The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes instructions for executing local scripts such as scripts/init_mcp_app.py and common development utilities like npx cloudflared for local tunnel creation during testing.
[EXTERNAL_DOWNLOADS]: Projects generated by this skill depend on standard Node.js packages from the official Model Context Protocol ecosystem and well-known libraries such as Express and Vite.
[PROMPT_INJECTION]: The skill establishes an interaction layer where UI components can process data from server tools and update the AI model's conversation context. This functionality represents an inherent surface for indirect prompt injection.
Ingestion points: Untrusted data enters the UI through the app.ontoolresult callback as demonstrated in the boilerplate logic.
Boundary markers: The UI runs within a sandboxed iframe on the host side, providing a core security boundary.
Capability inventory: UI components can trigger server-side tool execution via app.callServerTool and modify the host's context via app.updateModelContext.
Sanitization: The provided boilerplate code includes an escapeHtml utility to sanitize data before rendering it in the UI, mitigating potential cross-site scripting (XSS) risks.
[SAFE]: The skill is transparently documented, relies on trusted libraries, and does not contain malicious obfuscation, persistence mechanisms, or unauthorized data exfiltration patterns.

mcp-apps