The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the capability to fetch data from external APIs via HTTP and persist it in database tables. This architecture creates a vulnerability surface for indirect prompt injection, as untrusted data from an external service could be incorporated into the database state.
Ingestion points: Data is ingested through ctx.http.fetch and similar methods in the provided code examples within references/functions-procedures.md.
Boundary markers: The examples do not demonstrate the use of delimiters or explicit boundary markers to separate untrusted external content from agent instructions.
Capability inventory: Procedures have access to network operations (ctx.http), database read/write operations (ctx.withTx), and the ability to trigger other reducers.
Sanitization: The documentation does not provide examples of sanitization or validation for API responses before they are stored in the database.
[EXTERNAL_DOWNLOADS]: Code examples demonstrate procedures that accept a user-supplied url argument used directly in HTTP requests. This pattern introduces a potential surface for Server-Side Request Forgery (SSRF) if the input is not validated by the end user's implementation.

spacetimedb-procedures