The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone reference code and examples from the official repository at https://github.com/modelcontextprotocol/ext-apps.git. This is used to provide access to JSDoc documentation and framework templates for integration.
[COMMAND_EXECUTION]: The instructions include several shell commands for development setup:
git clone to fetch the SDK and examples into a temporary directory.
npm install for various Model Context Protocol vendor packages and development dependencies like vite and zod.
npm run build which executes vite and tsc to bundle the UI and compile server code.
[INDIRECT_PROMPT_INJECTION]: The skill defines an architecture for creating interactive UIs that process and display data originating from tool calls.
Ingestion points: Data enters the UI context via app.ontoolinput and app.ontoolresult handlers in the frontend script (mcp-app.ts).
Boundary markers: The guidance recommends using zod for strict schema validation of tool inputs.
Capability inventory: The resulting servers are capable of reading local files (fs.readFile) to serve the UI and can be configured for network access.
Sanitization: The skill provides explicit instructions for implementing a Content Security Policy (CSP) to restrict domains for network connections, scripts, and frames, which helps mitigate risks associated with processing untrusted tool data.

add-app-to-server