prompt

/prompt — AI prompt generator (Codex port wrapper)

Codex image purpose pre-gate

If the user's request is one of image·picture·photo·poster·card news·infographic·thumbnail·image editing, perform the following before the general procedure.

1. In-plugin ../image-prompt-guide/references/full.md를 read from beginning to end. Do not substitute the mere fact that the file exists or a summary of this document.
2. On this path, do not call the image-prompt·gongnyang-photo skill outside the plugin. Use only the internal guide as the reference file. (The original Gongnyang Kit text is enclosed inside ../image-prompt-kit/the plugin — reading those files indicated by the guide §K5 routing table is not a bypass.)
3. The final output follows the guide §0 contract — gpt-image-2/Codex $imagegen target = Gongnyang Kit text prompt(Format A label 6-section or Format B pictorial comma-style, ending with only the AR x:y token · batch = jsonl record). ChatGPT/Gemini web UI simple path only uses the legacy JSON skeleton (purpose·hero·context·evidence·visual_re_description·constraints 6 keys).
4. Self-check immediately before output — Kit path: ①Format A/B structure ②zero iron-rule violations (fully affirmative phrasing for negatives·zero SD vocabulary·ending AR) ③rendered text uses the Tier-1 combination formula once ④**타겟: <모델> Is 1 line above the prompt** (🏛️ Great Principles §2 — specify the model auto-detection result) ⑤is there a real-world expert visual language anchor(🏛️ Great Principles §1 — for pictorial Format B, the Director signature:·Lens character: slot; for Format A, put real photographer/director styles as result descriptions in Camera·Lighting·Color grading) / web UI path: all 6 keys. If unmet, do not output; reapply the internal guide and regenerate once.

🚨 Post-output workflow — mandatory for all purposes including the image path (skip = contract violation)

After outputting the prompt code block, you must present the 5-option menu below exactly as is and wait for the user's selection. Even if the image pre-gate was taken, this step is not exempt (evidence that through v3.2.1 this sentence was weak and the menu was omitted on the image path — 2026-07-29 ChatGPT Works reproduction). The only exemption = --batch argument or AUTO mode (bot autonomous flow).

## 어떻게 하시겠습니까?

1️⃣ Execute now - Execute the task right away with that prompt 2️⃣ Auto improve - AI automatically strengthens the prompt → output the modified prompt (execute ❌) 3️⃣ Manual improve - Select among the presented options to modify → output the modified prompt (execute ❌) 4️⃣ Other - Another request or question 5️⃣ Agent mode - Complete prompts step by step while conversing with AI (deriving optimal results)

💬 Please select (예: "1", "2", "3", "4", "5")

산출 검증기(리포 `scripts/prompt-output-gate.mjs` — `--mode interactive --purpose image`)가 이 계약(A1 5옵션·A2 전문가 앵커·A3 타겟 표기·A4 끝 AR)을 기계 판정한다.

그 밖의 요청은 이 스킬 디렉터리의 [references/prompt-command.md](references/prompt-command.md)를 읽고, 그 문서 전체를 프롬프트 생성 절차로 삼아 사용자의 요청($ARGUMENTS)을 처리한다. 이미지 요청도 위 선행 게이트를 통과한 뒤 필요한 공통 절차를 이 문서에서 보충한다.

- 모델별 전략 문서는 플러그인 루트 `skills/` 아래에 있다 — flat: `claude-fable-5-prompt-strategies.md` / 대형 가이드(P5 분할): `skills/<이름>/references/full.md` (`gpt-5.6-prompt-enhancement`·`gemini-3.1-prompt-strategies`·`image-prompt-guide` 등 — 각 디렉터리의 SKILL.md 가 섹션 맵 인덱스). `references/prompt-command.md` 가 지시하는 상대경로 `skills/...` 는 이 플러그인 루트 기준으로 해석한다.
- `--batch` 인자가 있으면 질문 없이 바로 결과를 출력한다.