The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess.run to call external database utilities like psql, mysql, and sqlcmd. While it avoids shell injection by passing arguments as lists, the ability to execute commands based on user-provided database configurations and file content is a significant capability that requires careful usage.
Evidence: Found in scripts/database_connection_test.py and scripts/stored_procedure_deployer.py where database commands are constructed and executed.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes SQL files and database outputs which could contain adversarial instructions.
Ingestion points: stored_procedure_deployer.py reads local SQL files; database_connection_test.py captures output from version and permission queries.
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts were found in the skill's logic.
Capability inventory: The skill can perform file writes and execute database commands via shell-level clients.
Sanitization: No sanitization or filtering for LLM injection patterns is implemented in the validation or deployment scripts.
[CREDENTIALS_UNSAFE]: The provided scripts for testing and deployment accept database passwords and pass them to client tools. In the case of MySQL and SQL Server, passwords are passed as command-line arguments, which is a known security risk as it can expose credentials in system process listings.
Evidence: See the check_mysql and check_sqlserver functions in both scripts/database_connection_test.py and scripts/stored_procedure_deployer.py.

generating-stored-procedures