The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill demonstrates patterns for processing user messages and external data (via RAG or tool calling) within AI model prompts. This is a foundational capability for AI agents, and the provided examples show the use of structured roles (system, user, assistant) to help manage the interaction. Implementing robust boundary markers and validation when integrating untrusted data remains a best practice for developers.
Ingestion points: Data enters the agent via onMessage (WebSocket input), onChatMessage (NL strings), and ingestDocument (vector data ingestion) as seen in SKILL.md and references/agent-patterns.md.
Boundary markers: Code examples utilize the chat completion API structure with roles (system, user, assistant) to distinguish between instructions and data.
Capability inventory: The skill utilizes this.env.AI.run for model inference, this.sql for database operations, and fetch for external API integration within tools.
Sanitization: The documentation correctly emphasizes the use of tagged template literals for this.sql queries to prevent SQL injection and uses JSON.parse for handling message payloads.
[Platform Tooling and Project Setup]: The instructions include standard commands for the Cloudflare environment, such as installing the Wrangler CLI and using npm create cloudflare to bootstrap new projects. These are the recommended workflows for deploying and managing services on the Cloudflare edge network.
[Trusted External Resources]: The skill frequently references official documentation and open-source repositories managed by Cloudflare. These links provide developers with access to the latest SDK updates, API references, and verified implementation examples.

building-ai-agent-on-cloudflare