO_PASS grading against the PR's own test files
+ *     (pool lookup, `./eval-runner.ts` — unchanged by #1891).
+ *   - live-issue: no gold tests exist prospectively, so grading is Stage 1's
+ *     thin mechanical evaluator (patch applies → typecheck → policy-scoped
+ *     tests, `./live-eval-runner.ts`) reading the spec straight off the task
+ *     — no pool lookup, since there is no gold to leak-protect.
+ *
+ * Spec: docs/superpowers/specs/2026-05-06-agent-harness-solvernet-design.md §3.4,
+ * spec/2026-07-20-autopilot-marketplace-execution.md §"No new SolverType".
+ */
+import { writeFile } from 'node:fs/promises';
+import { join } from 'node:path';
+import { JinnRepoSolutionPayloadSchema, } from '../../../vendor/@jinn-network/sdk/dist/solvernets/jinn-repo.js';
+import { JinnRepoTaskSchema, isLiveIssueTask, } from '../../../solver-types/jinn-repo.js';
+import { REQUIRES_LIVE_DAEMON_READINESS, SkippableError } from '../../types.js';
+import { SignedEnvelopeSchema, normalizeEnvelopeRole } from '../../../types/envelope.js';
+import { loadJinnRepoPool, } from '../../../solver-types/_jinn-repo-pool.js';
+import { JinnRepoEvaluator } from './evaluator.js';
+import { runJinnRepoLiveEval } from './live-eval-runner.js';
+const DEFAULT_MONO_REPO_URL = 'https://github.com/Jinn-Network/mono.git';
+/**
+ * The raw `source` field straight off an unparsed task spec — issue #1891
+ * Finding 2's primary discriminator. Deliberately does NOT go through
+ * `JinnRepoTaskSchema`: a full-schema parse can fail for two entirely
+ * different reasons that must be routed differently —
+ *   1. a merged-pr evaluation task's spec is the leak-controlled
+ *      solverView() projection (see `_jinn-repo-pool.ts`), which omits
+ *      required union fields and so never parses (raw `source` is absent, or
+ *      `'merged-pr'`) — the existing pool-lookup path, unchanged.
+ *   2. a live-issue spec with a field defect (missing `issue_number`, a
+ *      malformed `base_commit`, etc.) also fails the full parse, but its raw
+ *      `source` is unambiguously `'live-issue'` — that must be a loud,
+ *      specific error, never a silent fall-through to #1's pool-lookup path
+ *      (where it would surface a misleading `instance_not_in_pool` and be
+ *      re-attempted forever).
+ * Reading the raw field first makes the two cases distinguishable before any
+ * validation happens.
+ */
+function rawTaskSpecSource(spec) {
+    return spec?.['source'];
+}
+/**
+ * Short, single-line summary of a ZodError for error messages / rejection
+ * reasons. Structurally typed (not zod's ZodError) so it accepts errors from
+ * both zod majors — `JinnRepoTaskSchema` is authored in the SDK against
+ * zod/v3, while this package's own `zod` is v4.
+ */
+function summarizeZodError(error) {
+    return error.issues.map((i) => `${i.path.join('.') || '<root>'}: ${i.message}`).join('; ');
+}
+export class JinnRepoEvaluatorHarness {
+    name = 'jinn-repo-evaluator';
+    version = '1.0.0';
+    stub;
+    implStateDir;
+    loadPool;
+    grade;
+    gradeLive;
+    constructor(opts = {}) {
+        this.stub = opts.stub ?? false;
+        this.implStateDir = opts.implStateDir;
+        this.loadPool = opts.loadPool ?? (() => loadJinnRepoPool());
+        this.grade = opts.grade ?? ((args) => new JinnRepoEvaluator().grade(args));
+        this.gradeLive =
+            opts.gradeLive ??
+                ((args) => runJinnRepoLiveEval({
+                    spec: args.spec,
+                    patch: args.solution.patch,
+                    monoRepoUrl: process.env['JINN_MONO_REPO_URL'] ?? DEFAULT_MONO_REPO_URL,
+                }));
+    }
+    supports(ctx) {
+        return ctx.solverType === 'jinn-repo.v1' && ctx.role === 'evaluation';
+    }
+    async canAttempt(task) {
+        if (task.solverType !== 'jinn-repo.v1') {
+            return { ok: false, reason: 'solverType is not jinn-repo.v1' };
+        }
+        if (task.role !== 'evaluation') {
+            return { ok: false, reason: 'role is not evaluation' };
+        }
+        // A live-issue spec with a field defect (missing `issue_number`, a
+        // malformed `base_commit`, etc.) is rejected here with a specific reason
+        // — issue #1891 Finding 2 — rather than accepted blind and left to throw
+        // deep inside `run()` on every daemon re-attempt.
+        if (rawTaskSpecSource(task.spec) === 'live-issue') {
+            const parsedSpec = JinnRepoTaskSchema.safeParse(task.spec);
+            if (!parsedSpec.success) {
+                return {
+                    ok: false,
+                    reason: `malformed live-issue task spec: ${summarizeZodError(parsedSpec.error)}`,
+                };
+            }
+        }
+        // Both branches are gradeable (issue #1891): merged-pr against gold
+        // FAIL_TO_PASS tests, live-issue against the mechanical evaluator
+        // (applies/typecheck/tests). `run()` routes on the raw `source` field
+        // (same primary-discriminator check as above), never on parse-success
+        // alone.
+        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 };
+        // No Docker, no enable marker: the repo-native evaluator only needs git +
+        // the bundled pool, both present in any real daemon. Always ready when live.
+        return { ready: true };
+    }
+    async run(ctx) {
+        if (this.stub) {
+            throw new Error('jinn-repo-evaluator: stub registry cannot run evaluation (requires live daemon)');
+        }
+        // Read instance_id from the evaluation task spec (the leak-controlled
+        // solverView). The full pool item — including gold tests — is resolved
+        // locally, never sent over the wire.
