Design Agent Claude

Run a focused UI polish or mockup-to-implementation pass through Claude Code using the custom subagent `design-agent-claude`. Default to Opus 4.6 with high effort; fall back to Sonnet 4.6 when Opus is

SKILL.md

GitHub

Design Agent Claude

Use this skill when you, the orchestrating agent, need to launch Claude Code for a visual-first UI pass.

What this skill is for

This is an orchestrator-facing launch skill. It tells the current agent how to route the task, shape the prompt, and choose the right Claude-side agent.

Use it to:

  • choose the correct Claude agent/subagent
  • shape the delegation prompt for visual/design work
  • avoid generic coding-agent prompting when the task is really UI/UX polish

Do not use it by telling the downstream Claude worker to read or activate this skill itself unless the user explicitly wants that. The normal flow is:

  1. You read this skill.
  2. You launch Claude correctly.
  3. The Claude worker does the task using its own agent instructions.

Preferred agent

On Claude Code, prefer the custom subagent:

  • design-agent-claude

Model guidance

  • Default: Opus 4.6 with high effort for ambiguous or high-stakes visual work.
  • Fallback: Sonnet 4.6 when Opus is unavailable, rate-limited, or more power than the task needs.
  • Claude Code treats model as preference rather than hard enforcement, so keep the prompt explicit when you really want Opus-first behavior.
  • Inspect screenshots or real UI before making design claims.

Recommended prompt shape

"Use design-agent-claude for a UI polish pass on <surface>. Inspect the current UI or screenshots first, state the design intent, make the smallest high-leverage improvements, then report visual QA findings, interaction QA findings, and remaining weaknesses."

Output checklist

  • Design intent summary
  • Top UI/UX issues found
  • Concrete changes recommended or applied
  • Breakpoint and interaction QA notes
  • Remaining risks / next pass