The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to the agent to 'Announce at start' and 'Design for PostgreSQL 18's strengths.' These are standard operational instructions for specialized skills and do not attempt to bypass safety guidelines or override the system prompt.
[COMMAND_EXECUTION]: The skill configuration allows for the use of the Bash tool. This is a common and expected capability for a tool intended to interact with database CLI utilities (like psql) or migration runners. The instructions within the skill focus on generating valid SQL and do not provide any malicious shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill defines a vulnerability surface by processing external files related to database migrations and models.
Ingestion points: Files matching patterns such as **/migrations/**, **/*schema*.sql, and **/models/** defined in SKILL.md.
Boundary markers: The skill suggests using a dedicated DATABASE_ARCHITECTURE markdown block to encapsulate its architectural summaries.
Capability inventory: The skill utilizes Bash and mcp__github__* tools for file and system interaction as listed in the allowed-tools section of SKILL.md.
Sanitization: The skill mandates the use of specific SQL templates and a multi-step checklist to validate that generated or modified schemas adhere to architectural standards.
[DATA_EXFILTRATION]: There is no evidence of network operations to untrusted domains, hardcoded secrets, or attempts to read sensitive system configuration files like SSH keys or environment variables.

database-architecture