+        const instanceId = ctx.task.spec?.['instance_id'];
+        if (typeof instanceId !== 'string' || instanceId.length === 0) {
+            throw new Error('jinn-repo-evaluator: task spec missing instance_id');
+        }
+        // Parse the solver's solution envelope and pull out the candidate patch.
+        // Mirrors the swe-rebench harness: context.restorationResult is the signed
+        // solution envelope JSON; assert solverType + solution role, then validate
+        // the typed payload.
+        const manifestJson = ctx.task.context['restorationResult'];
+        const envelope = SignedEnvelopeSchema.parse(JSON.parse(manifestJson));
+        if (envelope.solverType !== 'jinn-repo.v1' ||
+            normalizeEnvelopeRole(envelope.role) !== 'solution') {
+            throw new Error(`jinn-repo-evaluator: expected jinn-repo.v1/solution envelope, got ${envelope.solverType}/${envelope.role}`);
+        }
+        const solutionPayload = JinnRepoSolutionPayloadSchema.parse(envelope.payload);
+        // Route on the RAW `source` field, not full-parse success (issue #1891
+        // Finding 2). A merged-pr evaluation task's spec is the leak-controlled
+        // solverView() projection (see `_jinn-repo-pool.ts`), which has no raw
+        // `source` field at all (legacy) or `source: 'merged-pr'` — either way it
+        // falls through to the (unchanged) pool-lookup path below. A live-issue
+        // evaluation task's spec is the FULL JinnRepoLiveIssueTask (nothing to
+        // leak-protect, so no separate solverView()/pool-item split exists for
+        // it). Discriminating on the RAW field first — before the full schema
+        // parse — means a live-issue spec with a field defect (missing
+        // `issue_number`, a malformed `base_commit`, etc.) is recognized as a
+        // live-issue task that failed validation and throws a loud, specific
+        // error here, rather than silently falling through to the pool-lookup
+        // path below (where it would surface a misleading `instance_not_in_pool`
+        // SkippableError and be re-attempted forever).
+        const rawSource = rawTaskSpecSource(ctx.task.spec);
+        if (rawSource === 'live-issue') {
+            const parsedSpec = JinnRepoTaskSchema.safeParse(ctx.task.spec);
+            if (!parsedSpec.success || !isLiveIssueTask(parsedSpec.data)) {
+                const detail = parsedSpec.success
+                    ? 'parsed as a non-live-issue task despite source: live-issue (unexpected)'
+                    : summarizeZodError(parsedSpec.error);
+                throw new Error(`jinn-repo-evaluator: malformed live-issue task spec: ${detail}`);
+            }
+            return this.runLive(ctx, parsedSpec.data, instanceId, solutionPayload.patch);
+        }
+        // Resolve the full pool item (gold tests + base commit) for this instance.
+        // A missing instance must NOT silently pass — emit no verdict (SkippableError)
+        // so the orchestrator records a skip rather than a bogus FAIL.
+        const poolItem = this.loadPool().find((p) => p.instance_id === instanceId);
+        if (!poolItem) {
+            throw new SkippableError('instance_not_in_pool', `jinn-repo-evaluator: ${instanceId} is not present in the local pool`);
+        }
+        const result = await this.grade({ task: poolItem, solution: { patch: solutionPayload.patch } });
+        // An unscorable run (clone/install/spawn failure, or a patch that did not
+        // apply) carries no signal about the solver. Mirror swe-rebench's
+        // EvalCouldNotGradeError → SkippableError: emit no verdict so the engine
+        // records a skip. NEVER coerce unscorable → passed:false.
+        if (result.unscorable || result.passed === null) {
+            throw new SkippableError('eval_unscorable', `jinn-repo-evaluator: could not grade ${instanceId}${result.logExcerpt ? `\n${result.logExcerpt}` : ''}`);
+        }
+        const verdictPayload = {
+            schemaVersion: 'jinn-repo-verdict.v1',
+            passed: result.passed === true,
+            test_log_excerpt: result.logExcerpt,
+        };
+        await writeFile(join(ctx.workingDir, 'jinn-repo-verdict.json'), `${JSON.stringify(verdictPayload, null, 2)}\n`, 'utf8');
+        // Derive the engine-facing `verdict` from `passed`. The engine's
+        // reputation-feedback hook (and `verdictCodeForTask`) keys on
+        // `gating.verdict` — mirror the swe-rebench harness (jinn-mono-uy6v.10).
+        const verdict = verdictPayload.passed ? 'PASS' : 'FAIL';
+        return {
+            venueRef: { name: 'jinn-repo' },
+            gating: {
+                passed: verdictPayload.passed,
+                verdict,
+            },
+            informational: {
+                instance_id: instanceId,
+            },
+            verdictPayload: verdictPayload,
+            artifacts: [
+                {
+                    path: 'jinn-repo-verdict.json',
+                    artifactType: 'jinn_repo_verdict',
+                    metadata: {
+                        passed: verdictPayload.passed,
+                    },
+                    access: { priceUsdc: '0' },
+                },
+            ],
+        };
+    }
+    /**
+     * Live-issue grading path (issue #1891): no gold, so no pool lookup — the
+     * mechanical evaluator (`./live-eval-runner.ts`) grades three AND-gated
+     * checks (applies/typecheck/tests) straight off the task's own spec.
+     */
+    async runLive(ctx, spec, instanceId, patch) {
+        const result = await this.gradeLive({ spec, solution: { patch } });
+        // Same unscorable convention as the merged-pr path: infra failure
+        // (clone/install/spawn) carries no signal about the solver — emit no
+        // verdict so the engine records a skip. NEVER coerce unscorable → FAIL.
