If your tools execute shell commands, access files, or interact with external systems, review and restrict those capabilities before using the agent with untrusted prompts. Python LLM agent ...