fix: retain agent challenge parse errors

examples/agent_challenges/browser_click_challenge/run_opencode_trials.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 import subprocess
 import sys
 from pathlib import Path
+from typing import Any
 
 # Support direct execution as `python examples/.../run_opencode_trials.py`.
 # ruff: noqa: I001 - imports must stay after sys.path.insert
@@ -135,7 +136,7 @@ def run_trial(
     *,
     index: int,
     results_dir: Path,
-) -> dict:
+) -> dict[str, Any]:
     return _generic_run_trial(
         config, index=index, results_dir=results_dir, classify_fn=classify_output
     )

examples/agent_challenges/runner.py

@@ -139,13 +139,18 @@ def run_trial(
         return payload
 
     parsed: dict[str, Any] | None
+    parse_error: dict[str, str] | None = None
     try:
         parsed = parse_opencode_output(completed.stdout)
         text = result_text(parsed)
         classification = classify_fn(text)
-    except Exception:
+    except Exception as exc:
         parsed = None
         classification = "parse_error"
+        parse_error = {
+            "type": type(exc).__name__,
+            "message": str(exc),
+        }
 
     payload = {
         "index": index,
@@ -158,6 +163,8 @@ def run_trial(
         "stderr": completed.stderr,
         "parsed": parsed,
     }
+    if parse_error is not None:
+        payload["parse_error"] = parse_error
     _write_trial_report(payload)
     _write_trial_result(results_dir, config=config, index=index, payload=payload)
     return payload

tests/examples/test_opencode_browser_click_challenge.py

@@ -254,6 +254,47 @@ def test_run_trial_records_report_save_error_for_timeout(
     assert saved_result["report_save_error"] == "result file is missing parsed output"
 
 
+def test_run_trial_records_parse_error_details(
+    tmp_path: Path,
+    monkeypatch,
+) -> None:
+    prompt = tmp_path / "prompt.md"
+    prompt.write_text("hello", encoding="utf-8")
+
+    def fake_run(*args: object, **kwargs: object) -> subprocess.CompletedProcess[str]:
+        return subprocess.CompletedProcess(
+            args=["opencode"],
+            returncode=0,
+            stdout="not-json",
+            stderr="",
+        )
+
+    from examples.agent_challenges import runner as generic_runner
+
+    monkeypatch.setattr(generic_runner.subprocess, "run", fake_run)
+    config = TrialConfig(
+        model="opencode/mimo-v2.5-free",
+        variant="high",
+        prompt_path=prompt,
+        attach_url=None,
+        timeout_seconds=120,
+        wf_command_prefix=LOCAL_WF_COMMAND_PREFIX,
+        server_context="Use local CLI mode.",
+    )
+
+    result = run_trial(config, index=1, results_dir=tmp_path / "results")
+
+    assert result["classification"] == "parse_error"
+    assert result["parse_error"]["type"] == "JSONDecodeError"
+    assert "Expecting value" in result["parse_error"]["message"]
+    saved_result = json.loads(
+        (tmp_path / "results" / "opencode_mimo-v2.5-free-trial-001.json").read_text(
+            encoding="utf-8"
+        )
+    )
+    assert saved_result["parse_error"] == result["parse_error"]
+
+
 def test_parse_opencode_output_reads_json_object() -> None:
     payload = {
         "text": "wf run start demo.default\nbefore.clicked false\nafter.clicked true"