The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive API keys (OpenAI, Anthropic, Gemini, S3, etc.) by storing them in a plaintext file at ~/.skills.env. While it enforces chmod 600 permissions to limit local access, plaintext storage remains a risk for credential exposure at rest.
[COMMAND_EXECUTION]: Documentation suggests using eval "$(skills-keys export)" to load keys into the shell environment. This pattern is vulnerable to command injection if the environment file contains malicious values with shell metacharacters, which would be executed upon evaluation.
[DATA_EXFILTRATION]: The verify command transmits API keys to official endpoints of trusted organizations (e.g., api.openai.com, api.anthropic.com) to confirm validity. This intended behavior involves sending high-entropy secrets over the network.
[DYNAMIC_EXECUTION]: The scripts/run.py entry point utilizes os.execv for process replacement, re-executing the script within a detected virtual environment.

skills-keys