Gongnyang Prompt Kit skill
A Claude Code skill that compiles a vague request into a finished gpt-image-2 prompt and checks it with a validator script
- 329 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill runs a local Node.js script that reads the prompt files you point it at
- The Codex route uses a ChatGPT login and spends subscription quota on every image
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/gongnyang-prompt-kitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill runs a local Node.js script that reads the prompt files you point it at
- The Codex route uses a ChatGPT login and spends subscription quota on every image
This is third-party code. Review the repository files before installing.
What it does
The image-prompt skill turns a vague request such as "make a poster" into a finished prompt for the gpt-image-2 model. Claude Code first fills in the missing decisions itself: category, shot type, subject, style, composition, on-image text and aspect ratio. It asks only for the exact wording of any caption, the brand name and sensitive subjects. A routing table then picks one template, and the skill reads a single reference file from the references folder without loading the rest. The templates fall into twelve categories C1–C12 (portrait shoots, posters, cinematic key art, infographics, comics, icons, 16:9 slides and more), twelve promo templates P1–P12 and seventeen typographic poster templates TP1–TP17. Nine lighting presets L1–L9 and concept axes sit on top and push a series of variants toward different ideas. The finished prompt is written in Korean and ends with a single token such as "AR 16:9". A Node.js script, check_prompt.mjs, ships with the skill. It returns JSON with the fields ok, format, tier, errors and warnings. The script flags as errors negations outside a whitelist, retired Stable Diffusion words such as masterpiece and 8k, canvas sizes outside the allowed list and unfilled slots. Empty adjectives and missing HEX colors raise a warning. The skill does not draw images: the prompt goes to Codex, and the author suggests a separate codex-imagegen skill for batches.
Who it is for. Designers and content makers who produce posters, covers, cards and slides in gpt-image-2 and accept that the finished prompts come out in Korean.
Good fit when
- A one-line prompt gives a bland poster, and you need a template with camera, light and palette
- One idea has to become a series of variants, and the concept axes push them toward different solutions
- A caption on the image has to sit on a grid, and picking a ready typographic poster template is faster
- Before batch generation, a JSONL file of prompts goes through check_prompt.mjs with the --jsonl flag to catch errors before they burn quota
Not a fit when
- You need the image itself: the skill writes the prompt and does not start generation
- The target model is different: the size rules and the ban on negations are written for gpt-image-2
- Prompts have to be in Russian or English: the skill outputs Korean text
- You have no ChatGPT Plus or Pro for Codex, and the author builds the path to the image through it
Example request
Write a prompt for a 16:9 jazz festival poster with the title in large letters and run it through check_prompt.mjsLimitations
- The rules target gpt-image-2 and the Codex ($imagegen) route. Six canvas sizes from 1024x1024 to 2048x2048 are allowed, and the value auto is forbidden.
- The skill does not create images. Generation needs a Codex CLI login with a ChatGPT Plus or Pro subscription, and the author hands batches to the codex-imagegen skill from the codex-fleet repository.
- Prompts, templates and examples are written in Korean. The README also exists in English and Japanese.
- The validator needs Node.js and catches format violations only. It does not judge whether the result looks good.
How to disable. Delete the ~/.claude/skills/image-prompt symlink and Claude Code stops loading the skill. The repository clone can be removed separately.
FAQ
How do I install Gongnyang Prompt Kit in Claude Code?
Clone the repository with git clone and create an ln -s link from skills/image-prompt to ~/.claude/skills/. The skill starts with /image-prompt or Korean trigger phrases such as "이미지 프롬프트 써줘". After git pull, updates arrive on their own because the link points at the clone.
Does the skill generate images itself?
No, it returns a finished prompt. For one image the author suggests passing the prompt to Codex, and for a series the codex-imagegen skill from the codex-fleet repository. Codex needs a ChatGPT Plus or Pro login.
What is the validator script for?
It stops a weak prompt before generation. node skills/image-prompt/scripts/check_prompt.mjs examples/poster.txt prints JSON with ok, errors and warnings. The --jsonl flag checks a batch of prompts, and --test runs the script's own self-test.
Why do the prompts avoid phrases like "no crowd"?
According to the author, gpt-image-2 draws the word from a negation. The skill states the same thing positively: instead of "no crowd" it describes what stands in the frame. Exceptions are allowed only to protect rendered text and for portrait shoots.
Which templates does the skill choose from?
Twelve categories C1–C12, twelve promo templates P1–P12, seventeen typographic templates TP1–TP17 and nine lighting presets L1–L9. The routing table decides which one applies, and the skill reads only that file.
Discussion
No questions or reviews yet.
