The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines patterns for handling untrusted external data within a web application context, establishing a surface for indirect prompt injection where malicious instructions could be embedded in user-supplied data.
Ingestion points: Untrusted data enters the application through formData in Server Actions (app/actions.ts), JSON payloads in API Route Handlers (app/api/projects/route.ts), and URL search parameters in Client Components (app/projects/search-bar.tsx).
Boundary markers: The provided code examples do not utilize specific natural language boundary markers or instructions to the model to ignore embedded commands within the processed data.
Capability inventory: The skill utilizes capabilities including database operations (db.project.create, db.project.delete), cache invalidation (revalidatePath, revalidateTag), and client-side redirection (redirect).
Sanitization: The skill includes a proactive mitigation by recommending the use of schema validation libraries like zod to parse and validate all incoming data before it is processed by server-side logic.

nextjs