rag

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly states it "Fetches documents from vector databases (e.g., Pinecone, FAISS)" (Key Capabilities) and shows examples where retrieved, potentially untrusted, third-party content is integrated into prompts ("Query -> Retrieve -> Augment -> Generate"), which allows indirect prompt injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches external documents at runtime (e.g., the configured API endpoint https://api.pinecone.io) and injects those retrieved documents directly into model prompts, so external content can directly control agent instructions.