The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-controlled data from URL search parameters into the application context.
Ingestion points: Data enters the agent and application context through the useQueryState, useQueryStates, and createSearchParamsCache tools as documented in SKILL.md and the references/ directory.
Boundary markers: The skill recommends using typed parsers and withDefault fallbacks to constrain parameters to expected values, providing clear boundaries for state.
Capability inventory: The managed state influences UI state, pagination, and client-side logic, which could be leveraged to affect application behavior if input is unvalidated.
Sanitization: The skill provides rigorous guidance on sanitizing input using Standard Schema libraries (e.g., Zod, Valibot) and custom type guards to handle potentially malicious JSON or enum inputs from the URL, specifically in references/parser-json-validation.md and references/state-standard-schema.md.

nuqs-v2