Gracefully handle corrupt or malformed run store files

- Catch OSError and JSONDecodeError in summarize() and return None
  instead of raising
- Validate all events are dicts and run_id is a non-empty string
  before producing a RunSummary

Author: johnteee

teaagent/run_store.py

@@ -91,10 +91,17 @@ def heartbeat_for_run(self, run_id: str) -> dict[str, Any]:
         }
 
     def summarize(self, path: Path) -> Optional[RunSummary]:
-        events = [json.loads(line) for line in path.read_text(encoding="utf-8").splitlines() if line.strip()]
+        try:
+            events = [json.loads(line) for line in path.read_text(encoding="utf-8").splitlines() if line.strip()]
+        except (OSError, json.JSONDecodeError):
+            return None
         if not events:
             return None
-        run_id = events[0]["run_id"]
+        if not all(isinstance(event, dict) for event in events):
+            return None
+        run_id = events[0].get("run_id")
+        if not isinstance(run_id, str) or not run_id:
+            return None
         task = ""
         status = "unknown"
         final_answer = None

tests/test_run_store.py

@@ -73,6 +73,32 @@ def test_cli_lists_and_shows_runs(self) -> None:
             self.assertEqual(json.loads(list_output.getvalue())[0]["run_id"], "run-2")
             self.assertEqual(json.loads(show_output.getvalue())[1]["event_type"], "run_failed")
 
+    def test_list_runs_skips_corrupt_jsonl_files(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp:
+            store = RunStore(tmp)
+            audit = store.audit_logger()
+            audit.record("run_started", "run-ok", task="demo")
+            audit.record("run_completed", "run-ok", answer="ok", metadata={})
+            store.logger_for_result(
+                RunResult(run_id="run-ok", final_answer=FinalAnswer("ok"), iterations=1, tool_calls=0, status="completed"),
+                audit,
+            )
+            (Path(tmp) / ".teaagent" / "runs" / "broken.jsonl").write_text("not json\n", encoding="utf-8")
+
+            summaries = store.list_runs()
+
+            self.assertEqual([summary.run_id for summary in summaries], ["run-ok"])
+
+    def test_list_runs_skips_records_without_run_id(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp:
+            store = RunStore(tmp)
+            (Path(tmp) / ".teaagent" / "runs" / "missing-id.jsonl").write_text(
+                json.dumps({"event_type": "run_started", "payload": {"task": "x"}}) + "\n",
+                encoding="utf-8",
+            )
+
+            self.assertEqual(store.list_runs(), [])
+
 
 if __name__ == "__main__":
     unittest.main()