The Agent Skills Directory

[SAFE]: The code templates correctly implement database interactions using parameterized queries ($1, $2, etc.) with the asyncpg library, which is a best practice for preventing SQL injection.
[SAFE]: The skill utilizes well-known, industry-standard libraries including asyncpg and elasticsearch for its implementation examples.
[INDIRECT_PROMPT_INJECTION]: The skill defines a pattern for ingesting data from an external event store which represents a potential attack surface. 1. Ingestion points: External data enters the system through the event.data field within the apply method of the Projection templates in SKILL.md. 2. Boundary markers: None are defined in these implementation templates. 3. Capability inventory: The templates demonstrate capabilities for database modification (PostgreSQL) and search index management (Elasticsearch). 4. Sanitization: The provided code prevents technical injection (SQLi) via parameterization, though it does not explicitly filter content for prompt injection instructions.

projection-patterns