The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to initialize the local environment, including npm install, npm run build, and npm run dev. It also instructs the agent to host the development server on 0.0.0.0, which makes the service accessible to other devices on the same local area network (LAN).
[EXTERNAL_DOWNLOADS]: During initialization, the skill triggers npm install to download dependencies from the public npm registry. Notably, the provided package.json specifies version numbers for several packages (e.g., vite@^8.0.12, typescript@~6.0.2) that are currently significantly higher than those available in stable releases.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted external data, such as screenshots and public URLs, to extract design tokens and components. While the instructions include explicit sanitization rules to replace sensitive customer data with fictional equivalents, malicious content embedded in the source pages or images could still attempt to influence the agent's behavior during the extraction process.
Ingestion points: The Add a design system workflow in SKILL.md accepts screenshots and URLs as input.
Boundary markers: The skill uses instructions like "Treat screenshot content as sensitive by default" and "Replace personal names... with neutral fictional equivalents" to guide the agent.
Capability inventory: The skill uses npm install, npm run build, and npm run dev to manage the local project.
Sanitization: The skill provides a specific list of data types to redact (names, avatars, URLs, emails, etc.) before generating documentation.

design-hub