The Agent Skills Directory

[COMMAND_EXECUTION]: The provided Python template in templates/connection_pool.py uses f-string interpolation to construct SQL commands, specifically in the _init_connection function: await conn.execute(f"SET statement_timeout = '{config.statement_timeout_ms}'"). While intended for configuration, this pattern creates a SQL injection vulnerability if the configuration object is populated from untrusted user input without strict validation.
[REMOTE_CODE_EXECUTION]: In templates/connection_pool.py, the check_sqlalchemy_engine_health function utilizes the __import__ call to dynamically load the sqlalchemy module. Although the module name is hardcoded as a static string, dynamic loading is a pattern often monitored as it can be used to bypass static analysis or facilitate the execution of arbitrary code.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as database schemas, migration files, and SQL queries. It lacks boundary markers or explicit instructions (e.g., 'ignore embedded instructions') to mitigate 'Indirect Prompt Injection' (Category 8), where malicious SQL comments or data payloads could attempt to hijack the agent's logic during analysis.
Ingestion points: Processes SQL files, migration files (Alembic/Django), and database schemas.
Boundary markers: Absent in instructions.
Capability inventory: Capability to execute SQL via op.execute and conn.execute templates.
Sanitization: No sanitization logic is provided in the Python templates for the configuration inputs.

production-postgres