start-middleware

Pass

Audited by Gen Agent Trust Hub on Aug 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates Indirect Prompt Injection through the sendContext mechanism which allows untrusted data to move from the client to the server environment.\n
  • Ingestion points: The sendContext property in createMiddleware (detailed in references/context.md and SKILL.md) is a direct entry point for client-supplied data into the agent's server-side context.\n
  • Boundary markers: The documentation explicitly includes strong warnings and "Critical Rules" advising developers that sendContext is untrusted and must be verified against server-trusted sessions (e.g., cookies or database records).\n
  • Capability inventory: Middleware scripts have the capability to perform database operations, modify request headers, and influence the execution flow of downstream server functions based on this untrusted data.\n
  • Sanitization: The skill provides templates using the zod library for schema validation and demonstrates database-level authorization checks to mitigate the risks associated with processing client-sent identifiers.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 23, 2026, 04:06 PM
Security Audit — agent-trust-hub — start-middleware