sanitize invalid ToolCall.Arguments before save; atomic write

pkg/agent/session.go

@@ -112,6 +112,23 @@ func (s *Session) saveTo(path, id, agentName, cwd string) error {
 }
 
 // saveToFull writes session state including backend/model info.
+func sanitizeMessages(msgs []backend.Message) []backend.Message {
+	out := make([]backend.Message, 0, len(msgs))
+	for _, m := range msgs {
+		for j, tc := range m.ToolCalls {
+			if len(tc.Arguments) > 0 {
+				var dummy json.RawMessage
+				if err := json.Unmarshal(tc.Arguments, &dummy); err != nil {
+					// Corrupted arguments — replace with empty object so save succeeds.
+					m.ToolCalls[j].Arguments = json.RawMessage("{}")
+				}
+			}
+		}
+		out = append(out, m)
+	}
+	return out
+}
+
 func (s *Session) saveToFull(path, id, agentName, backendName, modelName, cwd, remote string) error {
 	ps := PersistedSession{
 		ID:                       id,
@@ -120,7 +137,7 @@ func (s *Session) saveToFull(path, id, agentName, backendName, modelName, cwd, r
 		Model:                    modelName,
 		CWD:                      cwd,
 		Remote:                   remote,
-		Messages:                 s.messages,
+		Messages:                 sanitizeMessages(s.messages),
 		TaskState:                s.TaskState,
 		TotalInputTokens:         s.TotalInputTokens,
 		TotalCachedInputTokens:   s.TotalCachedInputTokens,
@@ -138,7 +155,12 @@ func (s *Session) saveToFull(path, id, agentName, backendName, modelName, cwd, r
 	if err != nil {
 		return fmt.Errorf("session save: %w", err)
 	}
-	return os.WriteFile(path, data, 0600)
+	// Atomic write: write to temp file, then rename to preserve last good snapshot.
+	tmpPath := path + ".tmp"
+	if err := os.WriteFile(tmpPath, data, 0600); err != nil {
+		return fmt.Errorf("session save tmp: %w", err)
+	}
+	return os.Rename(tmpPath, path)
 }
 
 // LoadPersistedSession reads a PersistedSession from a JSON file.