@brainbase-labs/cli 0.25.0-eng1209.2 → 0.25.0-eng1209.4

package/dist/index.js

@@ -36008,7 +36008,7 @@ function padStart(s, n) {
 // package.json
 var package_default = {
   name: "@brainbase-labs/cli",
-  version: "0.25.0-eng1209.2",
+  version: "0.25.0-eng1209.4",
   description: "Pack, share, and install agent templates across harnesses (Claude Code, Codex, ...).",
   type: "module",
   bin: {
@@ -78486,7 +78486,7 @@ var MAX_REMOTE_INPUT_BYTES = 2 * 1024 * 1024 * 1024;
 var MAX_ARCHIVE_EXTRACTED_BYTES = 2 * 1024 * 1024 * 1024;
 var MAX_ARCHIVE_SCAN_BYTES = 2 * 1024 * 1024 * 1024;
 var MAX_SANDBOX_COMMANDS = 20;
-var MAX_CRITERIA_PER_EVALUATOR = 100;
+var MAX_CRITERIA_PER_EVALUATOR = 200;
 var MAX_CRITERIA_RESULT_BYTES = 1024 * 1024;
 var MAX_CRITERION_EXPLANATION_LENGTH = 16384;
 var MAX_CRITERION_EVIDENCE_IDS = 100;
@@ -79114,10 +79114,15 @@ async function verifyRecordsUnchanged(records, spec) {
     const relative = safeRelPath(record3.path);
     assertNoSymlinkTraversal(root, relative);
     const candidate = path88.resolve(root, relative);
-    if (!isWithin(root, candidate) || !fs81.existsSync(candidate)) {
+    if (!isWithin(root, candidate)) {
+      throw new BenchmarkPhaseError("evidence_tampered", `evidence was removed during evaluation: ${record3.root}:${record3.path}`);
+    }
+    let stat;
+    try {
+      stat = fs81.lstatSync(candidate);
+    } catch {
       throw new BenchmarkPhaseError("evidence_tampered", `evidence was removed during evaluation: ${record3.root}:${record3.path}`);
     }
-    const stat = fs81.lstatSync(candidate);
     if (record3.kind === "symlink") {
       const target = stat.isSymbolicLink() ? fs81.readlinkSync(candidate) : null;
       if (target === null || Buffer.byteLength(target) !== record3.size || sha256(target) !== record3.sha256) {
@@ -79212,6 +79217,15 @@ function removeRemoteHydrationInputs(spec) {
     removedSources.add(material.source);
   }
 }
+function removePrivateHydrationSourcesBeforeSetup(spec, context) {
+  removeRemoteHydrationInputs(spec);
+  for (const temporary of context.temporaryRoots) {
+    fs81.rmSync(temporary, { recursive: true, force: true });
+  }
+  context.temporaryRoots.clear();
+  context.verifiedInputs.clear();
+  context.preparedArchiveFiles.clear();
+}
 async function downloadInputReference(stagingRoot, input, context) {
   const cacheKey = inputCacheKey(input);
   const cached2 = context.verifiedInputs.get(cacheKey);
@@ -80100,6 +80114,7 @@ async function executeHydrate(spec, context) {
       throw error2;
     }
   }
+  removePrivateHydrationSourcesBeforeSetup(spec, context);
   for (const command of spec.setup_commands) {
     assertBudget(context);
     let result2;
@@ -80155,7 +80170,6 @@ async function executeHydrate(spec, context) {
   }
   outputs.splice(0, outputs.length, ...finalOutputs);
   context.outputs.splice(0, context.outputs.length, ...finalOutputs);
-  removeRemoteHydrationInputs(spec);
   assertBudget(context);
   return outputs;
 }
@@ -80346,12 +80360,6 @@ async function copyCandidateOutput(output, manifest, spec, context) {
   return copied;
 }
 async function copyFrozenWorkspace(manifest, spec, context) {
-  for (const entry of manifest) {
-    assertBudget(context);
-    if (entry.kind === "symlink") {
-      throw new BenchmarkPhaseError("unsupported_workspace_symlink", `sandbox evaluator workspace cannot safely reproduce symlink: ${entry.path}`);
-    }
-  }
   const destinationRoot = fs81.mkdtempSync(path88.join(os16.tmpdir(), "brainbase-benchmark-workspace-"));
   fs81.chmodSync(destinationRoot, 448);
   context.temporaryRoots.add(destinationRoot);
@@ -80359,6 +80367,23 @@ async function copyFrozenWorkspace(manifest, spec, context) {
     assertBudget(context);
     const source = path88.resolve(spec.workspace_root, safeRelPath(frozenFile.path));
     const destination = path88.resolve(destinationRoot, safeRelPath(frozenFile.path));
+    if (frozenFile.kind === "symlink") {
+      const stat = fs81.lstatSync(source);
+      const target = stat.isSymbolicLink() ? fs81.readlinkSync(source) : null;
