The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses terminal commands to create and manage local file structures for design variants.
Evidence: Use of mkdir -p to create subdirectories under the sketches/ folder.
Evidence: Instructions to use platform-specific commands like open, xdg-open, and start to view the generated HTML artifacts.
[EXTERNAL_DOWNLOADS]: The skill incorporates external assets to enhance the visual fidelity of mockups.
Evidence: References the Tailwind CSS CDN (https://cdn.tailwindcss.com) in the generated HTML templates.
[REMOTE_CODE_EXECUTION]: The attribution and documentation sections mention external tools for advanced workflows.
Evidence: Mentions the installation and usage of the get-shit-done-cc package via npx from the gsd-build repository.
[PROMPT_INJECTION]: The skill features an indirect prompt injection surface as it processes user input to generate executable browser content.
Ingestion points: Captures user input for 'Feel', 'References', and 'Core action' during the intake phase.
Boundary markers: None present; user input is directly used to shape the content and logic of the generated HTML.
Capability inventory: Uses write_file to create the HTML/JS and browser_navigate / browser_vision to execute and inspect it.
Sanitization: No explicit sanitization or validation of the user-provided descriptions is specified before they are interpolated into the generated code.

sketch