The Agent Skills Directory

[SAFE]: The skill leverages the official DESIGN.md CLI from the Google organization via npx for validation and export operations. As Google is a trusted organization, this remote dependency is considered safe within the context of the skill's intended use.
[SAFE]: The skill includes a local Python script (check-contrast.py) for validating WCAG color contrast. The script is restricted to standard Python libraries and does not perform network operations or access sensitive system paths.
[SAFE]: A local-only preview server (preview-server.ts) implemented in Bun allows users to inspect design changes. The server is bound to 127.0.0.1 and contains explicit path traversal checks (isInsideSessionDir) to ensure it only serves files from the designated session directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files and external reference URLs, which constitutes an indirect prompt injection surface.
Ingestion points: Reads project documents like PRODUCT.md and PRD.md in the discovery workflow and fetches external URLs during design authoring.
Boundary markers: SKILL.md contains explicit instructions to ignore directives embedded in comments, strings, or metadata.
Capability inventory: The skill has permissions to read, write, and edit files, and can execute shell commands via bun, npx, and python3.
Sanitization: The skill employs strict YAML schema validation for design tokens and uses structural placeholders in generated content to prevent command or prompt leakage.

design-brief