@jinn-network/client 0.1.9 → 0.2.0

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>;

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

@@ -0,0 +1,73 @@
+/**
+ * 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).
+ */
+import { mkdtemp, mkdir, rm, writeFile } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { dirname, join } from 'node:path';
+import { execFile } from 'node:child_process';
+import { promisify } from 'node:util';
+const sh = promisify(execFile);
+/**
+ * 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 class PatchDoesNotApplyError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = 'PatchDoesNotApplyError';
+    }
+}
+export async function prepareRepro(args) {
+    const dir = await mkdtemp(join(tmpdir(), 'jinn-repo-eval-'));
+    const cleanup = () => rm(dir, { recursive: true, force: true });
+    try {
+        // Shallow single-commit fetch keeps the clone tractable even though the
+        // mono history is large. Works against both https and file:// remotes.
+        await sh('git', ['init', '-q', dir]);
+        await sh('git', ['-C', dir, 'remote', 'add', 'origin', args.monoRepoUrl]);
+        await sh('git', ['-C', dir, 'fetch', '-q', '--depth', '1', 'origin', args.baseCommit]);
+        await sh('git', ['-C', dir, 'checkout', '-q', args.baseCommit]);
+        // Apply the candidate patch (skip when empty — an empty patch is a valid
+        // "no-op" candidate that must still be graded, and `git apply` errors on
+        // empty input).
+        if (args.patch.trim().length > 0) {
+            const patchPath = join(dir, '.candidate.patch');
+            await writeFile(patchPath, args.patch);
+            try {
+                await sh('git', ['-C', dir, 'apply', '--3way', patchPath]);
+            }
+            catch (e) {
+                // A candidate patch that does not apply is a FAIL, not infra breakage.
+                // Surface a typed error so the eval-runner classifies it as a graded
+                // FAIL rather than silently skipping it (unscorable → no verdict).
+                const detail = String(e?.stderr ?? e).slice(0, 2000);
+                throw new PatchDoesNotApplyError(`candidate patch does not apply: ${detail}`);
+            }
+        }
+        // Overwrite the gold tests last so a candidate patch can never weaken
+        // them. relpath is from the repo root (e.g. client/test/...).
+        for (const [rel, contents] of Object.entries(args.goldTestFiles)) {
+            const dest = join(dir, rel);
+            await mkdir(dirname(dest), { recursive: true });
+            await writeFile(dest, contents);
+        }
+        return { dir, cleanup };
+    }
+    catch (e) {
+        await cleanup();
+        throw e;
+    }
+}
+//# sourceMappingURL=repro.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"repro.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/jinn-repo-evaluator/repro.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACjE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAEtC,MAAM,EAAE,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;AAO/B;;;;;;GAMG;AACH,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAC/C,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,IAKlC;IACC,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC7D,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC,EAAE,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,IAAI,CAAC;QACH,wEAAwE;QACxE,uEAAuE;QACvE,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC;QACrC,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC;QAC1E,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,QAAQ,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QACvF,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,UAAU,EAAE,IAAI,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QAEhE,yEAAyE;QACzE,yEAAyE;QACzE,gBAAgB;QAChB,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,kBAAkB,CAAC,CAAC;YAChD,MAAM,SAAS,CAAC,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC;YACvC,IAAI,CAAC;gBACH,MAAM,EAAE,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC;YAC7D,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,uEAAuE;gBACvE,qEAAqE;gBACrE,mEAAmE;gBACnE,MAAM,MAAM,GAAG,MAAM,CAAE,CAAyB,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC;gBAC9E,MAAM,IAAI,sBAAsB,CAAC,mCAAmC,MAAM,EAAE,CAAC,CAAC;YAChF,CAAC;QACH,CAAC;QAED,sEAAsE;QACtE,8DAA8D;QAC9D,KAAK,MAAM,CAAC,GAAG,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC,EAAE,CAAC;YACjE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC;YAC5B,MAAM,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAChD,MAAM,SAAS,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;QAClC,CAAC;QAED,OAAO,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC;IAC1B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,OAAO,EAAE,CAAC;QAChB,MAAM,CAAC,CAAC;IACV,CAAC;AACH,CAAC"}

