The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation encourages the installation of the 'get-shit-done-cc' package using the command npx get-shit-done-cc --hermes --global. Executing unverified packages from third-party sources via npx facilitates remote code execution.
[COMMAND_EXECUTION]: The skill instructs the agent to use the terminal tool to create project directories and recommends using shell commands (open, xdg-open, start) to launch the host browser and open locally generated files.
[EXTERNAL_DOWNLOADS]: The generated HTML variants are configured to load the Tailwind CSS library from an external content delivery network (https://cdn.tailwindcss.com).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input and interpolates it directly into generated HTML files, creating a cross-site scripting (XSS) attack surface.
Ingestion points: User-provided answers for the "Feel", "References", and "Core action" prompts in the Intake section of SKILL.md.
Boundary markers: None. Content is embedded directly into the HTML body and README files.
Capability inventory: The agent uses write_file to create mockups, browser_navigate to view them, and terminal to launch them in external browsers.
Sanitization: The instructions do not include any steps to escape or sanitize user-provided strings before they are written into the HTML structure.

sketch