The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of node-html-parser from the official NPM registry to facilitate HTML parsing within its generated scripts.
[COMMAND_EXECUTION]: Utilizes node to run generation scripts, gh (GitHub CLI) for Gist creation and secret management, and jq for JSON validation during the local and CI sync phases.
[DYNAMIC_EXECUTION]: During the setup phase, the agent generates custom JavaScript walkers and DTCG converters using templates provided in the skill's reference files. These scripts are then executed to process local project artifacts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data that could potentially contain malicious instructions.
Ingestion points: The generated walk-<site>.mjs script reads compiled HTML and CSS files (e.g., index.html, tailwind.css) from the build output directory.
Boundary markers: None identified in the script templates.
Capability inventory: Includes local script execution (node), file system read/write access for configuration and artifacts, and network operations via fetch to update GitHub Gists.
Sanitization: The skill parses unstructured HTML and CSS into a strictly defined JSON schema using node-html-parser and specific regex-based resolution logic, minimizing the risk of prompt interpolation.

code-to-figma