The Agent Skills Directory

[SAFE]: The skill explicitly enforces the use of parameterized queries and includes a specific evaluation case (unsafe-input) to reject unsafe string concatenation, mitigating SQL injection risks.
[SAFE]: Multi-tenant security is prioritized; the instructions and evaluation case multi-tenant-guard require that access-control filters never be removed or bypassed for performance gains.
[SAFE]: The skill mandates the protection of sensitive data, instructing agents to avoid logging PII or secrets and to use generic entities in all public-facing artifacts.
[SAFE]: Operational safety rules are defined, such as requiring authorization for destructive DDL or production migrations and recommending rollback plans for all changes.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted data from a user's codebase, schemas, and database logs as part of its optimization workflow.
Ingestion points: Project source code, database schemas, migration files, and query logs (referenced in references/project-discovery.md).
Boundary markers: The instructions do not specify strict technical delimiters for external data, but emphasize "inspect before prescribing" and "evidence vs assumptions" as reasoning guards.
Capability inventory: The skill is designed to propose and implement database query and schema modifications if the agent has file-write permissions.
Sanitization: No specific sanitization or escaping of ingested data is mentioned in the discovery instructions.

database-query-patterns