| name | gen-slide |
| description | Create professional PowerPoint (.pptx) presentations from synthesized content.
Two engines: Quick mode (pptxgenjs, 10 templates) for pipeline tasks, and Pro mode
(ppt-master SVG→PPTX) for consulting-grade output with native DrawingML shapes,
20+ layout templates, 50+ chart templates, and 6700+ icons.
Colors must be hex without # prefix (pptxgenjs format requirement).
Always use this skill when the user wants a PowerPoint, deck, or .pptx file — even casual
requests like "làm bài thuyết trình", "tạo slide", "tôi cần deck để present", "xuất ra
PowerPoint", "cho tôi file pptx", "tạo bài slide chuyên nghiệp" — even without saying
"gen-slide". Use Pro mode when user says "slide chuyên nghiệp", "consulting-grade",
"slide đẹp thật sự", "dùng ppt-master", or has source docs (PDF/DOCX/URL) to convert.
|
| argument-hint | [content] [mode: quick|pro] [template: corporate-blue|...|mckinsey|google-style|...] |
| version | 2 |
| compatibility | {"requires":["Node.js >= 18","pptxgenjs (npm install -g pptxgenjs)"],"optional":["Python 3.10+ with ppt-master requirements (for Pro mode)"],"tools":["run_in_terminal"]} |
Tạo Slide — PowerPoint Output Skill
References: references/template-styles.md | references/pro-mode.md
Governance: Read and follow .github/RULE.md — it overrides all instructions below.
Quality loop (RULE-2): After generating slides, self-review + auditor gate (>80/100).
Pivot strategies: 1) different layout template, 2) different content density, 3) switch engine.
Two engines — Pro là mặc định cho mọi bài thuyết trình mới:
| Mode | Engine | Khi nào dùng |
|---|
| Pro (mặc định) | ppt-master SVG→PPTX | Mọi bài thuyết trình mới — native DrawingML, consulting-grade |
| Quick | pptxgenjs (Node.js) | Chỉ khi user yêu cầu rõ ràng "simple", "nhanh", "prototype" |
Mode Selection
User request →
├─ Pipeline call từ synthesize HOẶC bất kỳ yêu cầu tạo slide?
│ → Pro mode (MẶC ĐỊNH)
│ Đọc `references/pro-mode.md`
│
├─ User nói rõ "slide đơn giản" / "quick" / "nhanh" / "prototype"?
│ → Quick mode
│ Tiếp tục đọc bên dưới
│
└─ User nói "tạo slide đơn giản" / "5 slides nhanh" / "prototype deck"?
→ Quick mode
Pro mode: đọc references/pro-mode.md.
Quick mode: tiếp tục đọc bên dưới.
All responses to the user are in Vietnamese.
Step 0: State Read-Back (RULE-13)
Call save_state.py read-context gen-slide as FIRST action before any processing. Check relevant_artifacts[] for upstream outputs to incorporate.
Step 0.5: Artifact Evidence Injection (US-18.3.2)
After read-context, check relevant_artifacts[] for evidence to enrich the output document.
This is ADDITIVE — it does not replace compose output, only enriches it with supporting material.
EVIDENCE_INJECTION:
1. Parse `relevant_artifacts[]` from read-context
2. For each artifact with `retention: keep` AND `quality_score >= 60`:
- type `chart` or `image` → embed or reference in document where topically relevant
- type `data_table` or `excel_data` → extract key rows/figures for inline tables
- type `search_results` → use as source citations
- type `gathered_content` → extract supporting quotes or data points
3. Injection rules:
- Place evidence NEAR the section it supports (not in a separate appendix)
- Add brief caption or attribution: "(Nguồn: {artifact_summary})"
- Do NOT inject if it would disrupt document flow — quality over completeness
4. Log usage to state for auditor verification:
python3 scripts/save_state.py update --step gen-slide --status completed \
--data '{"artifacts_injected": ["path1"], "artifacts_available_but_skipped": ["path2"]}'
Quick Mode (pptxgenjs)
pptxgenjs produces good visual output with proper support for gradients, shadows, and modern
slide layouts. Colors must be hex strings without the # prefix (e.g., "1F4E79" not
"#1F4E79"). The # prefix causes silent rendering errors.
