claude-design
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill orchestrates local project analysis and browser automation by executing scripts such as analyze_codebase.py, gather_references.py, collect_source.py, and drive_claude_design.cjs. These scripts use standard system commands like git for metadata retrieval.
- [DATA_EXFILTRATION]: The skill synthesizes project information (design tokens, component names, UI text, and image assets) and transmits it to claude.ai/design. This behavior is the core functionality of the skill and is explicitly disclosed to the user.
- [PROMPT_INJECTION]: The skill processes untrusted content from the user's codebase, creating a surface for indirect prompt injection. The instructions mitigate this risk by commanding the agent to treat extracted content as data only and ignore any embedded directives.
- [REMOTE_CODE_EXECUTION]: The skill manages browser automation via Playwright and Node.js. It directs the installation of these tools and provides a controlled environment for browser interaction, requiring human review for login and final submission.
Audit Metadata