Security hardening is required for the new tools.

• Action: Refine the system prompt to explicitly instruct the model on how to treat data returned by tools (logs/traces).

• Goal: Ensure the model treats tool output as data, not instructions, preventing malicious log entries from hijacking the chat behavior.