The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for creating MCP tools that ingest data from external APIs or user-supplied strings, creating a potential attack surface.
Ingestion points: The get_weather tool (SKILL.md, line 98) and query_db tool (SKILL.md, line 147) ingest external and user input respectively.
Boundary markers: The provided code snippets do not implement specific delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill leverages fetch for network requests and wrangler for deployment of generated code.
Sanitization: Although the database example uses direct interpolation, the skill includes a dedicated section on input validation using the Zod library (SKILL.md, lines 122-132).
[EXTERNAL_DOWNLOADS]: Fetches configuration and project templates from Cloudflare's official resources.
Downloads project templates from the cloudflare/ai repository using npm create cloudflare@latest.
References official documentation from developers.cloudflare.com and modelcontextprotocol.io.
[COMMAND_EXECUTION]: Utilizes CLI tools for project management and deployment.
Executes wrangler for deploying Worker scripts and managing secrets.
Uses npx @modelcontextprotocol/inspector to test the local MCP server environment.

building-mcp-server-on-cloudflare