p2

Author: rcholic

sentience/agent_runtime.py

@@ -326,19 +326,42 @@ def finalize_run(self, *, success: bool) -> None:
             return
         if success:
             if self._artifact_buffer.options.persist_mode == "always":
-                self._artifact_buffer.persist(reason="success", status="success")
+                self._artifact_buffer.persist(
+                    reason="success",
+                    status="success",
+                    snapshot=self.last_snapshot,
+                    diagnostics=getattr(self.last_snapshot, "diagnostics", None),
+                    metadata=self._artifact_metadata(),
+                )
             self._artifact_buffer.cleanup()
         else:
             self._persist_failure_artifacts(reason="finalize_failure")
 
     def _persist_failure_artifacts(self, *, reason: str) -> None:
         if not self._artifact_buffer:
             return
-        self._artifact_buffer.persist(reason=reason, status="failure")
+        self._artifact_buffer.persist(
+            reason=reason,
+            status="failure",
+            snapshot=self.last_snapshot,
+            diagnostics=getattr(self.last_snapshot, "diagnostics", None),
+            metadata=self._artifact_metadata(),
+        )
         self._artifact_buffer.cleanup()
         if self._artifact_buffer.options.persist_mode == "onFail":
             self.disable_failure_artifacts()
 
+    def _artifact_metadata(self) -> dict[str, Any]:
+        url = None
+        if self.last_snapshot is not None:
+            url = self.last_snapshot.url
+        elif self._cached_url:
+            url = self._cached_url
+        return {
+            "backend": self.backend.__class__.__name__,
+            "url": url,
+        }
+
     def begin_step(self, goal: str, step_index: int | None = None) -> str:
         """
         Begin a new step in the verification loop.

sentience/failure_artifacts.py

@@ -4,9 +4,10 @@
 import shutil
 import tempfile
 import time
+from collections.abc import Callable
 from dataclasses import dataclass
 from pathlib import Path
-from typing import Callable, Literal
+from typing import Any, Literal
 
 
 @dataclass
@@ -93,7 +94,20 @@ def _prune(self) -> None:
                     pass
         self._frames = keep
 
-    def persist(self, *, reason: str | None, status: Literal["failure", "success"]) -> Path | None:
+    def _write_json_atomic(self, path: Path, data: Any) -> None:
+        tmp_path = path.with_suffix(path.suffix + ".tmp")
+        tmp_path.write_text(json.dumps(data, indent=2))
+        tmp_path.replace(path)
+
+    def persist(
+        self,
+        *,
+        reason: str | None,
+        status: Literal["failure", "success"],
+        snapshot: Any | None = None,
+        diagnostics: Any | None = None,
+        metadata: dict[str, Any] | None = None,
+    ) -> Path | None:
         if self._persisted:
             return None
 
@@ -107,8 +121,23 @@ def persist(self, *, reason: str | None, status: Literal["failure", "success"])
         for frame in self._frames:
             shutil.copy2(frame.path, frames_out / frame.file_name)
 
-        steps_path = run_dir / "steps.json"
-        steps_path.write_text(json.dumps(self._steps, indent=2))
+        self._write_json_atomic(run_dir / "steps.json", self._steps)
+
+        snapshot_payload = None
+        if snapshot is not None:
+            if hasattr(snapshot, "model_dump"):
+                snapshot_payload = snapshot.model_dump()
+            else:
+                snapshot_payload = snapshot
+            self._write_json_atomic(run_dir / "snapshot.json", snapshot_payload)
+
+        diagnostics_payload = None
+        if diagnostics is not None:
+            if hasattr(diagnostics, "model_dump"):
+                diagnostics_payload = diagnostics.model_dump()
+            else:
+                diagnostics_payload = diagnostics
+            self._write_json_atomic(run_dir / "diagnostics.json", diagnostics_payload)
 
         manifest = {
             "run_id": self.run_id,
@@ -117,12 +146,12 @@ def persist(self, *, reason: str | None, status: Literal["failure", "success"])
             "reason": reason,
             "buffer_seconds": self.options.buffer_seconds,
             "frame_count": len(self._frames),
-            "frames": [
-                {"file": frame.file_name, "ts": frame.ts} for frame in self._frames
-            ],
+            "frames": [{"file": frame.file_name, "ts": frame.ts} for frame in self._frames],
+            "snapshot": "snapshot.json" if snapshot_payload is not None else None,
+            "diagnostics": "diagnostics.json" if diagnostics_payload is not None else None,
+            "metadata": metadata or {},
         }
-        manifest_path = run_dir / "manifest.json"
-        manifest_path.write_text(json.dumps(manifest, indent=2))
+        self._write_json_atomic(run_dir / "manifest.json", manifest)
 
         self._persisted = True
         return run_dir

tests/unit/test_failure_artifacts.py

@@ -34,11 +34,26 @@ def time_fn() -> float:
     buf.record_step(action="CLICK", step_id="s1", step_index=1, url="https://example.com")
     buf.add_frame(b"frame")
 
-    run_dir = buf.persist(reason="assert_failed", status="failure")
+    snapshot = {"status": "success", "url": "https://example.com", "elements": []}
+    diagnostics = {"confidence": 0.9, "reasons": ["ok"], "metrics": {"quiet_ms": 42}}
+    run_dir = buf.persist(
+        reason="assert_failed",
+        status="failure",
+        snapshot=snapshot,
+        diagnostics=diagnostics,
+        metadata={"backend": "MockBackend", "url": "https://example.com"},
+    )
     assert run_dir is not None
     manifest = json.loads((run_dir / "manifest.json").read_text())
     steps = json.loads((run_dir / "steps.json").read_text())
+    snap_json = json.loads((run_dir / "snapshot.json").read_text())
+    diag_json = json.loads((run_dir / "diagnostics.json").read_text())
 
     assert manifest["run_id"] == "run-2"
     assert manifest["frame_count"] == 1
+    assert manifest["snapshot"] == "snapshot.json"
+    assert manifest["diagnostics"] == "diagnostics.json"
+    assert manifest["metadata"]["backend"] == "MockBackend"
     assert len(steps) == 1
+    assert snap_json["url"] == "https://example.com"
+    assert diag_json["confidence"] == 0.9