The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages user credentials for Google APIs by storing them in standard configuration paths (e.g., ~/.config/claude-seo/). A comprehensive redaction mechanism is implemented in scripts/google_auth.py using regular expressions to prevent sensitive data like API keys or tokens from being exposed in error messages or logs.
[INDIRECT_PROMPT_INJECTION]: When fetching content from external URLs (in scripts/nlp_analyze.py), the skill includes robust SSRF protections. It validates that resolved IP addresses are globally routable and explicitly blocks private, loopback, and link-local addresses, minimizing the risk of internal network scanning.
[COMMAND_EXECUTION]: Command execution in scripts/run.py is used to launch sub-scripts within a managed virtual environment. The implementation includes validation to prevent directory traversal by ensuring the target script is located within the authorized scripts directory.
[EXTERNAL_DOWNLOADS]: External dependencies are installed using pip with strict hash verification against a lock file (scripts/requirements.lock), protecting against supply chain attacks and ensuring the integrity of the runtime environment.

blog-google