+      if (target === null || Buffer.byteLength(target) !== frozenFile.size || sha256(target) !== frozenFile.sha256) {
+        throw new BenchmarkPhaseError("evidence_tampered", `workspace changed after it was frozen: ${frozenFile.path}`);
+      }
+      if (path88.isAbsolute(target)) {
+        throw new BenchmarkPhaseError("unsafe_workspace_symlink", `sandbox evaluator workspace cannot reproduce an absolute symlink: ${frozenFile.path}`);
+      }
+      const resolvedTarget = path88.resolve(path88.dirname(source), target);
+      if (!isWithin(spec.workspace_root, resolvedTarget)) {
+        throw new BenchmarkPhaseError("unsafe_workspace_symlink", `sandbox evaluator workspace symlink escapes the workspace: ${frozenFile.path}`);
+      }
+      fs81.mkdirSync(path88.dirname(destination), { recursive: true, mode: 448 });
+      fs81.symlinkSync(target, destination);
+      continue;
+    }
     await atomicCopy(source, destination, frozenFile.mode, spec.workspace_root);
     const copied = await recordFile(destinationRoot, destination, "workspace");
     if (copied.sha256 !== frozenFile.sha256 || copied.size !== frozenFile.size || copied.mode !== frozenFile.mode) {
@@ -80510,7 +80535,21 @@ async function evaluateOne(evaluator, spec, manifest, finalOutput, trajectory, f
       }
       verdict = regexResult.exitCode === 0;
     }
-    return { ...base2, status: verdict ? "passed" : "failed", verdict, duration_ms: Date.now() - started };
+    return {
+      ...base2,
+      status: verdict ? "passed" : "failed",
+      verdict,
+      duration_ms: Date.now() - started,
+      details: {
+        operator: assertion.operator,
+        actual_size_bytes: Buffer.byteLength(finalOutput),
+        ...assertion.operator === "exact" ? {
+          expected_size_bytes: Buffer.byteLength(assertion.expected),
+          actual_sha256: sha256(finalOutput),
+          expected_sha256: sha256(assertion.expected)
+        } : {}
+      }
+    };
   }
   if (evaluator.type === "trajectory_assertion") {
     const count = trajectory.filter((event) => eventType(event) === evaluator.event_type).length;
@@ -80540,6 +80579,7 @@ async function evaluateOne(evaluator, spec, manifest, finalOutput, trajectory, f
     }
     const exists2 = stat !== null;
     let verdict = false;
+    let actualSha256;
     if (evaluator.assertion.operator === "exists")
       verdict = exists2;
     if (evaluator.assertion.operator === "not_exists")
@@ -80548,7 +80588,8 @@ async function evaluateOne(evaluator, spec, manifest, finalOutput, trajectory, f
       if (stat?.isFile()) {
         const opened = openRegularFileNoFollow(candidate, `workspace assertion ${relative}`, spec.workspace_root);
         try {
-          verdict = await sha256OfDescriptor(opened.fd) === evaluator.assertion.expected;
+          actualSha256 = await sha256OfDescriptor(opened.fd);
+          verdict = actualSha256 === evaluator.assertion.expected;
         } finally {
           fs81.closeSync(opened.fd);
         }
@@ -80564,7 +80605,22 @@ async function evaluateOne(evaluator, spec, manifest, finalOutput, trajectory, f
         }
       }
     }
-    return { ...base2, status: verdict ? "passed" : "failed", verdict, duration_ms: Date.now() - started };
+    return {
+      ...base2,
+      status: verdict ? "passed" : "failed",
+      verdict,
+      duration_ms: Date.now() - started,
+      details: {
+        operator: evaluator.assertion.operator,
+        path: relative,
+        exists: exists2,
+        ...stat ? {
+          actual_kind: stat.isFile() ? "file" : stat.isDirectory() ? "directory" : "other",
+          actual_size_bytes: stat.size
+        } : {},
+        ...actualSha256 ? { actual_sha256: actualSha256 } : {}
+      }
+    };
   }
   let isolatedWorkspace;
   let privateResultRoot;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@brainbase-labs/cli",
-  "version": "0.25.0-eng1209.2",
+  "version": "0.25.0-eng1209.4",
   "description": "Pack, share, and install agent templates across harnesses (Claude Code, Codex, ...).",
   "type": "module",
   "bin": {