building-mcp-server-on-cloudflare

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFE
Full Analysis
  • [Indirect Prompt Injection Considerations]: The skill demonstrates how to build tools that accept user-provided data for database queries and API calls. This is a functional requirement that creates an attack surface where maliciously crafted input could influence tool behavior. The skill mitigates this by teaching the use of Zod validation schemas and providing dedicated sanitization logic in its reference guides.
  • Ingestion points: User-defined tool parameters in SKILL.md (e.g., SQL queries and API endpoints).
  • Boundary markers: The skill demonstrates using Zod schemas to enforce data types and constraints.
  • Capability inventory: Tools are granted access to network resources (fetch) and database bindings (env.DB).
  • Sanitization: Explicit sanitization examples for HTML content and URL protocols are provided in references/oauth-setup.md.
  • [Credential Management]: The instructions correctly guide developers to use dedicated secret management tools (wrangler secret put) for sensitive OAuth credentials rather than hardcoding them. It also highlights advanced cookie security measures, such as the __Host- prefix, to protect against session attacks on shared domains.
  • [Trusted Development Workflow]: The skill utilizes official project templates and command-line utilities from well-known services like Cloudflare and the Model Context Protocol organization, ensuring that the development environment is built on verifiable sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 11:51 PM
Security Audit — agent-trust-hub — building-mcp-server-on-cloudflare