+        if (result.unscorable) {
+            throw new SkippableError('eval_unscorable', `jinn-repo-evaluator: could not grade ${instanceId}${result.logExcerpt ? `\n${result.logExcerpt}` : ''}`);
+        }
+        const verdictPayload = {
+            schemaVersion: 'jinn-repo-verdict.v2',
+            passed: result.passed,
+            test_log_excerpt: result.logExcerpt,
+            gates: {
+                applies: result.applies,
+                typecheck: result.typecheck,
+                tests: result.tests,
+            },
+        };
+        await writeFile(join(ctx.workingDir, 'jinn-repo-verdict.json'), `${JSON.stringify(verdictPayload, null, 2)}\n`, 'utf8');
+        const verdict = verdictPayload.passed ? 'PASS' : 'FAIL';
+        return {
+            venueRef: { name: 'jinn-repo' },
+            gating: {
+                passed: verdictPayload.passed,
+                verdict,
+            },
+            informational: {
+                instance_id: instanceId,
+            },
+            verdictPayload: verdictPayload,
+            artifacts: [
+                {
+                    path: 'jinn-repo-verdict.json',
+                    artifactType: 'jinn_repo_verdict',
+                    metadata: {
+                        passed: verdictPayload.passed,
+                    },
+                    access: { priceUsdc: '0' },
+                },
+            ],
+        };
+    }
+}
+//# sourceMappingURL=harness.js.map

package/dist/harnesses/impls/jinn-repo-evaluator/harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/jinn-repo-evaluator/harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC7C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EACL,6BAA6B,GAE9B,MAAM,wCAAwC,CAAC;AAChD,OAAO,EACL,kBAAkB,EAClB,eAAe,GAEhB,MAAM,oCAAoC,CAAC;AAE5C,OAAO,EAAE,8BAA8B,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAEhF,OAAO,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACzF,OAAO,EACL,gBAAgB,GAEjB,MAAM,0CAA0C,CAAC;AAClD,OAAO,EAAE,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AAEnD,OAAO,EAAE,mBAAmB,EAA+B,MAAM,uBAAuB,CAAC;AAmBzF,MAAM,qBAAqB,GAAG,0CAA0C,CAAC;AAEzE;;;;;;;;;;;;;;;;;GAiBG;AACH,SAAS,iBAAiB,CAAC,IAAyC;IAClE,OAAO,IAAI,EAAE,CAAC,QAAQ,CAAC,CAAC;AAC1B,CAAC;AAED;;;;;GAKG;AACH,SAAS,iBAAiB,CAAC,KAE1B;IACC,OAAO,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,QAAQ,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7F,CAAC;AA8BD,MAAM,OAAO,wBAAwB;IAC1B,IAAI,GAAG,qBAAqB,CAAC;IAC7B,OAAO,GAAG,OAAO,CAAC;IAEV,IAAI,CAAU;IACd,YAAY,CAAqB;IACjC,QAAQ,CAA2B;IACnC,KAAK,CAAU;IACf,SAAS,CAAc;IAExC,YAAY,OAAwC,EAAE;QACpD,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,KAAK,CAAC;QAC/B,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC;QACtC,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,CAAC,GAAG,EAAE,CAAC,gBAAgB,EAAE,CAAC,CAAC;QAC5D,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,iBAAiB,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QAC3E,IAAI,CAAC,SAAS;YACZ,IAAI,CAAC,SAAS;gBACd,CAAC,CAAC,IAAI,EAAE,EAAE,CACR,mBAAmB,CAAC;oBAClB,IAAI,EAAE,IAAI,CAAC,IAAI;oBACf,KAAK,EAAE,IAAI,CAAC,QAAQ,CAAC,KAAK;oBAC1B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,IAAI,qBAAqB;iBACxE,CAAC,CAAC,CAAC;IACV,CAAC;IAED,QAAQ,CAAC,GAAgE;QACvE,OAAO,GAAG,CAAC,UAAU,KAAK,cAAc,IAAI,GAAG,CAAC,IAAI,KAAK,YAAY,CAAC;IACxE,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,IAAU;QACzB,IAAI,IAAI,CAAC,UAAU,KAAK,cAAc,EAAE,CAAC;YACvC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,gCAAgC,EAAE,CAAC;QACjE,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,mEAAmE;QACnE,yEAAyE;QACzE,yEAAyE;QACzE,kDAAkD;QAClD,IAAI,iBAAiB,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,YAAY,EAAE,CAAC;YAClD,MAAM,UAAU,GAAG,kBAAkB,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC3D,IAAI,CAAC,UAAU,CAAC,OAAO,EAAE,CAAC;gBACxB,OAAO;oBACL,EAAE,EAAE,KAAK;oBACT,MAAM,EAAE,mCAAmC,iBAAiB,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE;iBACjF,CAAC;YACJ,CAAC;QACH,CAAC;QACD,oEAAoE;QACpE,kEAAkE;QAClE,sEAAsE;QACtE,sEAAsE;QACtE,SAAS;QACT,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,0EAA0E;QAC1E,6EAA6E;QAC7E,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IACzB,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,GAAmB;QAC3B,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CACb,iFAAiF,CAClF,CAAC;QACJ,CAAC;QAED,sEAAsE;QACtE,uEAAuE;QACvE,qCAAqC;QACrC,MAAM,UAAU,GAAI,GAAG,CAAC,IAAI,CAAC,IAA4C,EAAE,CAAC,aAAa,CAAC,CAAC;QAC3F,IAAI,OAAO,UAAU,KAAK,QAAQ,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC9D,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;QACxE,CAAC;QAED,yEAAyE;QACzE,2EAA2E;QAC3E,2EAA2E;QAC3E,qBAAqB;QACrB,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,IACE,QAAQ,CAAC,UAAU,KAAK,cAAc;YACtC,qBAAqB,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,UAAU,EACnD,CAAC;YACD,MAAM,IAAI,KAAK,CACb,qEAAqE,QAAQ,CAAC,UAAU,IAAI,QAAQ,CAAC,IAAI,EAAE,CAC5G,CAAC;QACJ,CAAC;QACD,MAAM,eAAe,GAAG,6BAA6B,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAE9E,uEAAuE;QACvE,wEAAwE;QACxE,uEAAuE;QACvE,0EAA0E;QAC1E,wEAAwE;QACxE,uEAAuE;QACvE,uEAAuE;QACvE,sEAAsE;QACtE,+DAA+D;QAC/D,sEAAsE;QACtE,qEAAqE;QACrE,sEAAsE;QACtE,yEAAyE;QACzE,+CAA+C;QAC/C,MAAM,SAAS,GAAG,iBAAiB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnD,IAAI,SAAS,KAAK,YAAY,EAAE,CAAC;YAC/B,MAAM,UAAU,GAAG,kBAAkB,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC/D,IAAI,CAAC,UAAU,CAAC,OAAO,IAAI,CAAC,eAAe,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7D,MAAM,MAAM,GAAG,UAAU,CAAC,OAAO;oBAC/B,CAAC,CAAC,yEAAyE;oBAC3E,CAAC,CAAC,iBAAiB,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;gBACxC,MAAM,IAAI,KAAK,CAAC,wDAAwD,MAAM,EAAE,CAAC,CAAC;YACpF,CAAC;YACD,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,UAAU,CAAC,IAAI,EAAE,UAAU,EAAE,eAAe,CAAC,KAAK,CAAC,CAAC;QAC/E,CAAC;QAED,2EAA2E;QAC3E,+EAA+E;QAC/E,+DAA+D;QAC/D,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,UAAU,CAAC,CAAC;QAC3E,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,MAAM,IAAI,cAAc,CACtB,sBAAsB,EACtB,wBAAwB,UAAU,mCAAmC,CACtE,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,eAAe,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAEhG,0EAA0E;QAC1E,kEAAkE;QAClE,yEAAyE;QACzE,0DAA0D;QAC1D,IAAI,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC;YAChD,MAAM,IAAI,cAAc,CACtB,iBAAiB,EACjB,wCAAwC,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CACzG,CAAC;QACJ,CAAC;QAED,MAAM,cAAc,GAA2B;YAC7C,aAAa,EAAE,sBAAsB;YACrC,MAAM,EAAE,MAAM,CAAC,MAAM,KAAK,IAAI;YAC9B,gBAAgB,EAAE,MAAM,CAAC,UAAU;SACpC,CAAC;QACF,MAAM,SAAS,CACb,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,wBAAwB,CAAC,EAC9C,GAAG,IAAI,CAAC,SAAS,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAC9C,MAAM,CACP,CAAC;QAEF,iEAAiE;QACjE,8DAA8D;QAC9D,yEAAyE;QACzE,MAAM,OAAO,GAAoB,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAEzE,OAAO;YACL,QAAQ,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE;YAC/B,MAAM,EAAE;gBACN,MAAM,EAAE,cAAc,CAAC,MAAM;gBAC7B,OAAO;aACR;YACD,aAAa,EAAE;gBACb,WAAW,EAAE,UAAU;aACxB;YACD,cAAc,EAAE,cAAoD;YACpE,SAAS,EAAE;gBACT;oBACE,IAAI,EAAE,wBAAwB;oBAC9B,YAAY,EAAE,mBAAmB;oBACjC,QAAQ,EAAE;wBACR,MAAM,EAAE,cAAc,CAAC,MAAM;qBAC9B;oBACD,MAAM,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE;iBAC3B;aACF;SACF,CAAC;IACJ,CAAC;IAED;;;;OAIG;IACK,KAAK,CAAC,OAAO,CACnB,GAAmB,EACnB,IAA2B,EAC3B,UAAkB,EAClB,KAAa;QAEb,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QAEnE,kEAAkE;QAClE,qEAAqE;QACrE,wEAAwE;QACxE,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YACtB,MAAM,IAAI,cAAc,CACtB,iBAAiB,EACjB,wCAAwC,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CACzG,CAAC;QACJ,CAAC;QAED,MAAM,cAAc,GAA2B;YAC7C,aAAa,EAAE,sBAAsB;YACrC,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,gBAAgB,EAAE,MAAM,CAAC,UAAU;YACnC,KAAK,EAAE;gBACL,OAAO,EAAE,MAAM,CAAC,OAAO;gBACvB,SAAS,EAAE,MAAM,CAAC,SAAS;gBAC3B,KAAK,EAAE,MAAM,CAAC,KAAK;aACpB;SACF,CAAC;QACF,MAAM,SAAS,CACb,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,wBAAwB,CAAC,EAC9C,GAAG,IAAI,CAAC,SAAS,CAAC,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAC9C,MAAM,CACP,CAAC;QAEF,MAAM,OAAO,GAAoB,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;QAEzE,OAAO;YACL,QAAQ,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE;YAC/B,MAAM,EAAE;gBACN,MAAM,EAAE,cAAc,CAAC,MAAM;gBAC7B,OAAO;aACR;YACD,aAAa,EAAE;gBACb,WAAW,EAAE,UAAU;aACxB;YACD,cAAc,EAAE,cAAoD;YACpE,SAAS,EAAE;gBACT;oBACE,IAAI,EAAE,wBAAwB;oBAC9B,YAAY,EAAE,mBAAmB;oBACjC,QAAQ,EAAE;wBACR,MAAM,EAAE,cAAc,CAAC,MAAM;qBAC9B;oBACD,MAAM,EAAE,EAAE,SAAS,EAAE,GAAG,EAAE;iBAC3B;aACF;SACF,CAAC;IACJ,CAAC;CACF"}

package/dist/harnesses/impls/jinn-repo-evaluator/live-eval-runner.d.ts

@@ -0,0 +1,68 @@
+/**
+ * Repo-native mechanical eval runner for jinn-repo LIVE-ISSUE tasks
+ * (issue #1891, spec/2026-07-20-autopilot-marketplace-execution.md
+ * §"No new SolverType").
