Powerpoint Business

PowerPoint Business — SVG-Native Slide Generation

Business/humanities presentation skill. Full lifecycle: analyze → interview → research → outline → plan → draft SVG → design SVG → export.

Execution model: LLM generates SVG (1280x720) per slide → SVG embedded in OOXML PPTX (editable in Office 365+, PNG fallback for older versions). Editing: natural language → semantic manifest element lookup → targeted SVG regeneration.

0. Quick Reference

Parse $ARGUMENTS to determine action. If none specified, ask.

1. Hard Rules

1. Semantic IDs on every SVG element — s{nn}-{role}-{index} format. No anonymous elements.
2. Safe area — 60px padding on all sides. Canvas 1280x720, usable area 1160x600.
3. Card gap >= 20px — never closer.
4. No external image references — all inline. User images: base64 embedded.
5. Draft ≠ Design — Design phase must not alter text content or card count. Minor size adjustments for theme fit OK.
6. Layout diversity — never use same layout for 3+ consecutive slides.
7. Information density — max 7 bullets, max 5 cards, max 80 CJK chars per card body.
8. Font size floor — minimum 12px.
9. Color contrast — WCAG AA >= 4.5:1. Verified by check_svg.py.
10. Manifest sync — after any SVG change, run sync_manifest.py. Script is source of truth.
11. Two HARD STOPs — user confirms requirements (Phase 1) and outline (Phase 3).
12. Research tags — web-sourced info tagged: confirmed / probable / unverified.
13. <foreignObject> for multi-line text — card bodies use <foreignObject> + HTML div. Single-line (titles, stats) use <text>.
14. Version before edit — copy current SVG to slides/history/ before any edit.

2. Themes & Layout

14 themes (10 business + 4 academic/humanities):

• Business: Corporate Blue (default), Warm Earth, Minimal White, Tech Dark, Emerald Gold, Royal Purple, Sunset Coral, Nordic Sage, Ocean Teal, Rose Dusk.
• Academic/Humanities: Zen Ink (禅意水墨), Academic Classic (学术经典), Parchment (自然书卷), Pure Academic (极简学术).

14 Bento Grid layouts + 3 semantic visualization components: content-driven, not template-based. 206 Lucide SVG icons across 9 categories: general, communication, data, business, education, humanities, nature, status, arrows. 28 chart templates indexed by analysis goal in references/charts-index.json.

Read references/themes.yaml for full theme definitions (color scheme, typography, card style, elevation, decoration). Read references/bento-grid.md for layout strategies, density targets, card constraints, and structural components. Read references/svg-conventions.md for SVG generation rules (IDs, fonts, multi-layer backgrounds, dynamic sizing, data viz). Read references/cognitive-design.md for evidence-based design principles (Miller's Law, Mayer, Gestalt, cognitive load, narrative arc).

3. Agent Roles

5 specialized roles, orchestrated by Lead Agent (this skill). Each role has a dedicated system prompt in prompts/. Dispatch as Agent subagents with the corresponding prompt.

Role boundaries: each role reads only upstream outputs, never modifies other roles' artifacts. Reviewer feedback routes to the correct role (content issues → Planner, visual → Designer).

4. Actions

4.1 create [topic/file] — Full Pipeline (Phase 0-7)

Phase 0: Material Analysis (Consultant)

Read user-provided materials (PDF, Word, text, URL). Extract key information, data points, logical structure, target audience cues. If user provides only a topic string with no materials, note that Phase 2 research will be essential.

Proceed directly to Phase 1 — do not present analysis yet.

Phase 1: Requirements Interview (Consultant) — HARD STOP

Dispatch Consultant agent with prompts/consultant.md. The Consultant handles:

• Interactive requirements interview via AskUserQuestion tool (questions, options, recommendations)
• Additional topic-specific questions as needed beyond baseline 5
• Produces requirements.md (audience, purpose, scope, style, content strategy, rough chapter outline)

User must confirm requirements before proceeding.

Phase 2: Web Research (Researcher) — Optional

Only if user requested. Dispatch Researcher agent. Read prompts/researcher.md.

