The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing external input through query arguments, creating a potential surface for indirect injection if data is later used in unsafe ways.
Ingestion points: Query handlers defined in SKILL.md and references/query-guidelines.md accept args from external callers.
Boundary markers: The skill explicitly mandates the use of argument validators (e.g., v.string(), v.id()) to enforce schemas on all inputs.
Capability inventory: Query handlers are restricted to database read operations (ctx.db.query) and calling other queries.
Sanitization: Input sanitization is performed via the framework's type-safe validation system, which is strongly recommended throughout the documentation.
[SAFE]: The documentation includes security-positive recommendations, such as using internalQuery to restrict function visibility and observing database read limits to prevent resource exhaustion.

convex-queries