The Agent Skills Directory

[DATA_EXPOSURE]: The skill describes the mechanism for securing the application's control channel using a local token file stored at ~/.local/share/solidcraft/control/PORT.token. It explicitly instructs the user or agent not to share this token in public communications such as issues or pull requests to prevent unauthorized local access to the running application.
[COMMAND_EXECUTION]: Instructions are provided for running the application using cargo, managing virtual framebuffers with Xvfb, and generating security tokens via openssl. These commands are standard for the development and headless operation of the project.
[INDIRECT_PROMPT_INJECTION]: The skill describes an architecture where the agent processes external data via a control protocol and scripts. It identifies that the system implements security boundaries for these tools, specifically enforcing path resolution to prevent directory traversal attacks (e.g., blocking .. or symlinks that exit allowed directories).
Ingestion points: TCP control channel (127.0.0.1:PORT) and local script files (script.json).
Boundary markers: The protocol uses a structured JSON-based command format.
Capability inventory: File writes (ui.screenshot), design manipulation (solid.extrude), and application state control (ui.start).
Sanitization: Explicit path validation and resolution are performed before file access to ensure tools stay within permitted folders.

solidcraft-agent-tools