+ *
+ * A live-issue task snapshots an OPEN GitHub issue — posted before a fix
+ * exists, so there is no gold FAIL_TO_PASS test to grade against (contrast
+ * `./eval-runner.ts`, which grades merged-pr tasks against the PR's own gold
+ * tests). Stage 1's evaluator for this branch is deliberately thin and
+ * mechanical: three AND-gated checks, in order:
+ *
+ *   1. applies    — does the candidate patch apply to a clean base_commit
+ *                    checkout? (via `./repro.ts`, same convention as the
+ *                    merged-pr path: a patch that does not apply is a graded
+ *                    FAIL, not an unscorable infra failure.)
+ *   2. typecheck  — does each TOUCHED package's FULL typecheck script pass?
+ *                    Always full (never scoped) — typechecking a subset of
+ *                    files is not meaningful. For `client`, `yarn typecheck`
+ *                    also builds `packages/sdk` (+ plugin/core) as a side
+ *                    effect (see `client/package.json`'s `build:sdk` chain),
+ *                    which is why this gate MUST run before the tests gate:
+ *                    a scoped `vitest run` on a fresh clone would otherwise
+ *                    fail to resolve `@jinn-network/sdk` (its `dist/` is
+ *                    gitignored and only exists after that build step).
+ *   3. tests      — for each touched package: if `scope-tests.ts` mirrors any
+ *                    changed file to an test path that actually exists in the
+ *                    repro, run those scoped (`vitest run <files>`); if none
+ *                    exist, fall back to the package's full test script.
+ *                    Multi-package patches AND-gate every touched package —
+ *                    each package's scoped-or-fallback tests must pass.
+ *
+ * Infra failure (clone / install / spawn) is NEVER coerced into a gate
+ * result — it is `unscorable`, mirroring `./eval-runner.ts` and the
+ * swe-rebench-v2 evaluator's `EvalCouldNotGradeError` convention. The engine
+ * records a skip (no verdict) rather than a bogus FAIL.
+ *
+ * A patch that changes files but touches NO gated package (e.g. only
+ * `contracts/`, `docs/`, or root config — anything outside `client/` and
+ * `packages/sdk/`) is likewise never coerced into a vacuous PASS: this
+ * evaluator has no applicable gate for it, which is a capability limit of the
+ * grader, not a verified solution — so it is `unscorable` too (see the
+ * `scopes.length === 0` check below). Contrast the true empty-patch case
+ * documented on `getChangedFiles`, where the empty changed-file list itself
+ * is the (by-design) vacuous case.
+ */
+import { type PackageSpec } from './scope-tests.js';
+export interface JinnRepoLiveEvalResult {
+    applies: boolean;
+    typecheck: boolean;
+    tests: boolean;
+    /** AND of the three gates above. */
+    passed: boolean;
+    /**
+     * True for grader-infra failure (clone/install/spawn) OR a patch that
+     * touches no gated package (no applicable gate exists) — see module doc.
+     * Never true alongside `passed: true`.
+     */
+    unscorable: boolean;
+    logExcerpt: string;
+}
+export declare function runJinnRepoLiveEval(args: {
+    spec: {
+        base_commit: string;
+    };
+    patch: string;
+    monoRepoUrl: string;
+    packages?: readonly PackageSpec[];
+}): Promise<JinnRepoLiveEvalResult>;

package/dist/harnesses/impls/jinn-repo-evaluator/live-eval-runner.js

@@ -0,0 +1,223 @@
+/**
+ * Repo-native mechanical eval runner for jinn-repo LIVE-ISSUE tasks
+ * (issue #1891, spec/2026-07-20-autopilot-marketplace-execution.md
+ * §"No new SolverType").
+ *
+ * A live-issue task snapshots an OPEN GitHub issue — posted before a fix
+ * exists, so there is no gold FAIL_TO_PASS test to grade against (contrast
+ * `./eval-runner.ts`, which grades merged-pr tasks against the PR's own gold
+ * tests). Stage 1's evaluator for this branch is deliberately thin and
+ * mechanical: three AND-gated checks, in order:
+ *
+ *   1. applies    — does the candidate patch apply to a clean base_commit
+ *                    checkout? (via `./repro.ts`, same convention as the
+ *                    merged-pr path: a patch that does not apply is a graded
+ *                    FAIL, not an unscorable infra failure.)
+ *   2. typecheck  — does each TOUCHED package's FULL typecheck script pass?
+ *                    Always full (never scoped) — typechecking a subset of
+ *                    files is not meaningful. For `client`, `yarn typecheck`
+ *                    also builds `packages/sdk` (+ plugin/core) as a side
+ *                    effect (see `client/package.json`'s `build:sdk` chain),
+ *                    which is why this gate MUST run before the tests gate:
+ *                    a scoped `vitest run` on a fresh clone would otherwise
+ *                    fail to resolve `@jinn-network/sdk` (its `dist/` is
+ *                    gitignored and only exists after that build step).
+ *   3. tests      — for each touched package: if `scope-tests.ts` mirrors any
+ *                    changed file to an test path that actually exists in the
+ *                    repro, run those scoped (`vitest run <files>`); if none
+ *                    exist, fall back to the package's full test script.
+ *                    Multi-package patches AND-gate every touched package —
+ *                    each package's scoped-or-fallback tests must pass.
