@brainbase-labs/cli 0.25.0-eng1209.5 → 0.25.0-eng1209.6

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-eng1209.5",
+  version: "0.25.0-eng1209.6",
   description: "Pack, share, and install agent templates across harnesses (Claude Code, Codex, ...).",
   type: "module",
   bin: {
@@ -78882,6 +78882,19 @@ class BenchmarkPhaseError extends Error {
     this.name = "BenchmarkPhaseError";
   }
 }
+
+class BenchmarkCommandExecutionError extends BenchmarkPhaseError {
+  stdout;
+  stderr;
+  durationMs;
+  constructor(code, message, stdout, stderr, durationMs) {
+    super(code, message);
+    this.stdout = stdout;
+    this.stderr = stderr;
+    this.durationMs = durationMs;
+    this.name = "BenchmarkCommandExecutionError";
+  }
+}
 var ZIP_EOCD_SIGNATURE = 101010256;
 var ZIP_CENTRAL_SIGNATURE = 33639248;
 var ZIP_LOCAL_SIGNATURE = 67324752;
@@ -79985,7 +79998,8 @@ async function runCommand(command, root, spec, context, additions = {}) {
   const remainingMs = context.deadline - Date.now();
   if (remainingMs <= 0)
     throw new BenchmarkPhaseError("phase_timeout", "phase budget expired");
-  const timeoutMs2 = Math.min(command.timeout_ms ?? remainingMs, remainingMs);
+  const reachesPhaseDeadline = command.timeout_ms === undefined || command.timeout_ms >= remainingMs;
+  const timeoutMs2 = reachesPhaseDeadline ? remainingMs : command.timeout_ms;
   const started = Date.now();
   return await new Promise((resolve, reject2) => {
     const child = spawn4(command.argv[0], command.argv.slice(1), {
@@ -80022,10 +80036,14 @@ async function runCommand(command, root, spec, context, additions = {}) {
     child.stdout?.on("data", (chunk2) => capture(stdout, chunk2));
     child.stderr?.on("data", (chunk2) => capture(stderr, chunk2));
     child.on("error", (error2) => {
-      fail(new BenchmarkPhaseError("command_start_failed", `failed to start ${command.id}: ${error2.message}`));
+      fail(new BenchmarkCommandExecutionError("command_start_failed", `failed to start ${command.id}: ${error2.message}`, Buffer.concat(stdout), Buffer.concat(stderr), Date.now() - started));
     });
     timer = setTimeout(() => {
-      fail(new BenchmarkPhaseError("command_timeout", `command timed out: ${command.id}`));
+      if (reachesPhaseDeadline) {
+        fail(new BenchmarkPhaseError("phase_timeout", "phase budget expired"));
+        return;
+      }
+      fail(new BenchmarkCommandExecutionError("command_timeout", `command timed out: ${command.id}`, Buffer.concat(stdout), Buffer.concat(stderr), Date.now() - started));
     }, timeoutMs2);
     child.on("close", (code, signal) => {
       if (settled)
@@ -80033,7 +80051,7 @@ async function runCommand(command, root, spec, context, additions = {}) {
       settled = true;
       clearTimeout(timer);
       if (code === null) {
-        reject2(new BenchmarkPhaseError("command_terminated", `command was terminated by ${signal ?? "an unknown signal"}: ${command.id}`));
+        reject2(new BenchmarkCommandExecutionError("command_terminated", `command was terminated by ${signal ?? "an unknown signal"}: ${command.id}`, Buffer.concat(stdout), Buffer.concat(stderr), Date.now() - started));
         return;
       }
       resolve({
@@ -80867,7 +80885,6 @@ async function executeEvaluate(spec, context) {
   for (const evaluator of executionOrder) {
     assertBudget(context);
     const started = Date.now();
-    let requiredResultError;
     try {
       const evaluated = await evaluateOne(evaluator, spec, manifest, finalOutput.buffer.toString("utf8"), trajectory, { finalOutputPath, trajectoryPath }, context);
       assertBudget(context);
@@ -80881,13 +80898,14 @@ async function executeEvaluate(spec, context) {
         outputs.push(evaluated.stderr);
         context.outputs.push(evaluated.stderr);
       }
-      if (evaluated.status === "errored" && evaluator.required) {
-        const errorCode = typeof evaluated.details?.error_code === "string" ? evaluated.details.error_code : "evaluator_execution_failed";
-        const errorMessage2 = typeof evaluated.details?.error_message === "string" ? evaluated.details.error_message : "required evaluator execution failed";
-        requiredResultError = new BenchmarkPhaseError(errorCode, errorMessage2);
-      }
     } catch (error2) {
       const normalized = stableError(error2);
+      let stdout;
+      let stderr;
+      if (error2 instanceof BenchmarkCommandExecutionError) {
+        stdout = await writeLog(spec.logs_root, `${evaluator.id}.stdout.log`, error2.stdout, spec);
+        stderr = await writeLog(spec.logs_root, `${evaluator.id}.stderr.log`, error2.stderr, spec);
+      }
       const errored = {
         id: evaluator.id,
         type: evaluator.type,
@@ -80897,20 +80915,28 @@ async function executeEvaluate(spec, context) {
         verdict: null,
         engine: "brainbase-cli",
         engine_version: VERSION,
-        duration_ms: Date.now() - started,
+        duration_ms: error2 instanceof BenchmarkCommandExecutionError ? error2.durationMs : Date.now() - started,
         details: {
           error_code: normalized?.code ?? "phase_failed",
           error_message: normalized?.message ?? "evaluator execution failed"
-        }
+        },
+        ...stdout ? { stdout } : {},
+        ...stderr ? { stderr } : {}
       };
       evaluators.push(errored);
       context.evaluators.push(errored);
-      if (evaluator.required)
+      if (stdout) {
+        outputs.push(stdout);
+        context.outputs.push(stdout);
+      }
+      if (stderr) {
+        outputs.push(stderr);
+        context.outputs.push(stderr);
+      }
+      if (!(error2 instanceof BenchmarkCommandExecutionError))
         throw error2;
       assertBudget(context);
     }
-    if (requiredResultError)
-      throw requiredResultError;
     if (evaluator.type === "sandbox_command") {
       assertEvaluateOutputBudget(spec, context);
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@brainbase-labs/cli",
-  "version": "0.25.0-eng1209.5",
+  "version": "0.25.0-eng1209.6",
   "description": "Pack, share, and install agent templates across harnesses (Claude Code, Codex, ...).",
   "type": "module",
   "bin": {