search-with-proof: ecosystem-aware build recipes and stderr in build failures

The agent passes a manifest file or project directory; pip uses -r for
requirements files and the directory for pyproject/setup, npm ci/install runs in
the directory, composer/maven/gradle likewise. Build and verify failures carry
phase, exit code and a sanitised stderr tail.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

benchmarks/ax/batch.py

@@ -724,7 +724,11 @@ class SearchExecutorTask:
         remaining = min(300, self.end - time.monotonic())
         result = self.client.submit("prepare_verification", spec, timeout_seconds=remaining)
         if result.get("status") != "ok":
-            raise ValueError(result.get("reason", "could_not_build"))
+            from openultrasast.search.task_storage import CommandFailure
+
+            raise CommandFailure(
+                result.get("phase", "build"), result.get("exit_code"), result.get("stderr", ""), result.get("reason", "could_not_build")
+            )
         stored = bounded_read(lambda: self.lane.store._get(self.keys[3]), self.end - time.monotonic())
         if stored is None or hashlib.sha256(stored[0]).hexdigest() != result.get("sha256"):
             raise ValueError("executor archive missing or digest mismatch")

benchmarks/ax/search_verify.py

@@ -110,4 +110,7 @@ class AXSideDispatcher:
             max(r.scratch_peak_bytes for r in records),
             isolation_mode="task-boundary",
             chromium_no_sandbox=any(r.chromium_no_sandbox for r in records),
+            phase=last.phase,
+            exit_code=last.exit_code,
+            stderr=last.stderr,
         )