+ *
+ * Infra failure (clone / install / spawn) is NEVER coerced into a gate
+ * result — it is `unscorable`, mirroring `./eval-runner.ts` and the
+ * swe-rebench-v2 evaluator's `EvalCouldNotGradeError` convention. The engine
+ * records a skip (no verdict) rather than a bogus FAIL.
+ *
+ * A patch that changes files but touches NO gated package (e.g. only
+ * `contracts/`, `docs/`, or root config — anything outside `client/` and
+ * `packages/sdk/`) is likewise never coerced into a vacuous PASS: this
+ * evaluator has no applicable gate for it, which is a capability limit of the
+ * grader, not a verified solution — so it is `unscorable` too (see the
+ * `scopes.length === 0` check below). Contrast the true empty-patch case
+ * documented on `getChangedFiles`, where the empty changed-file list itself
+ * is the (by-design) vacuous case.
+ */
+import { access } from 'node:fs/promises';
+import { join } from 'node:path';
+import { execFile } from 'node:child_process';
+import { promisify } from 'node:util';
+import { prepareRepro, PatchDoesNotApplyError } from './repro.js';
+import { scopeTestsForChangedFiles, KNOWN_LIVE_EVAL_PACKAGES, } from './scope-tests.js';
+const sh = promisify(execFile);
+const MAX_BUFFER = 64 * 1024 * 1024;
+const LOG_LIMIT = 4000;
+function unscorable(logExcerpt) {
+    return { applies: false, typecheck: false, tests: false, passed: false, unscorable: true, logExcerpt };
+}
+async function pathExists(path) {
+    try {
+        await access(path);
+        return true;
+    }
+    catch {
+        return false;
+    }
+}
+/**
+ * `prepareRepro` applies the candidate patch via `git apply --3way`, which
+ * stages cleanly-applied hunks — so a plain (unstaged) `git diff` sees
+ * nothing. `git diff --name-only HEAD` (working tree + index vs HEAD) is the
+ * form that reliably lists every changed file post-apply; verified against a
+ * scratch repo before relying on it here. An empty candidate patch (a valid
+ * no-op — `prepareRepro` skips `git apply` entirely for it) yields an empty
+ * changed-file list, so every gate below is vacuously satisfied for it — a
+ * known, documented limit of a mechanical (non-goal-aware) grader.
+ *
+ * Distinct from that: a NON-empty changed-file list that maps to zero gated
+ * packages (every changed file falls outside `client/` and `packages/sdk/`)
+ * must NOT also be vacuously satisfied — the caller checks
+ * `scopes.length === 0` against a non-empty `changedFiles` and treats it as
+ * `unscorable` (no applicable gate exists for this patch), never `passed`.
+ */
+async function getChangedFiles(repoDir) {
+    const { stdout } = await sh('git', ['-C', repoDir, 'diff', '--name-only', 'HEAD']);
+    return stdout
+        .split('\n')
+        .map((l) => l.trim())
+        .filter((l) => l.length > 0);
+}
+async function installDeps(pkgDir) {
+    // corepack is usually already enabled at the user / container level. A
+    // failure here (e.g. read-only global shim dir) must NOT block the eval —
+    // yarn install is the real gate. Swallow any corepack error (mirrors
+    // eval-runner.ts).
+    try {
+        await sh('corepack', ['enable'], { cwd: pkgDir });
+    }
+    catch {
+        // non-fatal
+    }
+    await sh('yarn', ['install', '--immutable'], { cwd: pkgDir, maxBuffer: MAX_BUFFER });
+}
+/**
+ * A *numeric* `code` on a spawn error means the command ran and exited
+ * non-zero — a real gate FAIL. A non-numeric `code` (e.g. ENOENT) is a
+ * spawn-level infra failure. Mirrors `eval-runner.ts`'s classification.
+ */
+function isRealFailure(e) {
+    return typeof e?.code === 'number';
+}
+function logOf(e) {
+    const out = e;
+    return `${out?.stdout ?? ''}\n${out?.stderr ?? ''}`.slice(0, LOG_LIMIT);
+}
+export async function runJinnRepoLiveEval(args) {
+    let repro;
+    try {
+        repro = await prepareRepro({
+            monoRepoUrl: args.monoRepoUrl,
+            baseCommit: args.spec.base_commit,
+            patch: args.patch,
+            goldTestFiles: {}, // no gold for a live-issue task — no-ops in prepareRepro
+        });
+    }
+    catch (e) {
+        if (e instanceof PatchDoesNotApplyError) {
+            return {
+                applies: false,
+                typecheck: false,
+                tests: false,
+                passed: false,
+                unscorable: false,
+                logExcerpt: String(e).slice(0, LOG_LIMIT),
+            };
+        }
+        return unscorable(`repro-prep-failed: ${String(e).slice(0, LOG_LIMIT)}`);
+    }
+    try {
+        const changedFiles = await getChangedFiles(repro.dir);
+        const scopes = scopeTestsForChangedFiles(changedFiles, args.packages ?? KNOWN_LIVE_EVAL_PACKAGES);
+        // A non-empty patch whose changed files all fall outside every gated
+        // package (e.g. only `contracts/`, `docs/`, or root config) has no
+        // applicable gate here — that is a capability limit of this mechanical
+        // grader, not a verified solution. NEVER return a vacuous PASS for it;
+        // mirror the infra-failure convention (unscorable, no verdict emitted).
+        // Contrast the true empty-patch case (changedFiles.length === 0), which
+        // stays vacuously satisfied by design — see getChangedFiles' doc above.
