The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for implementing HTTP actions (webhooks) and search functionality, which serve as ingestion points for external, potentially untrusted data.
Ingestion points: Data received via httpAction handlers in convex/http.ts and data indexed for searchIndex or vectorIndex operations.
Boundary markers: Not explicitly defined in the provided implementation rules.
Capability inventory: Handlers have access to ctx.db for database writes, ctx.scheduler for background task execution, and ctx.storage for file operations.
Sanitization: The instructions do not explicitly mandate input validation or sanitization for these entry points, representing a standard application security boundary rather than a malicious pattern.

convex-runtime