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

@@ -0,0 +1,62 @@
+/**
+ * Pure changed-file → mirrored-test-path mapping for the jinn-repo live-issue
+ * mechanical evaluator (issue #1891).
+ *
+ * No I/O here — this module only computes *candidate* test paths from a list
+ * of repo-root-relative changed file paths. It does NOT check whether those
+ * paths actually exist on disk; the caller (`live-eval-runner.ts`, which does
+ * do I/O against the throwaway repro clone) filters candidates by existence
+ * and falls back to a package's full test script when none exist. That
+ * two-step split keeps this module trivially unit-testable while the runner
+ * owns the filesystem truth.
+ *
+ * Mirror convention (docs/runbooks/testing.md): `<pkg>/src/foo/bar.ts` →
+ * `<pkg>/test/foo/bar.test.ts`. A file that is already a test file (matches
+ * `*.test.ts(x)`) — including co-located tests that live under `src/` (e.g.
+ * the dashboard SPA, which does not follow the src↔test split) — is used
+ * as-is rather than re-mirrored, so it is never double-suffixed.
+ *
+ * A changed file outside every known package's root contributes nothing (no
+ * test framework is wired up for it here); a changed file inside a package's
+ * root but outside both its `src/` and `test/` dirs (e.g. `package.json`)
+ * still marks that package as touched, with zero candidate test files — the
+ * runner reads that as "fall back to this package's full test script".
+ */
+export interface PackageSpec {
+    /** Repo-root-relative package directory, no trailing slash (e.g. `client`). */
+    root: string;
+    /** Repo-root-relative source directory (e.g. `client/src`). */
+    srcDir: string;
+    /** Repo-root-relative test directory (e.g. `client/test`). */
+    testDir: string;
+    /** `yarn <script>` name run for the FULL typecheck gate (never scoped). */
+    typecheckScript: string;
+    /** `yarn <script>` name run as the full-suite fallback when no test files are scoped. */
+    testScript: string;
+}
+export interface PackageScope {
+    pkg: PackageSpec;
+    /**
+     * Repo-root-relative candidate mirrored test file paths for this package.
+     * May contain paths that do not exist on disk — the runner is responsible
+     * for filtering. An empty array (with the package still present in the
+     * returned list) means "this package was touched but no test file could be
+     * derived — fall back to its full test script".
+     */
+    candidateTestFiles: string[];
+}
+/** The two TypeScript packages a jinn-repo live-issue patch can realistically
+ *  touch and that this evaluator knows how to gate: `client` (the daemon —
+ *  every jinn-repo pool instance mined so far is client/test-scoped, see
+ *  `jinn-repo-extract.ts`'s `TEST_PATH` filter) and `packages/sdk` (the
+ *  typed SolverNet contract layer `client` portals to). `packages/sdk` is
+ *  ordered first: it is a dependency-free leaf package, so grading it before
+ *  the heavier `client` gate fails fast on a broken contract surface. */
+export declare const SDK_PACKAGE: PackageSpec;
+export declare const CLIENT_PACKAGE: PackageSpec;
+export declare const KNOWN_LIVE_EVAL_PACKAGES: readonly PackageSpec[];
+/**
+ * Groups repo-root-relative changed file paths by owning package and maps
+ * each to a candidate test path per the mirror convention above.
+ */
+export declare function scopeTestsForChangedFiles(changedFiles: string[], packages?: readonly PackageSpec[]): PackageScope[];