+        if (scopes.length === 0 && changedFiles.length > 0) {
+            return unscorable(`no-gated-package-touched: changed files outside every gated package (${(args.packages ?? KNOWN_LIVE_EVAL_PACKAGES).map((p) => p.root).join(', ')}): ${changedFiles.join(', ')}`.slice(0, LOG_LIMIT));
+        }
+        // ── Gate 2: typecheck — full, per touched package, AND-gated ───────────
+        let typecheckFailLog = '';
+        for (const scope of scopes) {
+            const pkgDir = join(repro.dir, scope.pkg.root);
+            try {
+                await installDeps(pkgDir);
+            }
+            catch (e) {
+                return unscorable(`install-failed[${scope.pkg.root}]: ${String(e).slice(0, LOG_LIMIT)}`);
+            }
+            try {
+                await sh('yarn', [scope.pkg.typecheckScript], { cwd: pkgDir, maxBuffer: MAX_BUFFER });
+            }
+            catch (e) {
+                if (!isRealFailure(e)) {
+                    return unscorable(`typecheck-spawn-failed[${scope.pkg.root}]: ${String(e).slice(0, LOG_LIMIT)}`);
+                }
+                typecheckFailLog += `typecheck-failed[${scope.pkg.root}]:\n${logOf(e)}\n`;
+            }
+        }
+        if (typecheckFailLog.length > 0) {
+            return {
+                applies: true,
+                typecheck: false,
+                tests: false,
+                passed: false,
+                unscorable: false,
+                logExcerpt: typecheckFailLog.slice(0, LOG_LIMIT),
+            };
+        }
+        // ── Gate 3: tests — scoped-if-any-exist, else full fallback, per
+        //    touched package, AND-gated ─────────────────────────────────────
+        let testsFailLog = '';
+        for (const scope of scopes) {
+            const pkgDir = join(repro.dir, scope.pkg.root);
+            const existingScoped = [];
+            for (const f of scope.candidateTestFiles) {
+                if (await pathExists(join(repro.dir, f)))
+                    existingScoped.push(f);
+            }
+            try {
+                if (existingScoped.length > 0) {
+                    // Candidate paths are repo-root-relative; strip the package prefix
+                    // so vitest resolves them from cwd=pkgDir.
+                    const relArgs = existingScoped.map((f) => f.slice(scope.pkg.root.length + 1));
+                    await sh('yarn', ['vitest', 'run', ...relArgs], { cwd: pkgDir, maxBuffer: MAX_BUFFER });
+                }
+                else {
+                    await sh('yarn', [scope.pkg.testScript], { cwd: pkgDir, maxBuffer: MAX_BUFFER });
+                }
+            }
+            catch (e) {
+                if (!isRealFailure(e)) {
+                    return unscorable(`test-spawn-failed[${scope.pkg.root}]: ${String(e).slice(0, LOG_LIMIT)}`);
+                }
+                testsFailLog += `tests-failed[${scope.pkg.root}]:\n${logOf(e)}\n`;
+            }
+        }
+        const testsOk = testsFailLog.length === 0;
+        return {
+            applies: true,
+            typecheck: true,
+            tests: testsOk,
+            passed: testsOk,
+            unscorable: false,
+            logExcerpt: testsOk ? '' : testsFailLog.slice(0, LOG_LIMIT),
+        };
+    }
+    finally {
+        await repro.cleanup();
+    }
+}
+//# sourceMappingURL=live-eval-runner.js.map

package/dist/harnesses/impls/jinn-repo-evaluator/live-eval-runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"live-eval-runner.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/jinn-repo-evaluator/live-eval-runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4CG;AAEH,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC1C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAE,YAAY,EAAE,sBAAsB,EAAE,MAAM,YAAY,CAAC;AAClE,OAAO,EACL,yBAAyB,EACzB,wBAAwB,GAEzB,MAAM,kBAAkB,CAAC;AAE1B,MAAM,EAAE,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;AAE/B,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AACpC,MAAM,SAAS,GAAG,IAAI,CAAC;AAiBvB,SAAS,UAAU,CAAC,UAAkB;IACpC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;AACzG,CAAC;AAED,KAAK,UAAU,UAAU,CAAC,IAAY;IACpC,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;QACnB,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,KAAK,UAAU,eAAe,CAAC,OAAe;IAC5C,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;IACnF,OAAO,MAAM;SACV,KAAK,CAAC,IAAI,CAAC;SACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;SACpB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;AACjC,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,MAAc;IACvC,uEAAuE;IACvE,0EAA0E;IAC1E,qEAAqE;IACrE,mBAAmB;IACnB,IAAI,CAAC;QACH,MAAM,EAAE,CAAC,UAAU,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,CAAC,CAAC;IACpD,CAAC;IAAC,MAAM,CAAC;QACP,YAAY;IACd,CAAC;IACD,MAAM,EAAE,CAAC,MAAM,EAAE,CAAC,SAAS,EAAE,aAAa,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,CAAC,CAAC;AACvF,CAAC;AAED;;;;GAIG;AACH,SAAS,aAAa,CAAC,CAAU;IAC/B,OAAO,OAAQ,CAAoC,EAAE,IAAI,KAAK,QAAQ,CAAC;AACzE,CAAC;AAED,SAAS,KAAK,CAAC,CAAU;IACvB,MAAM,GAAG,GAAG,CAAqD,CAAC;IAClE,OAAO,GAAG,GAAG,EAAE,MAAM,IAAI,EAAE,KAAK,GAAG,EAAE,MAAM,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