The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of SQL commands against remote databases via the pymssql library. While it implements strict regex-based validation in scripts/query.py (is_read_only and validate_single_statement) to restrict commands to read-only operations like SELECT and SP_HELP, the ability to execute database queries is a significant capability that relies on client-side controls.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays data fetched from external SQL Server databases, creating a surface for indirect prompt injection.
Ingestion points: Database query results are fetched and processed in the execute_query function within scripts/query.py at the cursor.fetchmany(MAX_ROWS) call.
Boundary markers: The skill prints result sets to the console with column separators (" | ") but does not wrap external data in specific delimiters or include instructions for the agent to ignore embedded commands within the data.
Capability inventory: The agent can connect to databases, list tables, view schemas, and execute arbitrary read-only queries as defined in scripts/query.py using the pymssql package.
Sanitization: The script implements a MAX_ROWS limit (10,000) and a MAX_COLUMN_WIDTH cap (100 characters) to prevent denial-of-service or context overflow, but it does not perform semantic sanitization of the database content to filter for potential prompt injection patterns.

mssql