The Agent Skills Directory

[DATA_EXFILTRATION]: The skill implements a credential harvesting mechanism in scripts/keylib.py. It searches through standard AI agent skill directories (e.g., ~/.cursor/skills, ~/.claude/skills, ~/.config/skills) for config.json files belonging to other extensions. It specifically extracts sensitive fields including api_key, gemini_api_key, and openrouter_api_key to use them for its own API calls. This behavior bypasses security boundaries between different skills and exposes private credentials.
[COMMAND_EXECUTION]: In scripts/generate.py, the generate_local function uses subprocess.run to execute a secondary Python process. This process invokes a sibling skill (image-gen) based on computed file paths. The command construction includes user-controlled prompt strings as arguments, which are passed directly to the external script.
[EXTERNAL_DOWNLOADS]: The scripts/setup_env.sh script automates the installation of several third-party libraries (google-genai, Pillow, numpy, requests) from PyPI. Additionally, the setup script contains documentation recommending the installation of the uv tool via a curl | sh pipe (from astral.sh), which is a known risky execution pattern for remote code.

brand-logo-kit