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.
npx skills add Global-mindee/WAY --skill visual-verdictgit clone --depth 1 https://github.com/Global-mindee/WAYWrote 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.
[](https://agentmods.dev/skills/global-mindee/way/visual-verdict)<a href="https://agentmods.dev/skills/global-mindee/way/visual-verdict"><img src="https://agentmods.dev/badge/skills/global-mindee/way/visual-verdict/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/global-mindee/way/visual-verdict"><img src="https://agentmods.dev/badge/skills/global-mindee/way/visual-verdict.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00069 | $0.00950 |
| Opus 5.5 | $0.00028 | $0.00380 |
| Sonnet 5.5 | $0.00014 | $0.00190 |
| Haiku 4.5 | $0.00007 | $0.00095 |
Grade A, and why
visual-verdict 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 1mo 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.
What it actually says
visual-verdict (비주얼 판정)
생성된 UI 스크린샷을 하나 이상의 기준 이미지와 대조해, 다음 수정 반복을 구동할 수 있는 엄격한 JSON 판정을 반환한다.
언제 사용하나
- 과제에 시각 충실도 요건(레이아웃·간격·타이포그래피·컴포넌트 스타일)이 있을 때
- 생성된 스크린샷과 기준 이미지가 1장 이상 있을 때
- 수정을 계속하기 전에 결정론적 통과/실패 가이드가 필요할 때
이 스킬은 메모리 html-deliverable-render-verify(HTML 산출물 렌더링 검수 의무) 절차의 판정 표준으로 연결된다 — 정적 검사만으로 불충분한 HTML 산출물의 시각 검수 결과를 정량화한다.
입력
reference_images[]— 기준 이미지 경로 1개 이상generated_screenshot— 현재 산출 이미지- 선택:
category_hint— 의도한 UI 카테고리/스타일 (예:dashboard,report,landing)
출력 계약
JSON만 반환한다. 다음 정확한 형태를 따른다.
{
"score": 0,
"verdict": "revise",
"category_match": false,
"differences": ["..."],
"suggestions": ["..."],
"reasoning": "짧은 설명"
}
규칙:
score: 정수 0~100verdict: 짧은 상태값 (pass/revise/fail)category_match: 생성 스크린샷이 의도한 UI 카테고리/스타일에 부합하면truedifferences[]: 구체적인 시각 불일치 (레이아웃·간격·타이포그래피·색상·위계)suggestions[]: differences에 직결된 실행 가능한 다음 수정안reasoning: 1~2문장 요약
임계와 루프
- 통과 목표 임계는 90 이상.
score < 90이면 수정을 계속하고, 다음 시각 검수 패스 전에 visual-verdict를 재실행한다.- 다음 스크린샷이 임계를 넘기 전까지 시각 과제를 완료로 취급하지 않는다.
디버그 시각화
불일치 진단이 어려울 때:
- visual-verdict를 권위 있는 결정 기준으로 유지한다.
- 픽셀 단위 diff 도구(pixelmatch 오버레이 등)를 보조 디버그 수단으로만 사용해 hotspot을 국소화한다.
- 픽셀 diff hotspot을 구체적인
differences[]·suggestions[]갱신으로 변환한다.
주의: 캡처는 16,384px 텍스처 한계가 있으므로 동적 scale 후 픽셀 열람으로 측정한다(HTML 렌더링 검수 교훈). 보고서형 산출물은 등장 애니메이션을 금지한다.
예시
{
"score": 87,
"verdict": "revise",
"category_match": true,
"differences": [
"상단 내비게이션 간격이 기준보다 좁음",
"주요 버튼이 더 작은 폰트 굵기를 사용함"
],
"suggestions": [
"내비 항목 좌우 패딩을 4px 늘릴 것",
"주요 버튼 font-weight를 600으로 설정할 것"
],
"reasoning": "핵심 레이아웃은 일치하나 스타일 세부가 아직 어긋남."
}
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.
- 1mo ago First seen · 81 lines · 69 tokens per session scan A 289e3866e24e
visual-verdict is a skill published in the GitHub repository Global-mindee/WAY (11 stars, last pushed 1mo ago), licensed MIT. It adds 69 tokens to every session and 950 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-08-30.
Other skills, from other repositories
Art
Static visual content in 20+ formats (diagrams, infographics, comics, icons, wallpaper) via Nano Banana Pro by default, Nano Banana, and Flux, plus technical diagrams and a few essay illustrations that draw themselves as looping GIFs. USE WHEN art, illustration, diagram, flowchart, animated diagram, diagram GIF…
RichFormat
The rich-answer rendering mod, a Claude Code function-hooks plugin that draws assistant replies in the terminal with boxed flow diagrams, bar charts, sparklines, trees and timelines, coloured section labels, insight and caution callout bars, red and green diffs, and green and red VERIFY marks, leaving the stored text…
Excalidraw
Sketches ideas as hand-drawn Excalidraw diagrams on a live local canvas via the pinned mcp-excalidraw-server CLI, with mermaid conversion, file I/O, and share links. USE WHEN excalidraw, whiteboard, canvas, sketch a diagram, draw this on a canvas, hand-drawn diagram, draw my idea, architecture sketch, live canvas, put…
health
Audits agent config, instruction drift, hooks or MCP, and AI maintainability. Use when Claude, Codex, or Pi setup looks wrong. Not for application bugs or PR review.
seed-verify-figma-mcp-transports
A verification tool for Figma MCP tools that runs the same request through REST and WebSocket connections and compares the answers. MCP is a way for an AI agent to call external tools; REST and WebSocket are two ways to communicate with a service.
visual-acceptance
A visual final-review method for user-interface changes. It checks the rendered result with repeatable screenshots, pixel values, computed styles, and CSS layering checks across themes.