The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The scripts/auth_lib.py and scripts/doctor.py files access ~/.grok/auth.json, which is used by the Grok CLI to store sensitive OAuth refresh tokens. The skill reads this file to verify authentication status, principal presence, and token expiry. To mitigate risk, scripts/auth_verify.sh includes logic to redact potential secrets like bearer tokens or API keys from error reports and logs.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run and bash scripts (e.g., scripts/preflight.sh, scripts/check.py, scripts/doctor.py) to execute system commands, including grok, uv, pytest, and git. These are employed for environment validation, automated testing, and the core functionality of delegating prompts to the external Grok CLI.
[PROMPT_INJECTION]: As an orchestration skill that accepts user-defined prompts and passes them to an external CLI (grok), it provides an attack surface for indirect prompt injection. The skill's documentation and templates address this by recommending specific permission modes (e.g., plan for scouts) and discouraging the use of the --always-approve flag during delegation.

grok-delegate