Process: Planner produces rough outline first → Researcher supplements per chapter using grok-search → Planner refines. Max 1 iteration cycle.

Output: research/chapter-{nn}.md files, confidence-tagged.

Phase 3: Pyramid Outline (Planner) — HARD STOP

Dispatch Planner agent. Read prompts/planner.md.

Apply Pyramid Principle + select narrative framework (Pyramid/SCQA/PAS/Hero's Journey). Output structured outline.json.

Present outline to user. User must approve before proceeding.

Phase 4: Planning Draft (Planner)

Same Planner agent continues. For each page, produce 8-field planning card:

Output: planning-draft.md

Phase 5: Draft SVG (Designer)

Dispatch Designer agent. Read prompts/designer.md.

Generate one SVG per slide (1280x720). Focus on content + layout, neutral gray palette. Every element gets semantic ID. Batch: 3-5 slides per subagent.

Output: drafts/slide-{nn}-draft.svg

Phase 6: Design SVG + QA (Designer + Reviewer)

Designer injects theme style onto Draft SVGs. Does NOT change content/layout, only visual polish.

Then Reviewer runs two-layer QA:

1. scripts/check_svg.py — automated (well-formedness, safe area, contrast, gaps)
2. LLM review — layout balance, density, typography, aesthetics

Quality gate: >= 7.0 pass, 5.0-6.9 fix loop (max 2 rounds), < 5.0 redesign.

Output: slides/slide-{nn}.svg + slide-manifest.json + reviews/

Phase 7: Export

# Generate manifest from final SVGs
python ~/.claude/skills/powerpoint-business/scripts/sync_manifest.py slides/ slide-manifest.json

# Export PPTX (SVG-embed, editable in Office 365+)
python ~/.claude/skills/powerpoint-business/scripts/export_pptx.py slides/ output/presentation.pptx

# Generate HTML preview
python ~/.claude/skills/powerpoint-business/scripts/preview.py slides/ output/index.html

# Open preview
open output/index.html

Post-Creation Checklist

[ ] All SVGs well-formed (check_svg.py passes)
[ ] Manifest synced (sync_manifest.py run)
[ ] QA score >= 7.0
[ ] Layout diversity — no 3+ consecutive same layout
[ ] PPTX generated and non-empty
[ ] HTML preview opens correctly
[ ] Speaker notes in output/speaker-notes.md

4.2 edit [project]

Natural language editing. User describes change in plain language.

1. Read slide-manifest.json to locate target element(s)
2. Backup current SVG to slides/history/slide-{nn}.v{ver}.svg
3. Dispatch Designer with current SVG + targeted instruction
4. Run sync_manifest.py to update manifest
5. Re-export PPTX + HTML

Edit granularity:

• Element edit: single text/color/size change
• Slide redesign: regenerate entire slide
• Theme switch: → use retheme action instead
• Batch replace: find-and-replace across all slides

4.3 undo [project] [slide]

Restore previous version from slides/history/. Update manifest. Re-export.

4.4 retheme [project] [theme]

Regenerate all Design SVGs from Draft SVGs using new theme. Drafts in drafts/ are preserved. Re-run full QA cycle. Re-export.

4.5 export [project]

Re-export from current SVGs:

python ~/.claude/skills/powerpoint-business/scripts/export_pptx.py slides/ output/presentation.pptx
python ~/.claude/skills/powerpoint-business/scripts/preview.py slides/ output/index.html

4.6 preview [project]

Generate HTML preview and open in browser:

python ~/.claude/skills/powerpoint-business/scripts/preview.py slides/ output/index.html
open output/index.html

4.7 review [project]

Run Reviewer on existing slides. Dispatch Reviewer agent with prompts/reviewer.md. Produces reviews/ directory with per-slide reports and scores.

5. Dependencies

Required:

pip install python-pptx lxml cairosvg

Optional:

• matplotlib — complex statistical charts (simple charts are inline SVG)
• soffice (LibreOffice) — PPTX → PDF QA verification

Check before first run:

python -c "import cairosvg; print('cairosvg OK')"
python -c "from pptx import Presentation; print('python-pptx OK')"

6. Troubleshooting