The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell scripts (start-server.sh, stop-server.sh) to manage a local Node.js web server. It invokes the node runtime and uses process management commands like ps and kill to handle the server lifecycle.
[DATA_EXFILTRATION]: The skill initiates a local HTTP/WebSocket server to display design mockups. The server binds to 127.0.0.1 by default, which restricts access to the local machine. It serves files only from a designated session directory and implements path sanitization using path.basename to prevent directory traversal attacks.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its context-gathering behavior.
Ingestion points: The skill is instructed to explore project context by reading files, documentation, and recent commits (SKILL.md).
Boundary markers: The skill lacks explicit instructions to wrap project content in delimiters or to ignore instructions embedded within the processed data.
Capability inventory: The agent can write design specifications to the filesystem, execute shell scripts for server management, and dispatch subagents for spec reviews.
Sanitization: There is no evidence of sanitization or validation performed on the ingested project data before it is processed by the model.

brainstorming