The Agent Skills Directory

[COMMAND_EXECUTION]: The bundled script scripts/zoodata.py uses a background process to perform maintenance. Specifically, it uses subprocess.Popen for a self-invocation that cleans up stale temporary result files. This operation is restricted to a private, user-owned temporary namespace created with secure (0700) permissions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as Amazon product titles and review text, which it incorporates into instructions for the AI agent. This ingestion surface is managed using standard boundary markers (triple quotes) and strict output formatting requirements (JSON) to mitigate accidental instruction overrides.
[DATA_EXPOSURE]: The skill is designed to handle sensitive credentials via environment variables or a local configuration file (~/.zoodata/config.json). The CLI implementation includes a whitelist check to ensure the ZOODATA_API_KEY is only transmitted to trusted vendor hosts or localhost, protecting the token from exfiltration to arbitrary third-party servers.

zoodata