rag-architect

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow can ingest and pass retrieved document text into the LLM context (RAG), and those documents are often outsider-authored (e.g., public web content or third-party documents) via the retrieval pipeline’s “retrieved context” step; the provided SKILL.md does not constrain the ingestion source to user-authored content.

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

• Potentially malicious external URL detected (high risk: 0.80). The LateChunker calls AutoModel.from_pretrained("jinaai/jina-embeddings-v2-base-en", trust_remote_code=True), which will fetch and potentially execute remote model repo code from the Hugging Face model page (e.g. https://huggingface.co/jinaai/jina-embeddings-v2-base-en) at runtime and the skill depends on that load.