Available Templates (Quick Mode)
LIGHT:
corporate-blue: "Professional navy/blue — business reports"
corporate-red: "Bold red — executive summaries, annual reports"
academic-serif: "Scholarly serif — research, lectures, thesis"
minimal-white: "Clean whitespace — product demos, pitches"
minimal-gray: "Soft gray — internal memos, team updates"
creative-gradient: "Purple-amber — marketing, launch events"
creative-warm: "Earthy warm — non-profit, community events"
tech-modern: "Teal/blue — product launches, SaaS demos"
DARK:
dark-gradient: "Indigo/cyan — tech conferences, engineering"
dark-neon: "Neon cyan/magenta — gaming, hackathons"
STYLE_ALIASES:
corporate → corporate-blue | academic → academic-serif
minimal → minimal-white | dark-modern → dark-gradient
creative → creative-gradient
Full color/font specs: references/template-styles.md
Step 1: Pre-flight Check
- Check pptxgenjs:
node -e "require('pptxgenjs')" → if fail: "Chạy: npm install -g pptxgenjs"
- Confirm content available; if missing: redirect to gather + compose
- Determine style (user-specified, pipeline-inferred, or ask)
- Determine output path (default:
./<title>.pptx)
Template-First Protocol (US-13.4.1 / US-13.4.3)
When structured_requirements available with content_requirements:
python3 scripts/create_placeholder.py slide output/<filename>.pptx \
--slide-titles '["Title 1","Title 2","Title 3"]'
Thin Content Guard (STRICT — reject and loop back)
A presentation with only bullet-point titles and no substance looks amateurish. This is the
last line of defense against thin output. Better to loop back and enrich content than to
deliver a deck that embarrasses the user.
Automatic rejection criteria (when called from pipeline):
- < 500 words for a multi-section presentation: REJECT. Do not generate. Signal back to
synthesize: "❌ Content quá mỏng ({word_count} từ) cho presentation. Cần biên soạn lại
ở mức comprehensive." This triggers synthesize's quality loop to re-run compose.
- Sections with only 1 bullet or 1 sentence: these will produce nearly-empty slides.
If more than 40% of sections are this thin, REJECT and loop back.
- All sections are surface-level (no data, no examples, no specifics): REJECT. A
presentation without concrete content wastes the audience's time.
- No speaker notes content: For comprehensive content, compose should provide enough
material for both slides AND speaker notes. If content is only enough for slide text,
it's too thin.
When called standalone:
- Warn: "⚠️ Nội dung chỉ có ~{word_count} từ. Bài thuyết trình sẽ khá mỏng. Bạn muốn
bổ sung thêm không?"
- Proceed if user insists
Step 2: Plan Slide Structure
Before mapping content mechanically to slides, analyze the content to make intelligent
presentation decisions. A good presentation tells a story — it doesn't just dump information
onto slides. This analysis step is what separates a "slide deck" from a "presentation."
Narrative Analysis
Read the synthesized content and identify the narrative arc:
NARRATIVE_PATTERNS:
problem_solution:
structure: "Situation → Problem → Analysis → Solution → Results"
slide_design: "Open with compelling problem statement, build tension, resolve"
best_for: "Business proposals, project pitches, case studies"
timeline_journey:
structure: "Past → Present → Future (or chronological events)"
slide_design: "Visual timeline flow, milestone highlights"
best_for: "Progress reports, history overviews, roadmaps"
comparison:
structure: "Option A vs Option B (or multi-option)"
slide_design: "Side-by-side layouts, comparison tables, pros/cons"
best_for: "Decision reports, market analysis, product comparisons"
deep_dive:
structure: "Overview → Detail 1 → Detail 2 → ... → Synthesis"
slide_design: "Start broad, zoom in per topic, end with big picture"
best_for: "Research presentations, technical deep-dives, training"
data_story:
structure: "Key Insight → Supporting Data → Implications → Action"
slide_design: "Lead with the conclusion, then show evidence"
best_for: "Analytics reports, quarterly reviews, KPI dashboards"
Audience & Density Calibration
Think about who will see this presentation — it changes everything about slide density:
- Executive audience: fewer slides, more white space, 1 key takeaway per slide, big
numbers and conclusions up front, skip methodology details
- Technical audience: can handle denser slides, include methodology and data tables,
more content per slide is acceptable
- General/mixed audience: balance — use progressive disclosure (simple slide → detail
slide for those who want to dig in)
Default to the content context when the user doesn't specify audience. A quarterly business
report implies executive; a research paper implies technical; a workshop implies general.
Content-to-Slide Intelligence
For each piece of content, decide the optimal slide treatment:
| Content type | Bad (mechanical) | Good (intelligent) |
|---|
| 10 bullet points | 1 slide with 10 bullets | 2-3 slides, grouped by theme |
| Raw data table | Full table on one slide | Key insight as headline + simplified table |
| Long paragraph | Text dump slide | Extract 1 key message + supporting visual |
| Statistics/numbers | Buried in text | "Big number" highlight slide |
| Process/steps | Numbered list | Step-by-step with visual flow |
| Before/after | Two separate slides | Side-by-side comparison slide |
Engagement Techniques
Weave these into the structure naturally (not forced):
- Opening hook: start with a question, surprising stat, or bold statement — not a
table of contents (TOC slides are rarely engaging)
- Data highlights: when you have a striking number, give it its own "big number" slide
(e.g., huge "47%" centered with a one-line explanation)
- Breathing slides: between dense sections, add a section divider or visual-only slide
so the audience can reset
- Strong close: end with a clear call-to-action or key takeaway, not just "Thank you"
Report the analysis:
🎯 Phân tích nội dung:
- Narrative: {pattern} (ví dụ: problem → solution)
- Đối tượng: {audience_type}
- Mật độ: {density_recommendation}
- Kỹ thuật: {engagement_notes}
- Đề xuất: {N} slides (thay vì {M} nếu làm cơ học)
Map to Slides
Map to Slides
Map content to slide types using the narrative analysis above:
- H1 → title slide; H2 → section divider; H3 → content slide title
- Bullet lists → content bullets (max 6/slide); Tables → table slide; Images → image slide
- Key data points → highlight/callout slide; Auto-add closing slide with call-to-action
Content overflow: when a section has more than 6 bullet points, split across multiple
slides (e.g., "Key Findings (1/2)" and "Key Findings (2/2)"). Long paragraphs should be
condensed to bullets — slides are visual aids, not documents. Aim for 15-25 slides total;
more than 30 slides usually means content needs more aggressive condensation.
