Devstral Small 2: serve.sh, baseline.py without thinking args, repair rate, activation error message fix, stop rule in chain (run base 21000)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014aUaQeLnwbb1zTpN7kHeat

harness/proxy.py

@@ -43,8 +43,8 @@ def activation_messages(text, is_error=False):
         return []
     msgs = (obj.get("activation") or {}).get("messages") or obj.get("messages") or []
     out = [str(m.get("message", ""))[:300] for m in msgs if isinstance(m, dict) and m.get("severity") == "E"]
-    if not out and is_error:
-        out = [str(obj.get("message") or text)[:300]]
+    if not out and (is_error or obj.get("success") is False):
+        out = [str(obj.get("error") or obj.get("message") or text)[:300]]  # {"success":false,"error":...} of a failed write
     return out
 
 

harness/runner.py

@@ -91,6 +91,26 @@ def trajectory_stats(path):
             "final_report": final, "agent_seconds": round((t_last or 0) - (t_first or 0), 1)}
 
 
+def repair_stats(path):
+    """Repair rate: pushes whose source changed after a failed push of the same object / pushes after a failed push."""
+    import hashlib
+    last = None  # (object, md5, failed)
+    after = changed = 0
+    for line in open(path):
+        e = json.loads(line)
+        if e.get("tool") != "sap_push_source" or (e["result"].startswith("Tool ") and e["is_error"]):
+            continue
+        args = e.get("args", {})
+        key = (str(args.get("objectName", "")).upper(), str(args.get("includeType", "")))
+        md5 = hashlib.md5(str(args.get("source", "")).encode()).hexdigest()
+        if last and last[2] and last[0] == key:
+            after += 1
+            changed += md5 != last[1]
+        last = (key, md5, e["is_error"] or '"success":false' in e["result"].replace(" ", ""))
+    return {"pushes_after_error": after, "pushes_changed_after_error": changed,
+            "repair_rate": round(changed / after, 3) if after else None}
+
+
 class Runner:
     def __init__(self, tasks_root, runs_root):
         self.tasks_root = tasks_root