---

name: prompt-gallery description: "Choose and adapt GPT Image 2 prompt patterns from the curated gallery and prompt-craft guidance." disable-model-invocation: true metadata: disco-role: operating license: MIT

Prompt Gallery

Use this sub-skill when the user needs help choosing, adapting, or composing GPT Image 2 prompts from the gallery patterns: research figures, diagrams, UI mockups, posters, typography, photography, game/anime styles, product imagery, reference-image edits, or hybrid visual directions.

Do not run the image API, inspect credentials, or build final shell commands here. For actual gpt-image CLI/API execution and preflight, route to ../cli-and-api/SKILL.md. Do not add, remove, or update gallery/repository files here; route repository maintenance to ../repo-maintenance/SKILL.md.

Gallery-first operating loop

1. Classify the user request. Identify the deliverable type, output surface, required text, subject/domain, style family, and whether it is a generation or edit.
2. Select categories before writing. Use references/gallery-catalog.md or scripts/select_gallery_categories.py to choose one primary category. Load guidance from 2-3 categories only when the user explicitly asks for a hybrid result.
3. Extract a pattern, not a prompt dump. Reuse the category's structure: canvas, layout grammar, visible entities, exact labels, style bounds, material/light/palette controls, and avoid-lines.
4. Apply craft checklists. Use references/prompt-craft.md for exact text, diagrams, UI specs, multi-panel boards, research figures, edits, and size/quality choices.
5. Return a generation-ready prompt plus routing note. Provide the final prompt and any parameter suggestions. If the user wants execution, hand off to ../cli-and-api/SKILL.md.
6. Preserve provenance discipline. If adapting an outside-source or community pattern into a public gallery entry, preserve Author + Source. Use Curated only for repo-created, repo-curated, or substantially reworked material.

Required user-facing output

For normal prompt-help requests, return:

• selected primary category and optional hybrid categories;
• short rationale for the category choice;
• a polished prompt or structured prompt skeleton;
• exact text block list when typography/labels matter;
• suggested canvas/size family and quality intent;
• API/CLI handoff note without executing anything.

Quick routing

• Research paper figure, method diagram, data chart, scientific/technical diagram -> Research Paper Figures, Data Visualization, Scientific & Educational, or Technical Illustration.
• UI/app/dashboard/product mockup -> UI/UX Mockups; combine with Brand Systems & Identity if the design system itself is part of the deliverable.
• Posters, covers, dense typography, Chinese copy, event flyers -> Typography & Posters, Infographics & Field Guides, or Events & Experience.
• Product hero image, packaging, beverage/food campaign -> Product & Food; combine with Photography or Brand Systems & Identity when needed.
• Anime/manga/game stills/HUD/worldbuilding -> Anime & Manga, Gaming, Character Design, Pixel Art, Isometric, Retro & Cyberpunk, or Cinematic & Animation.
• Reference-image transformations, inpainting, or multi-reference composition -> Edit Endpoint Showcase for prompt shape, then route execution to ../cli-and-api/SKILL.md.

References

• references/gallery-catalog.md — distilled category map, loading policy, and hybrid-selection signals.
• references/prompt-craft.md — compact prompt recipes and checklists.
• references/troubleshooting.md — failure diagnosis and prompt repair patterns.
• scripts/select_gallery_categories.py — safe no-network category selector for task phrases.