start-middleware
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns that allow the ingestion of untrusted data from the client using the
sendContextmechanism. - Ingestion points: Untrusted data enters the server context via the
.server()phase of middleware as shown inreferences/templates/auth-authorization.mdandreferences/context.md. - Boundary markers: The instructions do not specify LLM-level delimiters or instruction-ignore warnings, though they recommend application-level validation.
- Capability inventory: The scripts do not contain high-risk capabilities such as arbitrary shell execution or direct file system writing based on the untrusted data.
- Sanitization: The skill enforces security best practices by requiring Zod schema validation (
z.string().uuid().parse) and server-side membership verification before processing any client-sent identifiers.
Audit Metadata