The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates (such as templates/rag-agent.ts and templates/websocket-agent.ts) that process data from external requests and messages, which is then used in prompts for LLMs. This ingestion of untrusted data creates a potential surface for indirect prompt injection.
Ingestion points: onRequest and onMessage handlers across multiple templates, including those for RAG, web scraping, and chat applications.
Boundary markers: The templates demonstrate standard prompt interpolation but lack explicit boundary markers or safety instructions to prevent the model from following commands embedded in external content.
Capability inventory: Templates include file-system access (SQLite), network operations (AI API calls), and web browsing capabilities via Puppeteer.
Sanitization: The templates do not implement specific sanitization or filtering for the external content before it is processed by the AI models.
[EXTERNAL_DOWNLOADS]: The skill references and downloads resources from well-known and trusted providers.
Project scaffolding is performed via Cloudflare's official npm create templates.
Dependencies are managed through standard registries (NPM) and target official packages from Cloudflare, Vercel, and major AI providers.
[COMMAND_EXECUTION]: Instructions are provided for using standard development CLI tools.
The documentation guides the user to use npm, npx, and wrangler for project setup, dependency management, and deployment within the Cloudflare ecosystem.

cloudflare-agents