package/dist/harnesses/impls/jinn-repo-evaluator/scope-tests.js

@@ -0,0 +1,91 @@
+/**
+ * Pure changed-file → mirrored-test-path mapping for the jinn-repo live-issue
+ * mechanical evaluator (issue #1891).
+ *
+ * No I/O here — this module only computes *candidate* test paths from a list
+ * of repo-root-relative changed file paths. It does NOT check whether those
+ * paths actually exist on disk; the caller (`live-eval-runner.ts`, which does
+ * do I/O against the throwaway repro clone) filters candidates by existence
+ * and falls back to a package's full test script when none exist. That
+ * two-step split keeps this module trivially unit-testable while the runner
+ * owns the filesystem truth.
+ *
+ * Mirror convention (docs/runbooks/testing.md): `<pkg>/src/foo/bar.ts` →
+ * `<pkg>/test/foo/bar.test.ts`. A file that is already a test file (matches
+ * `*.test.ts(x)`) — including co-located tests that live under `src/` (e.g.
+ * the dashboard SPA, which does not follow the src↔test split) — is used
+ * as-is rather than re-mirrored, so it is never double-suffixed.
+ *
+ * A changed file outside every known package's root contributes nothing (no
+ * test framework is wired up for it here); a changed file inside a package's
+ * root but outside both its `src/` and `test/` dirs (e.g. `package.json`)
+ * still marks that package as touched, with zero candidate test files — the
+ * runner reads that as "fall back to this package's full test script".
+ */
+/** The two TypeScript packages a jinn-repo live-issue patch can realistically
+ *  touch and that this evaluator knows how to gate: `client` (the daemon —
+ *  every jinn-repo pool instance mined so far is client/test-scoped, see
+ *  `jinn-repo-extract.ts`'s `TEST_PATH` filter) and `packages/sdk` (the
+ *  typed SolverNet contract layer `client` portals to). `packages/sdk` is
+ *  ordered first: it is a dependency-free leaf package, so grading it before
+ *  the heavier `client` gate fails fast on a broken contract surface. */
+export const SDK_PACKAGE = {
+    root: 'packages/sdk',
+    srcDir: 'packages/sdk/src',
+    testDir: 'packages/sdk/test',
+    typecheckScript: 'typecheck',
+    testScript: 'test',
+};
+export const CLIENT_PACKAGE = {
+    root: 'client',
+    srcDir: 'client/src',
+    testDir: 'client/test',
+    typecheckScript: 'typecheck',
+    testScript: 'test',
+};
+export const KNOWN_LIVE_EVAL_PACKAGES = [SDK_PACKAGE, CLIENT_PACKAGE];
+const TEST_FILE_RE = /\.test\.tsx?$/;
+const SOURCE_EXT_RE = /\.tsx?$/;
+function ownerPackage(file, packages) {
+    return packages.find((p) => file === p.root || file.startsWith(`${p.root}/`));
+}
+function mirrorSrcToTest(file, pkg) {
+    const relFromSrc = file.slice(pkg.srcDir.length + 1);
+    const dot = relFromSrc.lastIndexOf('.');
+    const ext = dot >= 0 ? relFromSrc.slice(dot) : '';
+    const withoutExt = dot >= 0 ? relFromSrc.slice(0, dot) : relFromSrc;
+    const testExt = ext === '.tsx' ? '.test.tsx' : '.test.ts';
+    return `${pkg.testDir}/${withoutExt}${testExt}`;
+}
+/**
+ * Groups repo-root-relative changed file paths by owning package and maps
+ * each to a candidate test path per the mirror convention above.
+ */
+export function scopeTestsForChangedFiles(changedFiles, packages = KNOWN_LIVE_EVAL_PACKAGES) {
+    const touched = new Map();
+    for (const file of changedFiles) {
+        const pkg = ownerPackage(file, packages);
+        if (!pkg)
+            continue; // outside every known package — nothing to gate here
+        if (!touched.has(pkg))
+            touched.set(pkg, new Set());
+        const scoped = touched.get(pkg);
+        if (TEST_FILE_RE.test(file)) {
+            // Already a test file wherever it lives (including co-located src/
+            // tests, e.g. the dashboard SPA) — run it directly, never re-mirror.
+            scoped.add(file);
+        }
+        else if (SOURCE_EXT_RE.test(file) && (file === pkg.srcDir || file.startsWith(`${pkg.srcDir}/`))) {
+            scoped.add(mirrorSrcToTest(file, pkg));
+        }
+        // Else: a package-root file outside src/test (config, package.json, a
