feat(runtime): capture step evidence in console

storage/artifact_store.py

@@ -20,21 +20,36 @@ class ArtifactStore:
         *,
         task_id: str,
         index: int,
-        screenshot: bytes | None,
+        before_screenshot: bytes | None,
+        after_screenshot: bytes | None,
         record: dict[str, Any],
     ) -> dict[str, str | None]:
         task_dir = self.task_dir(task_id)
         task_dir.mkdir(parents=True, exist_ok=True)
         stem = f"{index:03d}"
-        screenshot_path: Path | None = None
-        if screenshot is not None:
-            screenshot_path = task_dir / f"{stem}.png"
-            screenshot_path.write_bytes(screenshot)
+        before_screenshot_path: Path | None = None
+        after_screenshot_path: Path | None = None
+        if before_screenshot is not None:
+            before_screenshot_path = task_dir / f"{stem}-before.png"
+            before_screenshot_path.write_bytes(before_screenshot)
+        if after_screenshot is not None:
+            # Preserve the original filename as the compatibility alias for the
+            # screenshot captured after a step.
+            after_screenshot_path = task_dir / f"{stem}.png"
+            after_screenshot_path.write_bytes(after_screenshot)
 
         json_path = task_dir / f"{stem}.json"
         payload = {
             **record,
-            "screenshot_path": str(screenshot_path) if screenshot_path else None,
+            "before_screenshot_path": (
+                str(before_screenshot_path) if before_screenshot_path else None
+            ),
+            "after_screenshot_path": (
+                str(after_screenshot_path) if after_screenshot_path else None
+            ),
+            "screenshot_path": (
+                str(after_screenshot_path) if after_screenshot_path else None
+            ),
         }
         json_path.write_text(
             json.dumps(_jsonable(payload), ensure_ascii=False, indent=2),
@@ -42,7 +57,15 @@ class ArtifactStore:
         )
         return {
             "json_path": str(json_path),
-            "screenshot_path": str(screenshot_path) if screenshot_path else None,
+            "before_screenshot_path": (
+                str(before_screenshot_path) if before_screenshot_path else None
+            ),
+            "after_screenshot_path": (
+                str(after_screenshot_path) if after_screenshot_path else None
+            ),
+            "screenshot_path": (
+                str(after_screenshot_path) if after_screenshot_path else None
+            ),
         }
 
     def read_steps(self, task_id: str) -> list[dict[str, Any]]:

storage/timeline.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from datetime import datetime
 from typing import Any
 
@@ -17,6 +17,9 @@ class TimelineRecord:
     tool_call: dict[str, Any]
     result: dict[str, Any]
     timestamp: str
+    ocr_results: list[dict[str, Any]] = field(default_factory=list)
+    before_screenshot_path: str | None = None
+    after_screenshot_path: str | None = None
     screenshot_path: str | None = None
 
 
@@ -33,20 +36,29 @@ class Timeline:
         tool_call: dict[str, Any],
         result: dict[str, Any],
         screenshot: bytes | None = None,
+        before_screenshot: bytes | None = None,
+        after_screenshot: bytes | None = None,
+        ocr_results: list[dict[str, Any]] | None = None,
     ) -> TimelineRecord:
         index = len(self.read(task_id)) + 1
+        resolved_after_screenshot = (
+            after_screenshot if after_screenshot is not None else screenshot
+        )
+        resolved_ocr_results = list(ocr_results or [])
         record = {
             "index": index,
             "scene": scene,
             "prompt": prompt,
             "tool_call": tool_call,
             "result": result,
+            "ocr_results": resolved_ocr_results,
             "timestamp": datetime.now().astimezone().isoformat(),
         }
         paths = self.artifact_store.write_step(
             task_id=task_id,
             index=index,
-            screenshot=screenshot,
+            before_screenshot=before_screenshot,
+            after_screenshot=resolved_after_screenshot,
             record=record,
         )
         return TimelineRecord(
@@ -56,6 +68,9 @@ class Timeline:
             tool_call=tool_call,
             result=result,
             timestamp=record["timestamp"],
+            ocr_results=resolved_ocr_results,
+            before_screenshot_path=paths["before_screenshot_path"],
+            after_screenshot_path=paths["after_screenshot_path"],
             screenshot_path=paths["screenshot_path"],
         )