The Agent Skills Directory

[SAFE]: The skill implements several defensive measures for database access, including enforced read-only sessions (via PostgreSQL readonly=True mode) and strict query validation that only permits data retrieval commands (SELECT, SHOW, EXPLAIN, WITH).
[DATA_EXPOSURE]: The skill documentation describes using a local connections.json file for database configuration. It correctly identifies this as a security-sensitive file and provides instructions for the user to restrict file permissions (chmod 600). The provided example uses placeholders for sensitive values.
[COMMAND_EXECUTION]: The skill operates by executing a local Python script (scripts/query.py). While the script's source is not provided in this file, the documented interface indicates it is a standard wrapper for database operations.
[INDIRECT_PROMPT_INJECTION]: Like any tool that retrieves content from external sources (databases), the skill is a potential vector for indirect prompt injection if queried tables contain malicious instructions. The skill mitigates this risk by capping column widths and limiting row counts, reducing the complexity of data the agent must process.

postgres