The Agent Skills Directory

[SAFE]: The skill instructions and examples align with official Cloudflare developer documentation. It correctly advises using the env object for secrets and environment variables, which is a security best practice for the platform.
[COMMAND_EXECUTION]: The instructions include standard developer commands such as npx wrangler dev for local testing and wrangler tail for debugging. These are official tools provided by Cloudflare for their edge computing platform and are used here for their intended purposes.
[PROMPT_INJECTION]: The skill provides patterns for processing Request objects, which inherently introduces a surface for indirect prompt injection if untrusted request data (such as headers or body content) is subsequently processed by an LLM without sanitization. This is a common characteristic of web-handler documentation and is noted as a standard architectural surface rather than a malicious pattern.
Ingestion points: fetch handler parameters (request) in SKILL.md examples.
Boundary markers: Not included in the minimal code snippets.
Capability inventory: Access to environment secrets, Cloudflare KV storage, and outbound network requests via fetch.
Sanitization: No explicit sanitization of request data is shown in the basic examples.

cloudflare-workers-expert