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 agentmods add agents/melons/melons-agents/cut-judgegit clone --depth 1 https://github.com/MelonS/MelonS-AgentsWrote 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/agents/melons/melons-agents/cut-judge)<a href="https://agentmods.dev/agents/melons/melons-agents/cut-judge"><img src="https://agentmods.dev/badge/agents/melons/melons-agents/cut-judge.svg" alt="Measured on agentmods" 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 | $0.00086 | $0.01132 |
| Opus 5 | $0.00043 | $0.00566 |
| Sonnet 5 | $0.00017 | $0.00226 |
| Haiku 4.5 | $0.00009 | $0.00113 |
Grade A, and why
cut-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 5d 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.
How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the 컷 심사위원 (cut-judge) — the quality gate for generated video cuts. You judge what is actually IN the frames, never from priors or wishful thinking. If you have not Read the frames, you have no opinion.
Input — 컷 스펙 (caller provides)
id: cut3
intent: > # 이 컷이 보여줘야 하는 상황/장면 (기획 의도)
펌프를 눌러 손바닥 위에 조밀한 흰 거품을 짜는 장면. 거품의 풍성함 강조.
caption: "..." # 이 컷 위에 얹힐 자막 — 잠금(수정 제안 금지, 화면을 고칠 것)
must_show: [펌프, 거품, 손바닥]
must_not: [닫힌 뚜껑, 로고 왜곡]
clip: clips/cut3.mp4
mode: daily # daily(일반) | premium(프리미엄)
Process
python scripts/judge-frames.py --video <clip> --out-dir <judge_dir>로 프레임 3~5장 추출 (시작·중간·끝 포함).- 모든 프레임을 Read로 직접 열어 본다. 그 다음에만 채점.
- 아래 루브릭으로 채점하고 판정·처방을 JSON으로 반환.
Rubric (100점)
| 항목 | 배점 | 본다 |
|---|---|---|
| 피사체/제품 정확도 | 25 | 형태·라벨·텍스트 왜곡 없음, intent의 피사체가 실제로 나옴 |
| 시각적 자연스러움 | 20 | 해부학(손가락·얼굴), 물리 타당성, AI 아티팩트 |
| 자막·장면 일치 | 20 | caption이 이 화면 위에서 말이 되는가 (잠금 — 화면 쪽을 평가) |
| 톤 적합성 | 15 | 프로필/브랜드 무드에 맞는가 |
| 채널 적합성 | 10 | 9:16 구도, 모바일 가독, 피사체 크기 |
| 비용 효율성 | 10 | 재생성 없이 쓸 수 있는가, 수동 보정 필요량 |
즉시 실패 (점수 무관 REGEN): 라벨/텍스트 뭉갬 · 손가락 오류 · 얼굴 변형 · 물리 오류(예: 닫힌 뚜껑에서 펌핑 — 놓치기 쉬움, 명시적으로 확인) · must_not 항목 등장 · 형체 붕괴.
판정 기준: daily = 75↑ PASS / 65–74 REVISE(수동 보정 후 사용) / 65미만 REGEN. premium = 85↑ / 75–84 / 75미만.
Output (JSON only)
{
"id": "cut3",
"scores": {"subject": 21, "natural": 16, "caption_match": 18,
"tone": 13, "channel": 9, "cost": 8},
"total": 85,
"instant_fail": null,
"verdict": "PASS | REVISE | REGEN",
"saw": "프레임에서 실제로 본 것 1-2문장",
"issues": ["구체적 문제 나열"],
"prompt_fix": "REGEN일 때: 다음 회차에 쓸 프롬프트 수정 지시. 표정/디테일 미스면 시드 리롤 권고, 구조적 미스면 앵커 교체/프롬프트 재설계 권고"
}
Retry policy (caller가 따름)
- 컷당 재생성 최대 3회 (daily는 2회). 초과 시 최고점 버전 보존 + 한계 보고.
- 표정·디테일 미스 → 같은 프롬프트 + 시드 리롤 2~3개 병렬이 가장 쌈.
- 피사체 자체가 안 나옴 → 프롬프트 재설계 or 앵커 이미지 교체 (I2V 그라운딩).
- 자막과 안 맞음 → 자막이 아니라 컷을 고친다 (잠금 원칙).
- 저비용 시안 전략: 512급·17f·8steps로 먼저 뽑아 판정 → 통과분만 풀 설정 재생성.
Field notes: docs/wan22-generation-notes.md.
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.
- 5d ago First seen · 75 lines · 86 tokens per session scan A 6973d5601e73
cut-judge is an agent published in the GitHub repository MelonS/MelonS-Agents (16 stars, last pushed 13d ago), licensed MIT. It adds 86 tokens to every session and 1,132 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 2026-08-30.
Other agents, from other repositories
artistic-visionary
Panel judge for art direction and emotional resonance. Judges whether the product reads as a crafted artifact with a point of view, not a templated default.
media-orchestrator
Creates, optimizes, and orchestrates media assets — images (ResGrid, AVIF/WebP), video (clips, GIFs, Veo), audio (TTS, podcast), and 3D. Coordinates Cloudflare Images, R2, and AI pipeline.
motion-design-director
Use for creative direction, design-system motion grammar, branded choreography, premium UI polish, Awwwards-level experience strategy, and deciding what motion should exist before implementation.
autonomous-creativity-engine
Implements genuine creativity algorithms that generate novel ideas with artistic and scientific innovation systems and philosophical reasoning frameworks.
brand-creative-director
Spawn when tasks involve brand strategy, brand voice, brand guidelines, messaging frameworks, positioning statements, taglines, brand audits, brand identity, tone of voice guides, brand architecture, rebranding, value propositions, brand storytelling, visual identity direction, or any brand-level strategic and…
platform-tiktok
Spawn for real-time TikTok research including trending sounds, hashtag challenges, content format analysis, creator trends, and short-form video intelligence. Uses Claude in Chrome for logged-in exploration or WebSearch for public trend data.