The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local repository (code, TODOs, comments) and external sources (research, issues, URLs) to drive the design process. Malicious content within these files could potentially influence the agent's output or subsequent actions.
Ingestion points: Reads repository files (routes, components, copy), user-provided screenshots, and external research/issues.
Boundary markers: Lacks explicit delimiting or 'ignore' instructions when processing repository content.
Capability inventory: Has the ability to write multiple files to disk, execute shell commands for local servers, and register MCP servers.
Sanitization: No explicit sanitization of ingested content before use in analysis or code generation.
[COMMAND_EXECUTION]: The skill utilizes shell commands to facilitate the design workflow, such as serving prototypes via python3 -m http.server or npx serve, and managing MCP tools using claude mcp add and npx agentation-mcp doctor.
[REMOTE_CODE_EXECUTION]: The skill involves the execution of the agentation-mcp package from the npm registry using npx, which downloads and runs the server code locally.
[EXTERNAL_DOWNLOADS]: The generated prototypes and process maps fetch external JavaScript libraries and CSS from public CDNs, including mermaid from cdn.jsdelivr.net and react, react-dom, and agentation from esm.sh.
[DYNAMIC_EXECUTION]: The skill dynamically generates a functional HTML/JavaScript prototype (prototype/index.html) based on analyzed project data and design tokens. It iteratively modifies this code based on feedback received via the Agentation tool.

co-design