Cloudflare Platform Products
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill outlines architectural patterns (such as RAG and content management) that ingest untrusted user data and process it through LLMs and other services.
- Ingestion points: User data is ingested through
request.json()andrequest.formData()in several functions withinexamples/multi-product-architecture.js. - Boundary markers: The RAG implementation in
examples/multi-product-architecture.jsincludes a system prompt ("Answer the question using only the provided context") to delimit user queries from knowledge base content, though this is a partial mitigation. - Capability inventory: The examples demonstrate access to sensitive capabilities including relational database writes (
env.DB), object storage writes (env.MEDIA,env.ATTACHMENTS), AI inference (env.AI), and headless browser automation (env.BROWSER). - Sanitization: Input sanitization and validation are not explicitly implemented in the example code snippets.
- [SAFE]: No signs of malicious command execution, data exfiltration, or obfuscation were detected. All external references target official Cloudflare documentation or use standard placeholder domains.
Audit Metadata