building-ai-agent-on-cloudflare
Pass
Audited by Gen Agent Trust Hub on Mar 26, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill documents patterns for agents to ingest user-supplied messages via WebSockets and process them using LLMs (e.g., in SKILL.md and references/agent-patterns.md). This architecture presents a surface for indirect prompt injection, where untrusted input could potentially influence the agent's logic or tool usage.
- Ingestion points: onMessage handlers in SKILL.md and references/agent-patterns.md.
- Boundary markers: Code examples use system prompts to define roles, though explicit delimiters for user input are not always shown.
- Capability inventory: Agents can perform SQL queries (this.sql), network requests (fetch), and call external tools.
- Sanitization: Documentation in references/troubleshooting.md specifically advises using tagged templates for SQL to prevent injection attacks.
- [EXTERNAL_DOWNLOADS]: Instructions include bootstrapping projects using the official Cloudflare CLI (npm create cloudflare@latest) and reference implementation examples from Cloudflare's official GitHub repositories. These are recognized as trusted sources for developer tooling.
- [COMMAND_EXECUTION]: The documentation provides examples for using standard development tools such as npm and the wrangler CLI for project management and deployment to the Cloudflare platform.
Audit Metadata