cloudflare-workers-ai
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes implementation templates for RAG (Retrieval Augmented Generation) and vision models that process untrusted external data.
- Ingestion points:
templates/ai-embeddings-rag.tsretrieves data from a Vectorize database, andtemplates/ai-vision-models.tsperforms network fetches using user-provided URLs. - Boundary markers: While
templates/ai-embeddings-rag.tsincludes instructions to the model to use specific context, the templates lack explicit sanitization or escaping of the ingested content before interpolation into prompts. - Capability inventory: The skill templates demonstrate the ability to perform network requests (
fetch), execute LLM inference (env.AI.run), and write to persistent storage (env.BUCKET.put). - Sanitization: The
references/best-practices.mdfile correctly identifies the need for input sanitization and provides examples, but these practices are not enforced in the primary functional templates, leaving a vulnerability surface for developers who use the templates as-is.
Audit Metadata