cohere-deploy-integration
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes boilerplate code for a streaming chat endpoint that processes external user input. While standard for chat applications, this creates a potential surface for indirect prompt injection if the LLM is not properly constrained.
- Ingestion points: The
messagefield from the JSON request body inapi/chat/route.ts. - Boundary markers: Absent in the example code.
- Capability inventory: Network communication with the Cohere API.
- Sanitization: Absent in the provided code snippets.
- [SAFE]: The skill correctly implements security best practices for secret management by using platform-specific CLI tools (
vercel env,fly secrets,gcloud secrets) rather than hardcoding credentials or environment variables in plain text. - [SAFE]: All external service references target well-known and trusted infrastructure providers including Vercel, Fly.io, and Google Cloud Platform for deployment operations.
Audit Metadata