By name: gorden-super-ppt description: >- One-shot end-to-end PPT: first generate an image-format PPT (gorden-ppt-gen), then reverse it into a fully editable .pptx (gorden-img2pptx). Orchestrates image-PPT generation and image→editable-PPTX conversion A→B and delivers image-PPT + editable pptx + all intermediates. 주제만 주면 보기 좋고 편집도 되는 완성 PPT, 원샷 PPT 전체 파이프라인. Use when the user gives only a topic/content and wants a complete "good-looking AND editable" deck, or names no specific function: "PPT 만들어줘 (편집 가능하게)", "완성 PPT", "make a full editable PPT", "end-to-end deck". Do NOT use when the user explicitly wants only image slides (use gorden-ppt-gen) or only image→editable conversion (use gorden-img2pptx). Depends on both sibling skills. Ported from @Gorden Sun (github.com/GordenSun/GordenSuperPPTSkills), adapted for Claude Code.
gorden-super-ppt — End-to-end PPT Orchestration
Runs two sub-skills in sequence (no wheel reinvention — it fully executes each):
content ─▶ 【Stage 1 gorden-ppt-gen】 image-PPT (per-page .png + image .pptx)
─▶ 【Stage 2 gorden-img2pptx】 per page: background+frame+icons+text → editable .pptx
- Stage 1:
../gorden-ppt-gen/SKILL.md(function A) - Stage 2:
../gorden-img2pptx/SKILL.md(function B) - End-to-end runbook:
references/pipeline.md
Both sub-skills must be installed alongside this one (same skills dir). Image backend = Claude Code adaptation (gpt-image-2 primary, HF MCP fallback, Claude native vision) — see each sub-skill's references/runtime-notes.md.
Which to use
| User intent | Skill |
|---|---|
| "make a PPT / image-version PPT / AI slide images" | gorden-ppt-gen only |
| "convert these PPT images/screenshots to editable PPTX / extract icons / pull text" | gorden-img2pptx only |
| no named function, just a topic; or "both viewable AND editable" | this skill (A→B) |
User names a specific function → go straight to that sub-skill. No name → default to Super: generate first, then reverse each page to editable, deliver everything at once.
Orchestration (check each box)
== Stage 1: gorden-ppt-gen (run function A fully) ==
- [ ] Read ../gorden-ppt-gen/SKILL.md, execute A1–A5
- [ ] Each page actually calls the image backend to render the finished slide
- [ ] Produce: outline.json, prompts/NN-*.md, imagegen-manifest.json, slides/NN-*.png, image .pptx
- [ ] Law: every page image contains ALL real text verbatim — never placeholders/text-less templates
- [ ] Gate: no imagegen-manifest.json or any page missing generated_source → Stage 1 fails, do not enter Stage 2
== Stage 2: gorden-img2pptx (run function B fully on each Stage-1 image) ==
- [ ] Read ../gorden-img2pptx/SKILL.md, execute B0–B9 per page
- [ ] Forced 4 layers: background → frame(chroma-key) → icons(chroma-key) → text(vision) — none skipped
- [ ] background/frame/icon image layers must come from the image backend (extraction), not PPT shapes/code/cropping
- [ ] Produce per page: imagegen-assets-manifest.json + background/frame/icons/layout.json + composed editable .pptx
== Delivery ==
- [ ] Deliver at once: ① all slide images ② per-page background ③ frame ④ icons/decor ⑤ text (layout.json) ⑥ image .pptx ⑦ editable .pptx
Key constraints (both stages)
- 🔴 Stage 1 MUST call the image backend (never code-drawn, never placeholder); Stage 2 background/frame/icon layers MUST also be backend-generated (never skip frame + icons). These two are the root cause of past regressions — hold them.
- Unified aspect ratio A and B end-to-end (default 16:9; user wants 3:2 → all 3:2).
- Unified palette, no repeated framework (Stage 1); color/line-preserving chroma-key, icon sizes matched to source (Stage 2).
- Zero fabricated data; language follows user; clean footer; avoid large pure-green areas (chroma-key conflict).
Output layout (suggested)
<topic-slug>/
├── outline.json / imagegen-manifest.json / prompts/ / slides/ # Stage 1
├── out/<topic>-image-deck.pptx # Stage 1 image deck
└── editable/ # Stage 2
├── NN/{background.png, frame.png, icons/*.png, layout.json}
├── deck.json / preview/slide_*.png
└── <topic>-editable.pptx # Stage 2 editable deck
Sub-skills carry their own scripts & references; this skill only orchestrates and delivers. If the sub-skills aren't co-located, install them first or call them separately.