groq-deploy-integration
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill correctly implements secret management by using platform-specific secret stores (Vercel environment variables, Google Cloud Secret Manager, and Fly.io secrets) instead of hardcoding sensitive API keys in the source code or container images.
- [INDIRECT_PROMPT_INJECTION]: The skill defines an API endpoint that ingests untrusted user messages and passes them to the Groq LLM API, creating an inherent surface for indirect prompt injection.
- Ingestion points: The
POST /api/chathandler inreferences/implementation.md(Steps 1, 5, and 6) reads user messages directly from the request JSON body. - Boundary markers: No specific delimiters or boundary markers are implemented in the prompt construction to mitigate the risk of user input being interpreted as instructions.
- Capability inventory: The skill provides deployment scripts and a basic inference wrapper; it does not grant the LLM capabilities to execute code, perform network requests, or modify the file system based on its generated output.
- Sanitization: The provided implementation does not include input validation or sanitization for the messages received from the user.
Audit Metadata