The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (start-server.sh, stop-server.sh) and a Node.js server (server.cjs) to provide a browser-based "Visual Companion" UI for design mockups.
The startup script uses standard process management tools (ps, kill) to monitor the parent agent process and ensure the server shuts down automatically when no longer needed.
The WebSocket server binds to the local interface (127.0.0.1) by default and serves static content from a session-specific directory.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it reads project files, documentation, and commit history to establish context.
Ingestion points: Project context is explored by reading existing codebase files and documentation into the agent's context in the first step of the workflow.
Boundary markers: None explicitly defined in the prompt interpolation; however, the skill relies on natural language processing of the ingested data.
Capability inventory: The skill has the capability to write design specifications to the disk and trigger implementation planning via the writing-plans skill.
Sanitization: No explicit sanitization of ingested file content is performed.
Mitigation: The skill includes several "Hard Gates" and "User Review Gates" that mandate human approval of the design specification before any implementation code or plans are generated, effectively neutralizing the risk of autonomous malicious action derived from untrusted project content.

brainstorming