The Agent Skills Directory

[SAFE]: The skill provides developer tools for local prompt analysis and optimization. \n- [SAFE]: The use of uv run and importlib.util are standard patterns for running and loading local Python code for testing purposes. \n- [SAFE]: The use of FakeLLMClient ensures that rendered prompts are not sent to external APIs during the diagnostic process, preventing unintentional data exposure. \n- [SAFE]: The instructions promote security best practices by recommending the use of markers like @hidden and nosnapshot to prevent sensitive internal state or lifecycle data from being exposed to the LLM or persisted in session snapshots.

refine-agent-prompt