Related
SkillsA set of skills for Claude Code and Codex that assemble detailed image prompts from an element library and composition schemes
GPT-Image2 Style Library
gpt-image-2-style-library
A skill picks a matching template and style from a library of GPT-Image2 cases and assembles it into a finished prompt
A Claude Code skill that turns an idea into a five-stage prompt for Sora, Kling, Veo and Seedance, with templates for 21 genres
A library of 150 GPT Image prompts, each credited to its author with a link to the post, grouped into nine use cases
A skill that searches a library of 10,000+ image prompts, shows sample images and tailors the chosen prompt to your article or script
A skill and CLI for generating and editing images with GPT Image 2 and 2.5 plus a library of ready prompts
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Gongnyang Prompt Kit (this card) | no | 329 | |
| GPT-Image2-Skill | no | 6,011 | draws and edits images through Codex, while this skill only writes and checks the prompt |
| GPT-Image2 Style Library | no | 35,497 | picks a style from a case library, while this one centers on template routing and script validation |
README in short
The page opens with a side-by-side: the same line asking for a cool picture, once without the skill and once after compilation. Next come the request routing table, the key rules, a gallery of the best frames and the skill's folder tree. Links lead to a demo site, releases and the release checklist.
README badge
Original SKILL.md
---
name: image-prompt
version: "4.0.0"
description: 막연한 요청을 gpt-image-2(Codex `$imagegen`) 완성 프롬프트로 컴파일. 커버 — 카테고리 C1~C12·화보 Format B·홍보물 P1~P12·타이포 포스터 TP1~TP17·룩 L1~L9·컨셉 축 M/R/X/T·jsonl·검증기. 트리거 — "공냥 프롬프트", "이미지 프롬프트 써줘", "화보 프롬프트", "키아트", "타이포 포스터", "홍보물/판촉물", "표지/앨범커버/북커버", "활자 견본/워드마크", "포스터/카드뉴스/만화", "슬라이드/피피티 이미지", "글자 배치", "프롬프트 jsonl", "컨셉부터 잡아줘", "시안 여러 개". ※ 생성·양산은 [codex-imagegen].
---
# 🐾 공냥 프롬프트 킷 VOL.2 — gpt-image-2 프롬프트 컴파일러
모호한 요청을 `$imagegen`용 완성 한국어 프로덕션 프롬프트로 컴파일한다. 생성·양산은 [codex-imagegen](1장은 codex 직접).
## 워크플로우
1. 빠진 결정(카테고리·컷타입·피사체·스타일·구도·텍스트·AR)은 추론해 채운다. 묻는 건 한글 문구 원문·브랜드명·민감 소재뿐.
2. 라우팅 표에서 읽을 파일 결정, 포맷은 §포맷 A/B.
3. 철칙 9개를 지켜 작성, 끝에 `AR` 토큰.
4. 고가치 산출물은 응답 전 `node scripts/check_prompt.mjs` 검증(`ok:true`).
5. 생성 연계 시 컴파일된 프롬프트만 넘긴다(거친 원문 금지).
확장 예: 포스터→C3·화보→C1 B·키아트→C11·아이콘→C9("텍스트 없음")·제품→C4·만화→C10·피피티→C12(16:9).
출력 계약: 단일=본문+끝 `AR x:y`만(설명 없이) · 다중=엔트리당 `Title/Category(Cn)/Cut type/Prompt` · 생성 요청=조용히 컴파일 후 툴 호출.
## 라우팅 표 (유일 라우팅 지점)
| 요청 신호 | 카테고리/포맷 | 읽을 파일 |
|---|---|---|
| 단독 인물 화보·에디토리얼 | C1·Format B | `references/editorial-hwabo.md` (룩북·시퀀스·패션 21종 +`references/style-taxonomy.md`) |
| 타이포 포스터·글자가 곧 이미지 | TP1~TP17 | `references/typo-poster-router.md`→`references/typo-poster/TPn-*.md` 1개 |
| 활자 견본·글리프 세트·손절단 워드마크·인쇄 그라디언트 금속 | TP15~TP17 | `references/typo-poster-router.md`→ 해당 `TPn-*.md` 1개 |
| 홍보판촉물·브랜드 포스터·"디자인 잘된 포스터" | P1~P12 | `references/promo-router.md`→`references/promo/Pn-*.md` 1개. 카드뉴스 밀도 문법 금지(미감 사망) |
| 표지 판면(앨범커버·북커버·패키징 라벨)·색면 분할·회화 표지·간판체 콜라주 | P9~P12 | `references/promo-router.md`→ 해당 `Pn-*.md` 1개 |
| 포스터·키아트·인포그래픽·카드뉴스·만화·도감·아이콘·뷰티·캠페인·목업 | C2~C11 | `references/category-patterns.md` 해당 §. C6·C7=밀도 기본값·돌파 전술 §C6 |
| 프레젠테이션·슬라이드 덱 | C12 | `references/category-patterns.md` §C12 |
| 무드("있어보이게"·"럭셔리"·"영화처럼") | 룩 L1~L9 | `references/look-presets.md` 프리셋 1개 드롭인 |
| 시안 다변화·양산 컨셉·"차별화"·"컨셉부터" | M/R/X/T축 | `references/concept-axes.md` 축 1개 변주 |
| 글자 배치·폰트·그리드·밀집 텍스트 | — | `references/typography-layout.md` |
| 카메라·조명·색 어휘 | — | `references/photo-vocab.md` |
| jsonl 배치·모델 팩트·완성 예제·codex 골격·8섹션 변형 | — | `references/jsonl-and-examples.md` |
라우터(P/TP)는 패턴 1개 선택 후 해당 파일 **하나만** 로드.