+        // non-.ts/.tsx source file, …). No candidate test file, but the package
+        // stays "touched" via the Map entry created above, so an empty candidate
+        // list still triggers the fallback rather than being silently ignored.
+    }
+    return [...touched.entries()].map(([pkg, files]) => ({
+        pkg,
+        candidateTestFiles: [...files].sort(),
+    }));
+}
+//# sourceMappingURL=scope-tests.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"scope-tests.js","sourceRoot":"","sources":["../../../../src/harnesses/impls/jinn-repo-evaluator/scope-tests.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AA2BH;;;;;;yEAMyE;AACzE,MAAM,CAAC,MAAM,WAAW,GAAgB;IACtC,IAAI,EAAE,cAAc;IACpB,MAAM,EAAE,kBAAkB;IAC1B,OAAO,EAAE,mBAAmB;IAC5B,eAAe,EAAE,WAAW;IAC5B,UAAU,EAAE,MAAM;CACnB,CAAC;AAEF,MAAM,CAAC,MAAM,cAAc,GAAgB;IACzC,IAAI,EAAE,QAAQ;IACd,MAAM,EAAE,YAAY;IACpB,OAAO,EAAE,aAAa;IACtB,eAAe,EAAE,WAAW;IAC5B,UAAU,EAAE,MAAM;CACnB,CAAC;AAEF,MAAM,CAAC,MAAM,wBAAwB,GAA2B,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;AAE9F,MAAM,YAAY,GAAG,eAAe,CAAC;AACrC,MAAM,aAAa,GAAG,SAAS,CAAC;AAEhC,SAAS,YAAY,CAAC,IAAY,EAAE,QAAgC;IAClE,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,CAAC,CAAC,IAAI,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC;AAChF,CAAC;AAED,SAAS,eAAe,CAAC,IAAY,EAAE,GAAgB;IACrD,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACrD,MAAM,GAAG,GAAG,UAAU,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACxC,MAAM,GAAG,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAClD,MAAM,UAAU,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC;IACpE,MAAM,OAAO,GAAG,GAAG,KAAK,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,UAAU,CAAC;IAC1D,OAAO,GAAG,GAAG,CAAC,OAAO,IAAI,UAAU,GAAG,OAAO,EAAE,CAAC;AAClD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,yBAAyB,CACvC,YAAsB,EACtB,WAAmC,wBAAwB;IAE3D,MAAM,OAAO,GAAG,IAAI,GAAG,EAA4B,CAAC;IAEpD,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAChC,MAAM,GAAG,GAAG,YAAY,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;QACzC,IAAI,CAAC,GAAG;YAAE,SAAS,CAAC,qDAAqD;QAEzE,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC;YAAE,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,GAAG,EAAE,CAAC,CAAC;QACnD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC;QAEjC,IAAI,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YAC5B,mEAAmE;YACnE,qEAAqE;YACrE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;aAAM,IAAI,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,KAAK,GAAG,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC;YAClG,MAAM,CAAC,GAAG,CAAC,eAAe,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC;QACzC,CAAC;QACD,sEAAsE;QACtE,wEAAwE;QACxE,yEAAyE;QACzE,uEAAuE;IACzE,CAAC;IAED,OAAO,CAAC,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;QACnD,GAAG;QACH,kBAAkB,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,EAAE;KACtC,CAAC,CAAC,CAAC;AACN,CAAC"}

package/dist/harnesses/impls/learner/adapters/claude-code.js

@@ -299,7 +299,13 @@ export class ClaudeCodeHarnessAdapter {
             };
             let stderr = '';
             child.stdout?.on('data', (d) => {
-                stdoutLog.write(d);
+                // Guard against ERR_STREAM_WRITE_AFTER_END: once the session settles,
+                // closeLogs() ends these streams, but the reaped child (SIGTERM, not
+                // instant) can still emit bytes — e.g. after onResult() fires on a
+                // mid-session tool-result envelope. An unguarded write here throws
+                // "write after end", which surfaces as an engine tick failure.
+                if (!stdoutLog.writableEnded)
+                    stdoutLog.write(d);
                 if (settled)
                     return;
                 stdoutBuf += d.toString();
@@ -320,7 +326,8 @@ export class ClaudeCodeHarnessAdapter {
                 }
             });
             child.stderr?.on('data', (d) => {
-                stderrLog.write(d);
+                if (!stderrLog.writableEnded)
+                    stderrLog.write(d);
                 stderr += d.toString();
             });
             child.on('exit', (code, signal) => {