The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides code templates for tools that ingest data from external APIs or take raw SQL input, creating a surface for indirect prompt injection. If the external data sources are compromised or the input is manipulated, the agent could be tricked into performing unauthorized database operations or data exfiltration.
Ingestion Points: In SKILL.md, the get_weather tool fetches data from an external API, and the query_db tool accepts a raw sql string input.
Boundary Markers: The basic tool registration examples do not include explicit instructions or markers to distinguish between system instructions and untrusted data.
Capability Inventory: The tools demonstrate significant capabilities, including network access via fetch() and database manipulation via env.DB in SKILL.md.
Sanitization: Although the skill provides a dedicated security reference (references/oauth-setup.md) covering input sanitization, the primary tool examples in SKILL.md omit these protections for brevity.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to download project templates from Cloudflare's official demo repositories using the npm create cloudflare@latest command. These templates are sourced from Cloudflare, a well-known service provider, and are used for initial project scaffolding.
[COMMAND_EXECUTION]: The instructions include standard developer commands to manage the lifecycle of the MCP server, such as npm start for local testing, npx wrangler deploy for production deployment, and npx @modelcontextprotocol/inspector for debugging.

building-mcp-server-on-cloudflare