Present plan to user (interactive):
📊 Kế hoạch slide ({N} slides):
1. Title: "{title}"
2. Section: "{section_1}"
3. Content: "{point_1}" (bullets)
...
Bạn muốn điều chỉnh gì không?
Step 3: Prepare Data & Run Script
- Prepare content as JSON → save to
tmp/{timestamp}_slides.json
- Select template from STYLE_ALIASES or user-specified name
- Run script:
node .github/skills/gen-slide/scripts/<template>.js --input <json> --output <output.pptx>
- JSON format:
{
"title": "...",
"slides": [
{"type": "title", "title": "...", "subtitle": "..."},
{"type": "section", "title": "..."},
{"type": "content", "title": "...", "bullets": ["..."], "notes": "Talking points"},
{"type": "two-column", "title": "...", "left": ["..."], "right": ["..."]},
{"type": "table", "title": "...", "headers": ["Col1"], "rows": [["a"]]},
{"type": "image", "title": "...", "image_path": "..."},
{"type": "quote", "text": "...", "author": "..."},
{"type": "closing", "title": "Cảm ơn!", "subtitle": "Questions?"}
]
}
- Notes field: optional on any slide → written as PowerPoint speaker notes (View → Notes)
- Auto-generate notes when synthesize sets
include_notes: true (compose provides notes)
Step 4: Post-Generation Verification & Report
╔══════════════════════════════════════════════════════════════════╗
║ 🔴 DO NOT SKIP: Read the generated .pptx BEFORE reporting ║
║ A "successful" script can produce slides with no real content. ║
╚══════════════════════════════════════════════════════════════════╝
- Check exit code; on error: read traceback, fix data, retry (max 2)
- READ:
read_file the .pptx (via markitdown) or re-read the input JSON
- COUNT: slides generated, bullets per slide, total words across all slides
- CHECK DEPTH: Each content slide should have ≥ 3 bullet points with specific data/examples.
Slides with only a title and 1 generic bullet = THIN → re-generate with richer content.
- CHECK COVERAGE: Compare slide topics vs input sections — any section missing?
- If thin/missing → enrich the JSON data → re-run script (max 2 retries)
- Clean up tmp JSON file
- Report:
✅ File PowerPoint đã tạo:
📄 {output_path} | 📏 {file_size} | 🎨 {style} | 📊 {slide_count} slides
Verified: {total_words} từ across slides, {thin_count} thin slides fixed
Examples
Example 1 — Business presentation:
Input: Quarterly report content, 6 sections, with data tables
Output: corporate-blue .pptx, 20 slides (title + 6 sections + closing), speaker notes, 180 KB
Example 2 — Tech conference deck:
Input: Product launch content with code samples and screenshots
Output: dark-gradient .pptx, 15 slides, code slides with syntax, image slides, 250 KB
Example 3 — Academic lecture:
Input: Research findings, 8 sections, bullet-heavy content (needs splitting)
Output: academic-serif .pptx, 25 slides (long sections split across 2-3 slides), 150 KB
Step 5: Shared Auditor Agent Call (Post-Generation)
AUDITOR_GATE:
when: After slide generation and verification
how:
1. READ .github/agents/auditor.agent.md
2. BUILD prompt with:
user_request: original user request
output_content: slide titles + content summaries from .pptx
output_format: "slides"
required_fields: topics/sections user wanted
structured_requirements: from `python3 scripts/save_state.py check-requirements` (if available)
3. CALL runSubagent(agentName="auditor", prompt=<built_prompt>, description="Audit Slide output")
4. PARSE response:
IF VERDICT == PASS → return to orchestrator
IF VERDICT == FAIL → re-generate slides with IMPROVEMENTS guidance (max 2 retries)
budget: Counts toward max 5 auditor calls per pipeline run
skip_when: Standalone quick generation
Step 6: Artifact Registration (RULE-13)
Call save_state.py register-artifact --step gen-slide --path <file> --type draft_output --summary "<text>" for every file created in tmp/ or output/.
What This Skill Does NOT Do
- Does NOT read input files — that's gather
- Does NOT synthesize content — that's compose
- Does NOT generate Word/PDF/HTML — use respective gen-* skills
- Does NOT install dependencies — redirects to setup
- Does NOT generate chart images — that's gen-image (receives chart PNGs as input)
