The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages a local Node.js environment by executing shell scripts (start-server.sh, stop-server.sh) and a server script (server.cjs). This infrastructure supports a 'Visual Companion' that renders HTML mockups. The server implementation is defensive, defaulting to the local loopback interface (127.0.0.1) and employing path.basename to prevent directory traversal when serving files from the session directory.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks common to agents that analyze local project environments.
Ingestion points: The agent reads project files, documentation, and commit history during the 'Explore project context' phase defined in SKILL.md.
Boundary markers: The instructions lack explicit delimiters or 'ignore' instructions for content ingested from the project files.
Capability inventory: The skill possesses the ability to execute local scripts, write new documentation files to the disk, and trigger downstream implementation planning skills.
Sanitization: No specific filtering or escaping mechanisms are defined for data read from the local file system before it enters the agent's context.
[SAFE]: The background server includes significant safety features, such as a 30-minute idle timeout and a watchdog that terminates the server if the parent process (the agent) exits. The skill also enforces a 'Hard Gate' requiring explicit user approval of a design before any implementation tools can be used.

brainstorming