The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download external code from an unverified source during the local development installation process.
Evidence: git clone https://github.com/bergside/design-md-chrome.git in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites and generates markdown files intended to be consumed by AI coding agents, creating a potential surface for indirect prompt injection.
Ingestion points: The extractDesignTokens function in the extension code iterates through all elements on a webpage using document.querySelectorAll('*') and reads their computed styles.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded natural language in the generated DESIGN.md or SKILL.md examples provided in the skill documentation.
Capability inventory: The skill generates markdown files which are specifically designed to be integrated into AI agent contexts (e.g., .claude/DESIGN.md, .cursorrules/design-system.md) to guide their coding behavior.
Sanitization: While the code performs normalization and deduplication of tokens (e.g., rounding pixels, converting colors to hex), it does not appear to sanitize style values for embedded natural language instructions that could override agent behavior.

design-md-chrome-extractor