The Agent Skills Directory

[DATA_EXFILTRATION]: The pre_tool_use.py hook records all tool interactions, including their full inputs, to logs/pre_tool_use.json. This practice stores potentially sensitive data like API keys, secrets, or private information in plain text within the project's filesystem.
[COMMAND_EXECUTION]: The orchestrator's command filtering logic is susceptible to bypasses. It allows any command starting with a 'safe prefix' (like echo or git ), which permits an attacker or a confused agent to execute restricted write operations via command chaining or redirects (e.g., echo 'malicious content' > config.yaml).
[COMMAND_EXECUTION]: The skill requires administrative-like changes to the project environment during setup, including modifying the AI agent's core configuration file (.claude/settings.json) and altering file execution permissions via chmod +x.

orchestrator