C — Moderate risk signals Graded October 16, 2026Several concerns worth a closer look before you install.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.
AI agent skill · Developer tools, Design & media
baoyu-design
Run Claude Design locally as an Agent Skill — Cursor, Claude Code & more. Produce polished UI mockups, prototypes, decks & wireframes as self-contained HTML, without claude.ai/design. Best with Opus or Claude.
Bottom line: Signal-based grade (3/7 points): well starred on GitHub, public source code. This reflects adoption and transparency, not a code review.
- Popularity score
- 41/100
- Rank
- #296 of 500
- GitHub stars
- 4.4k
What it does
Baoyu-design enables Claude to generate polished UI mockups, prototypes, presentations, and wireframes as self-contained HTML files directly within agent workflows—bypassing the need for claude.ai/design. It works as an Agent Skill in Cursor, Claude Code, and similar environments, with best results on Claude Opus.
Who it’s for
Designers, product managers, and builders who want to automate UI/UX artifact generation within their development tools or agent pipelines. Teams seeking faster iteration on low-fidelity and high-fidelity prototypes without leaving their coding environment.
Copy-paste setup
Check these steps against the project’s own instructions before running anything.
Verify against the repo README at https://github.com/JimLiu/baoyu-design (opens external site) for exact integration steps. Likely requires cloning the repository and configuring as a local MCP server or Claude Desktop plugin; exact npm/config command patterns should be confirmed in the repo's documentation.
Risks
- Prompt injection: Generated HTML is returned to Claude and may be re-processed; malicious design prompts could inject scripts or exfiltrate context.
- Local file system access: Agent skill may need read/write permissions to save HTML artifacts; misconfigured permissions could expose sensitive project files.
- Unvetted HTML output: Generated HTML is self-contained but not sandboxed; running untrusted designs in browsers poses XSS and client-side attack vectors.
- Dependency chain: Relies on local Claude installation and agent framework; updates or breaking changes upstream could degrade functionality.
- Data retention: No clear documentation on whether design prompts or generated artifacts are logged or retained locally.
How to stay secure
Least-privilege permissions
- Restrict file access: Configure the skill to only read and write to a dedicated project folder; block access to sensitive directories like passwords, financial data, or employee records.
- Use read-only mode when possible: If you're only reviewing designs, disable the skill's ability to save files to your computer.
- Separate user accounts: Run the skill under a limited user account rather than administrator-level access; this prevents accidental damage if something goes wrong.
Network isolation
- Disable cloud sync: Ensure design prompts and generated HTML files are not automatically uploaded to cloud services or external servers.
- Block outbound connections: If your IT team manages your network, request that the skill be restricted from sending data outside your organization's network.
- Avoid public networks: Only use this skill on trusted, corporate networks; never on public WiFi or shared devices where others could intercept your design work.
Monitoring
- Review generated files before opening: Always inspect the HTML code in a text editor before viewing it in a browser; look for suspicious script tags or external links.
- Keep an audit log: Maintain a simple record of which designs were created and when, in case you need to investigate unexpected changes.
- Check local storage: Periodically verify that old design files are being deleted as expected and not accumulating unintended copies on your hard drive.
Red flags to watch for
- Unsolicited scripts or links in designs: If generated HTML contains code that tries to collect data, redirect you to websites, or pop up unexpected windows, stop using that design immediately.
- Designs that request credentials: Never enter passwords or sensitive information into design prompts; the skill should never ask for login details or personal data.
- Unexplained file changes: If files outside your design folder are modified, or if you see new files created without your request, report it to your IT team right away.
Alternatives
- Figma API + Agent Connectors — Programmatic Figma file generation with built-in security controls and team collaboration.
- Claude.ai/design — Official design endpoint; cloud-hosted with Anthropic's security model, no local setup friction.
- Framer AI — Browser-based design-to-code with real-time collaboration and component libraries.
- v0 by Vercel — AI-generated React components and UI with built-in code quality checks and deployment integration.