Vision Judge — Design agent for Claude Code
Use this agent (typically via the vision-verify workflow) to LOOK AT a rendered UI screenshot and judge whether it visibly demonstrates a requirement and is legible — catching defects that code review.
How to install Vision Judge
Installs to ~/.claude/agents/koldovsky-2026-agentic-engineering-crash-course-day03-alternative-vision-judge.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/koldovsky/2026-agentic-engineering-crash-course-day03-alternative/HEAD/.claude/agents/vision-judge.md -o ~/.claude/agents/koldovsky-2026-agentic-engineering-crash-course-day03-alternative-vision-judge.md Restart Claude Code, or start a new session, for it to be picked up.
What Vision Judge does
name: vision-judge description: Use this agent (typically via the vision-verify workflow) to LOOK AT a rendered UI screenshot and judge whether it visibly demonstrates a requirement and is legible — catching defects that code review and DOM assertions are blind to (dead-looking controls, poor color contrast, broken/overlapping layout, wrong-moment frames). Reads the image with fresh eyes; never the agent that built the UI. tools: Read, Grep, Glob
You are a visual QA judge with fresh eye
Alternatives in Design
- UI Verify — 2.1k ★
- Template Cloner — Use this agent whenever Pulkit gives a UI template/website URL in the fable repo and wants it reproduced same 509 ★
- Judge Runner — melta shadow judge の 1 trial を実行する 201 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Design Judge
The judge of the stage-2 review round — rules every finding sustained/deferred/dismissed against the scrutiny
Render Critic
Reviews rendered shots of a 3D character against a reference image or against itself and returns a ranked list
Visual Judge
THE single visual acceptance pass for a rendered deliverable of these types only — pptx, docx, xlsx, pdf, post
Critic Visual
Independent visual judge that looks at rendered screenshots/thumbnails and the color/style bible, and returns
Kimi Challenge
Use this agent when Claude wants Kimi to run an adversarial review that challenges the implementation approach
iOS UX Reviewer
iOS UI/UX review expert grounded in Apple's Human Interface Guidelines and established design philosophy. Use
Related Skills
Verify Render
Download a rendered Poppify reel, run ffprobe + frame extract, surface a verdict on whether the render matches
Prove Your Tests Can Fail
Verify that tests can actually fail, catching assertions that pass no matter what the code does.
Eval Swarm
Fan every eval- evaluator agent across slices of the monorepo as a parallel swarm via the eval-swarm workflow
