fix: classify opencode text events and helper scripts

examples/agent_challenges/browser_click_challenge/README.md

@@ -55,6 +55,8 @@ workflow output.
 Each trial is classified as one of:
 
 - `success`: output shows workflow usage and before/after clicked states.
+- `workflow_script`: output shows a workflow run, but the agent drove it through
+  a new helper script instead of the product-facing CLI/server path.
 - `workflow_not_used`: output appears to solve the task without `wf`,
   `wf-rpc-server`, deployment, or run evidence.
 - `run_failed`: output includes workflow usage but reports a failure.

examples/agent_challenges/browser_click_challenge/prompt.md

@@ -12,6 +12,14 @@ Use this repository's workflow product path. That means you should use the
 the deployment through the workflow API. Do not solve the challenge with only a
 standalone Playwright/Python script.
 
+Do not create a new helper script whose only job is to drive `WorkflowApi`
+directly. The challenge is about whether the product-facing CLI/server workflow
+can be discovered and used.
+
+If you need to write a workflow definition, write a declarative JSON/YAML file
+and then apply/run it through the product-facing workflow tools. Do not hide the
+workflow construction inside a Python script.
+
 The repository already includes a deterministic source example at:
 
 ```text

examples/agent_challenges/browser_click_challenge/run_opencode_trials.py

@@ -10,6 +10,7 @@ from typing import Any, Literal
 
 Classification = Literal[
     "success",
+    "workflow_script",
     "workflow_not_used",
     "run_failed",
     "timeout",
@@ -71,6 +72,7 @@ def parse_opencode_output(stdout: str) -> dict[str, Any]:
 
 def _parse_jsonl_tail(text: str) -> dict[str, Any]:
     last_error: json.JSONDecodeError | None = None
+    parsed_events: list[dict[str, Any]] = []
     for line in reversed(text.splitlines()):
         stripped = line.strip()
         if not stripped:
@@ -81,22 +83,55 @@ def _parse_jsonl_tail(text: str) -> dict[str, Any]:
             last_error = exc
             continue
         if isinstance(parsed, dict):
-            return parsed
+            parsed_events.append(parsed)
+            event_text = _event_text(parsed)
+            if event_text is not None:
+                return {"text": event_text, "event": parsed}
+    if parsed_events:
+        return parsed_events[0]
     if last_error is not None:
         raise last_error
     raise ValueError("opencode output did not contain JSON lines")
 
 
+def _event_text(event: dict[str, Any]) -> str | None:
+    """Extract assistant text from opencode JSON events.
+
+    `opencode run --format json` emits many events. The final event can be a
+    `step_finish`; the answer text is usually the previous `text` event under
+    `part.text`.
+    """
+    part = event.get("part")
+    if isinstance(part, dict):
+        text = part.get("text")
+        if isinstance(text, str):
+            return text
+    text = event.get("text")
+    return text if isinstance(text, str) else None
+
+
 def classify_output(text: str) -> Classification:
     lowered = text.lower()
-    workflow_markers = [
+    product_command_markers = [
         "wf ",
         "wf-rpc-server",
+    ]
+    workflow_evidence_markers = [
         "deployment",
         "run id",
         "run_",
     ]
-    used_workflow = any(marker in lowered for marker in workflow_markers)
+    used_product_command = any(
+        marker in lowered for marker in product_command_markers
+    )
+    has_workflow_evidence = any(
+        marker in lowered for marker in workflow_evidence_markers
+    )
+    used_helper_script = (
+        "uv run python" in lowered
+        or "python examples/" in lowered
+        or "run_workflow.py" in lowered
+    )
     failed = any(
         marker in lowered
         for marker in [
@@ -107,26 +142,33 @@ def classify_output(text: str) -> Classification:
             "validation failed",
         ]
     )
-    before_false = (
+    before_false = _contains_bool_marker(lowered, "before.clicked", "false") or (
-        "before.clicked is false" in lowered
+        '"before"' in lowered and '"clicked": false' in lowered
-        or '"before"' in lowered
-        and '"clicked": false' in lowered
     )
-    after_true = (
+    after_true = _contains_bool_marker(lowered, "after.clicked", "true") or (
-        "after.clicked is true" in lowered
+        '"after"' in lowered and '"clicked": true' in lowered
-        or '"after"' in lowered
-        and '"clicked": true' in lowered
     )
 
-    if used_workflow and before_false and after_true and not failed:
+    if used_product_command and before_false and after_true and not failed:
         return "success"
-    if used_workflow and failed:
+    if has_workflow_evidence and used_helper_script and before_false and after_true:
+        return "workflow_script"
+    if (used_product_command or has_workflow_evidence) and failed:
         return "run_failed"
-    if not used_workflow and (before_false or after_true or "playwright" in lowered):
+    if not has_workflow_evidence and (
+        before_false or after_true or "playwright" in lowered
+    ):
         return "workflow_not_used"
     return "unknown"
 
 
+def _contains_bool_marker(text: str, marker: str, value: str) -> bool:
+    marker_index = text.find(marker)
+    if marker_index == -1:
+        return False
+    return value in text[marker_index : marker_index + 80]
+
+
 def trial_output_path(results_dir: Path, *, model: str, index: int) -> Path:
     safe_model = model.replace("/", "_").replace(":", "_")
     return results_dir / f"{safe_model}-trial-{index:03d}.json"

tests/examples/test_opencode_browser_click_challenge.py

@@ -72,6 +72,27 @@ def test_parse_opencode_output_reads_last_jsonl_object() -> None:
     assert parsed["text"] == "final"
 
 
+def test_parse_opencode_output_prefers_text_event_before_step_finish() -> None:
+    payload = "\n".join(
+        [
+            json.dumps(
+                {
+                    "type": "text",
+                    "part": {
+                        "type": "text",
+                        "text": "deployment id: demo.default\nbefore.clicked is false",
+                    },
+                }
+            ),
+            json.dumps({"type": "step_finish", "part": {"type": "step-finish"}}),
+        ]
+    )
+
+    parsed = parse_opencode_output(payload)
+
+    assert parsed["text"] == "deployment id: demo.default\nbefore.clicked is false"
+
+
 def test_classify_output_success() -> None:
     result = classify_output(
         """
@@ -87,6 +108,20 @@ def test_classify_output_success() -> None:
     assert result == "success"
 
 
+def test_classify_output_workflow_script() -> None:
+    result = classify_output(
+        """
+        uv run python examples/browser_click_workflow/run_workflow.py
+        Deployment id: browser_click_case_study.default
+        Run id: run_123
+        `before.clicked`: `False`
+        `after.clicked`: `True`
+        """
+    )
+
+    assert result == "workflow_script"
+
+
 def test_classify_output_workflow_not_used() -> None:
     result = classify_output(
         """