fix: validate an OpenHands setup before using it

responses_api_agents/swe_agents/app.py

@@ -1716,6 +1716,45 @@ def _openhands_setup_target(self, commit: str) -> Path:
             return legacy_dir
         return self.setup_root / "swe_openhands_setup" / _repo_slug(self.config.agent_framework_repo) / commit
 
+    def _probe_openhands_venv(self, openhands_dir: Path) -> Optional[str]:
+        """Probe an existing OpenHands venv's deps; return None if healthy, else the failure."""
+        venv_python = openhands_dir / ".venv" / "bin" / "python"
+        probe = subprocess_run(
+            [
+                str(venv_python),
+                "-c",
+                (
+                    "from importlib.metadata import version; "
+                    "from packaging.version import Version; "
+                    "assert Version(version('jinja2')) >= Version('3.1.3'); "
+                    "assert Version(version('pyjwt')) >= Version('2.9'); "
+                    "assert Version(version('sqlalchemy')) >= Version('2.0.40'); "
+                    "assert Version(version('flask')) >= Version('2.2'); "
+                    "import datasets, wandb"
+                ),
+            ],
+            capture_output=True,
+            text=True,
+            env={**os.environ, "PYTHONNOUSERSITE": "1"},
+        )
+        if probe.returncode == 0:
+            return None
+        return " ".join(probe.stderr.strip().splitlines()[-1:]) or f"exit code {probe.returncode}"
+
+    def _existing_setup_is_reusable(self, openhands_dir: Path) -> bool:
+        """Whether a pre-existing OpenHands checkout can be reused as-is."""
+        if not (openhands_dir.exists() and Path(openhands_dir / ".venv" / "bin" / "python").exists()):
+            return False
+        probe_failure = self._probe_openhands_venv(openhands_dir)
+        if probe_failure is not None:
+            print(
+                f"OpenHands venv at {openhands_dir} failed its runtime probe ({probe_failure}); "
+                "rebuilding the setup instead of reusing it",
+                flush=True,
+            )
+            return False
+        return True
+
     def setup(self) -> Path:
         repo = self.config.agent_framework_repo
         ref = self.config.agent_framework_commit
@@ -1739,7 +1778,7 @@ def setup(self) -> Path:
             openhands_dir = setup_dir / "OpenHands"
             miniforge_dir = setup_dir / "miniforge3"
 
-            if self._openhands_tree_valid(setup_dir):
+            if self._existing_setup_is_reusable(openhands_dir):
                 print(f"OpenHands already set up at {setup_dir}", flush=True)
                 self._sync_openhands_to_commit(openhands_dir, commit)
                 return setup_dir

responses_api_agents/swe_agents/setup_scripts/openhands.sh

@@ -150,10 +150,10 @@ done
 echo "Installing Python dependencies (creating .venv in OpenHands directory)..."
 poetry install --no-interaction --no-root
 
-# Install datasets package
+# Install datasets package.
 echo "Installing datasets package..."
 
-poetry run python -m pip install datasets huggingface_hub packaging==26.0
+poetry run python -m pip install datasets huggingface_hub packaging==26.0 wandb
 
 mkdir -p evaluation/oh
 mkdir -p logs

responses_api_agents/swe_agents/tests/test_app.py

@@ -1164,6 +1164,41 @@ def test_get_run_command_no_replay_messages_omits_replay_file(self) -> None:
             assert not (config.persistent_dir / "replay_messages.json").exists()
             assert "replay_messages.json" not in self._read_agent_script(config)
 
+    def _reuse_processor(self) -> OpenHandsHarnessProcessor:
+        return OpenHandsHarnessProcessor(config=_minimal_server_config())
+
+    def _fake_openhands_dir(self, tmp_path: Path, probe_script: str) -> Path:
+        openhands_dir = tmp_path / "OpenHands"
+        bin_dir = openhands_dir / ".venv" / "bin"
+        bin_dir.mkdir(parents=True)
+        python = bin_dir / "python"
+        python.write_text(probe_script)
+        python.chmod(0o755)
+        return openhands_dir
+
+    @pytest.mark.parametrize(
+        ("probe_script", "reusable"),
+        [
+            (None, False),  # no setup at all
+            ("#!/bin/sh\nexit 0\n", True),  # healthy venv
+            ("#!/bin/sh\nexit 1\n", False),  # interpreter exists but the probe fails -> rebuild
+        ],
+    )
+    def test_setup_reuse_gate(self, tmp_path, probe_script, reusable) -> None:
+        """A pre-existing setup is reused only when its venv passes the
+        runtime probe: publishing a corpse makes every subsequent episode
+        fail after the agent has done its real work."""
+        openhands_dir = self._fake_openhands_dir(tmp_path, probe_script) if probe_script else tmp_path / "OpenHands"
+
+        assert self._reuse_processor()._existing_setup_is_reusable(openhands_dir) is reusable
+
+    def test_probe_reports_the_last_stderr_line(self, tmp_path) -> None:
+        openhands_dir = self._fake_openhands_dir(
+            tmp_path, "#!/bin/sh\necho 'ModuleNotFoundError: No module named wandb' >&2\nexit 1\n"
+        )
+        failure = self._reuse_processor()._probe_openhands_venv(openhands_dir)
+        assert failure is not None and "ModuleNotFoundError" in failure
+
 
 ########################################
 # Workspace path + user-message resolver tests