The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the target database, which can be manipulated to influence the agent's behavior.
Ingestion points: The schema command in scripts/sql_query_helper.py extracts table names, column types, foreign key relationships, and sample row data to provide context for the agent.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard instructions embedded within the extracted schema or sample data.
Capability inventory: The skill can read local file systems (SQLite database files), perform network operations (PostgreSQL connections), and execute SQL queries through the explain and schema commands.
Sanitization: While the script implements a _validate_readonly function to block dangerous SQL keywords (e.g., INSERT, DROP, ATTACH) and multi-statement queries, it does not sanitize the content of the data for prompt injection patterns.
[COMMAND_EXECUTION]: The skill operates by executing a local Python script with various arguments to perform database operations.
Evidence: The workflow described in SKILL.md instructs the agent to run python3 scripts/sql_query_helper.py for schema extraction, optimization analysis, and EXPLAIN plan interpretation.

sql-tutor