The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes patterns for processing external input through middleware.
Ingestion points: The variables clientInput and parsedInput are used to receive raw and validated user data across all files.
Boundary markers: The skill instructs developers to use Zod inputSchema validation before processing data in actions.
Capability inventory: Examples include database operations, logging, and error reporting based on user-supplied data.
Sanitization: Zod validation is used as a standard sanitization mechanism for incoming input.
[DATA_EXFILTRATION]: The skill documents methods for managing sensitive information like API keys and user sessions.
Evidence: The auth-patterns.md file demonstrates extracting API keys from headers and validating session tokens.
Context: These patterns represent standard authorization workflows and do not include hardcoded credentials or unauthorized data transmission.
[EXTERNAL_DOWNLOADS]: The skill refers to well-known packages and monitoring services.
Evidence: Documentation includes examples for using @sentry/nextjs for error tracking and zod for data validation.
Context: These are established, trusted dependencies used for application observability and safety within the intended use-case.

safe-action-middleware