The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess module in scripts/detect_tools.py to execute local commands for environment discovery.
It runs claude mcp list to programmatically identify available Model Context Protocol servers so they can be included in the permission allowlist.
This execution is limited to system discovery and does not process external or untrusted data sources.
[SAFE]: The skill implements strong security defaults and validation when managing agent permissions.
Sensitive tools associated with network exfiltration or infrastructure mutation (such as curl, docker, and aws) are gated behind an optional flag to prevent accidental over-permissioning.
The script includes logic (the codex_path_safe function) to detect and prevent potential shell injection vulnerabilities when generating commands for the user to execute.
Configuration updates to agent settings are performed using atomic writes and non-destructive merging to ensure file integrity.

project-tool-allowlist