The Agent Skills Directory

[SAFE]: No security issues detected. The skill utilizes a local server for UI purposes, which is implemented with security best practices including path sanitization and restricted network binding.
[DATA_EXFILTRATION]: The skill launches a local Node.js server to provide a visual interface. This server binds to localhost (127.0.0.1) by default and includes path validation using path.basename to prevent directory traversal or unauthorized file access beyond the session content directory.
[PROMPT_INJECTION]: The skill includes an interaction loop where user-driven browser events (selections and clicks) are stored as structured JSON and read by the agent.
Ingestion points: Structured event data from state_dir/events recorded by the local server.
Boundary markers: Not explicitly present; however, the agent is instructed to merge this data with terminal text to understand user preferences.
Capability inventory: Includes file system writes (design documents) and git operations, all of which are gated by mandatory user approval steps in the core skill process.
Sanitization: The server uses JSON.parse to validate the structure of interaction events, and the agent's instructions focus on using these for design confirmation rather than direct command invocation.

brainstorming