;AAC1E,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,IAKzC;IACC,IAAI,KAAK,CAAC;IACV,IAAI,CAAC;QACH,KAAK,GAAG,MAAM,YAAY,CAAC;YACzB,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW;YACjC,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,aAAa,EAAE,EAAE,EAAE,yDAAyD;SAC7E,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,IAAI,CAAC,YAAY,sBAAsB,EAAE,CAAC;YACxC,OAAO;gBACL,OAAO,EAAE,KAAK;gBACd,SAAS,EAAE,KAAK;gBAChB,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,KAAK;gBACb,UAAU,EAAE,KAAK;gBACjB,UAAU,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC;aAC1C,CAAC;QACJ,CAAC;QACD,OAAO,UAAU,CAAC,sBAAsB,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,EAAE,CAAC,CAAC;IAC3E,CAAC;IAED,IAAI,CAAC;QACH,MAAM,YAAY,GAAG,MAAM,eAAe,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACtD,MAAM,MAAM,GAAG,yBAAyB,CAAC,YAAY,EAAE,IAAI,CAAC,QAAQ,IAAI,wBAAwB,CAAC,CAAC;QAElG,qEAAqE;QACrE,mEAAmE;QACnE,uEAAuE;QACvE,uEAAuE;QACvE,wEAAwE;QACxE,wEAAwE;QACxE,wEAAwE;QACxE,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACnD,OAAO,UAAU,CACf,wEAAwE,CAAC,IAAI,CAAC,QAAQ,IAAI,wBAAwB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CACpL,CAAC,EACD,SAAS,CACV,CACF,CAAC;QACJ,CAAC;QAED,0EAA0E;QAC1E,IAAI,gBAAgB,GAAG,EAAE,CAAC;QAC1B,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YAC/C,IAAI,CAAC;gBACH,MAAM,WAAW,CAAC,MAAM,CAAC,CAAC;YAC5B,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,OAAO,UAAU,CAAC,kBAAkB,KAAK,CAAC,GAAG,CAAC,IAAI,MAAM,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,EAAE,CAAC,CAAC;YAC3F,CAAC;YACD,IAAI,CAAC;gBACH,MAAM,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,eAAe,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,CAAC,CAAC;YACxF,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC;oBACtB,OAAO,UAAU,CAAC,0BAA0B,KAAK,CAAC,GAAG,CAAC,IAAI,MAAM,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,EAAE,CAAC,CAAC;gBACnG,CAAC;gBACD,gBAAgB,IAAI,oBAAoB,KAAK,CAAC,GAAG,CAAC,IAAI,OAAO,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;YAC5E,CAAC;QACH,CAAC;QACD,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,SAAS,EAAE,KAAK;gBAChB,KAAK,EAAE,KAAK;gBACZ,MAAM,EAAE,KAAK;gBACb,UAAU,EAAE,KAAK;gBACjB,UAAU,EAAE,gBAAgB,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC;aACjD,CAAC;QACJ,CAAC;QAED,kEAAkE;QAClE,sEAAsE;QACtE,IAAI,YAAY,GAAG,EAAE,CAAC;QACtB,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YAC/C,MAAM,cAAc,GAAa,EAAE,CAAC;YACpC,KAAK,MAAM,CAAC,IAAI,KAAK,CAAC,kBAAkB,EAAE,CAAC;gBACzC,IAAI,MAAM,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;oBAAE,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACnE,CAAC;YACD,IAAI,CAAC;gBACH,IAAI,cAAc,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;oBAC9B,mEAAmE;oBACnE,2CAA2C;oBAC3C,MAAM,OAAO,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC;oBAC9E,MAAM,EAAE,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,KAAK,EAAE,GAAG,OAAO,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,CAAC,CAAC;gBAC1F,CAAC;qBAAM,CAAC;oBACN,MAAM,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,CAAC,CAAC;gBACnF,CAAC;YACH,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC;oBACtB,OAAO,UAAU,CAAC,qBAAqB,KAAK,CAAC,GAAG,CAAC,IAAI,MAAM,MAAM,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,EAAE,CAAC,CAAC;gBAC9F,CAAC;gBACD,YAAY,IAAI,gBAAgB,KAAK,CAAC,GAAG,CAAC,IAAI,OAAO,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;YACpE,CAAC;QACH,CAAC;QAED,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,KAAK,CAAC,CAAC;QAC1C,OAAO;YACL,OAAO,EAAE,IAAI;YACb,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,OAAO;YACd,MAAM,EAAE,OAAO;YACf,UAAU,EAAE,KAAK;YACjB,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC;SAC5D,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,MAAM,KAAK,CAAC,OAAO,EAAE,CAAC;IACxB,CAAC;AACH,CAAC"}

package/dist/harnesses/impls/jinn-repo-evaluator/repro.d.ts

@@ -0,0 +1,32 @@
+/**
+ * Repo-native repro for the jinn-repo SolverType.
+ *
+ * Clones Jinn-Network/mono at a task's `base_commit` into a throwaway dir,
+ * applies a candidate patch, and overwrites the PR's gold test files. The
+ * caller (eval-runner) then installs deps and runs the targeted test to
+ * grade the patch (SWE-bench FAIL_TO_PASS pattern).
+ *
+ * Deliberately does NOT install or run anything — that is the eval-runner's
+ * job, so it can classify install failure as `unscorable` separately from a
+ * test failure (FAIL).
+ */
+export interface PreparedRepro {
+    dir: string;
+    cleanup: () => Promise<void>;
+}
+/**
+ * Thrown when the *candidate* patch fails to apply to a clean base_commit
+ * checkout. This is a property of the candidate (a bad / conflicting patch),
+ * NOT a grader-infra failure — so the eval-runner scores it FAIL, not
+ * unscorable. (Reserve `unscorable` for clone / fetch / install / test-spawn
+ * failures, which are the grader's environment breaking.)
+ */
+export declare class PatchDoesNotApplyError extends Error {
+    constructor(message: string);
+}
+export declare function prepareRepro(args: {
+    monoRepoUrl: string;
+    baseCommit: string;
+    patch: string;
+    goldTestFiles: Record<string, string>;
+}): Promise<PreparedRepro>;