The Agent Skills Directory

Command Execution (MEDIUM): The skill utilizes system CLIs (sqlite3, psql, mysql) to manage databases and specifically demonstrates a pattern of executing strings generated by an LLM (psql -c \"$QUERY\"). This constitutes dynamic execution of content derived from external user input. While central to the skill's purpose, this pattern requires strict oversight.
Indirect Prompt Injection (LOW): The "AI-Assisted Query Generation" feature creates an indirect prompt injection surface by allowing natural language to determine executable SQL. Mandatory Evidence Chain: 1. Ingestion points: User descriptions provided to the gemini tool. 2. Boundary markers: The skill recommends manual verification of the generated output (echo \"$QUERY\") prior to execution. 3. Capability inventory: Subprocess calls to database engines capable of data modification and deletion. 4. Sanitization: None; the skill relies on the user for security validation.

db