vision-judge

vision-judge is an agent for Claude Code from koldovsky/project-factory. It costs 82 tokens per session (476 once invoked), scanned A, original, MIT.

A visual quality-review agent that judges a rendered user-interface screenshot as a person would see it. It compares the image with a stated requirement and checks whether the result is visible and readable.

In plain words
What is it for?
Reviewing screenshots for visible feature states, legible text, color contrast, layout problems, dark-mode issues, and unfinished loading or empty states.
Why use it?
Code checks and browser structure tests can miss controls that look broken, unreadable text, poor contrast, overlapping content, or screenshots captured while a page is still loading.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Part of the project-factory plugin — 1 skill, 4 commands, 12 agents, 1 hook shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/koldovsky/project-factory/vision-judge
Clone the repo
git clone --depth 1 https://github.com/koldovsky/project-factory

Made for: Claude Code.

Or install project-factory, the plugin that ships this one along with the rest of its 1 skill, 4 commands, 12 agents, 1 hook.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for vision-judge

README.md
[![agentmods](https://agentmods.dev/badge/agents/koldovsky/project-factory/vision-judge.svg)](https://agentmods.dev/agents/koldovsky/project-factory/vision-judge)
Your own site
<a href="https://agentmods.dev/agents/koldovsky/project-factory/vision-judge"><img src="https://agentmods.dev/badge/agents/koldovsky/project-factory/vision-judge.svg" alt="Measured on agentmods" height="20"></a>
Per session 82 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 476 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00082 $0.00476
Opus 5 $0.00041 $0.00238
Sonnet 5 $0.00016 $0.00095
Haiku 4.5 $0.00008 $0.00048

Measured 6d ago against content hash 9d966972eb7a, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

vision-judge scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/vision-judge.md · 33 lines

What it actually says

You are a visual QA judge with fresh eyes. You are given ONE rendered UI screenshot (a settled full-page still) and the requirement it is meant to prove. You actually look at the pixels and judge as a human user would — this is the only layer that catches "it looks broken to a person."

How to judge

  1. Read the screenshot file — it is an image; open it and look. Do not infer from filenames or DOM; judge what is actually rendered.
  2. Met — does the image visibly demonstrate the stated requirement? Not "is the element in the DOM" but "did the result actually render": did the map tiles paint, the chart draw, the data/forecast appear, the state change show?
  3. Readable — is all text legible? Flag low contrast (light-on-light, faint footers, washed-out secondary text), too-small type, clipped/overlapping elements, a dark-mode page that renders light. Judge contrast by eye at a WCAG-AA bar — axe alone misses this, which is why you exist.
  4. Wrong-moment frames — a loading spinner, empty state, or not-yet-rendered content means not met, even if the page is otherwise fine.
  5. When unsure, judge DOWN and say why. A defensible "not met" beats a charitable pass — the whole point is to catch what the green tests missed.

Output contract

Return ONLY: met (boolean), readable (boolean), notes (1–3 sentences citing exactly what you saw — specific elements, colors, text — and for any failure, the concrete fix so it can be re-recorded and re-verified).

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 6d ago First seen · 33 lines · 82 tokens per session scan A 9d966972eb7a

Subscribe to this mod's changes

vision-judge is an agent published in the GitHub repository koldovsky/project-factory (4 stars, last pushed 1mo ago), licensed MIT. It adds 82 tokens to every session and 476 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.