The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (start-server.sh, stop-server.sh) and a Node.js process (server.cjs) to facilitate interactive UI mockups and architecture diagrams.
[PROMPT_INJECTION]: The skill captures user interaction data from the web companion to inform the design process, which introduces a surface for indirect prompt injection.
Ingestion points: User choices and click events are recorded to the local state directory and processed by the agent.
Boundary markers: The event log uses structured JSON lines but lacks explicit boundary markers for untrusted content.
Capability inventory: The agent can write specification documents to the filesystem and initiate project implementation via subsequent skills.
Sanitization: The server script uses path.basename to validate file requests and prevent path traversal when serving mockup assets.

brainstorming