The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (query_executor.py) using the uv run command to perform database operations.
[PROMPT_INJECTION]: The skill implements a Text-to-SQL workflow which is vulnerable to indirect prompt injection. A user could potentially craft natural language inputs designed to influence the generated SQL query.
Ingestion points: User-provided natural language queries are processed by the agent to generate SQL code.
Boundary markers: The SKILL.md file provides clear instructions to the agent to only generate SELECT statements and exclude specific columns.
Capability inventory: The skill has the capability to execute SQL queries via sqlalchemy and read the schema.sql reference file.
Sanitization: The query_executor.py script includes a validate_query function that checks for forbidden DDL/DML keywords (e.g., DROP, DELETE, INSERT) and dangerous system functions. It also wraps queries in a subquery to enforce a LIMIT and uses execution timeouts.

autorag-query