AI PPT Maker Skill

> [!IMPORTANT]

> ## 🌐 Language & Communication Rule

>

> - Response language: Always match the language of the user's input. If the user writes in Chinese, respond in Chinese; if in English, respond in English.

> - Explicit override: If the user requests a specific language (e.g. "reply in English" / "请用中文回答"), use that language instead.

Conventions

• SKILL_DIR: The directory containing this SKILL.md file — no need to locate it separately.
• dist/export-pptx.mjs is a ~2 MB compiled bundle. Reading it wastes tokens and the content is not human-readable. Execute it directly with node.

Reference Documents

How It Works

This skill uses a Research → HTML → PPTX three-stage pipeline:

1. research.md as the unified content layer — Before drafting the outline, consolidate all content material into /research.md. This applies whether the source is web research results, user-provided files, or AI-organized knowledge. Downstream outline planning and HTML generation always pull from this file — never from context memory.

1. presentation.json as the single source of truth — Records the title, theme, per-slide metadata, and file mappings. All create/read/update/delete operations go through this file, ensuring cross-slide consistency and reliable file discovery during export.

1. HTML as the intermediate representation — Each slide is a standalone HTML file (1280×720 canvas) using Tailwind CSS + Lucide icons for layout. HTML is chosen over direct PPTX manipulation because LLMs generate HTML/CSS naturally, the visual expressiveness far exceeds raw PPTX XML, and slides can be previewed directly in a browser.

1. Per-slide isolated generation — Each HTML file is generated independently with no shared context between slides. Style consistency is enforced solely through the theme field. Default mode is serial (one slide at a time); parallel mode (sub-agents) is only enabled when the current platform is confirmed to support sub-agent file writes (e.g. Claude Code).

Project File Structure

<project_dir>/
├── research.md          ← Content layer — single source for all slide content
├── presentation.json    ← State file — slide registry and metadata
└── slides/
    ├── slide_001.html
    ├── slide_002.html
    └── ...

presentation.json Format

{
  "title": "Presentation Title",
  "theme": "Natural-language style description (see ppt-design.md)",
  "slides": [
    {
      "file": "slide_001.html",
      "type": "title | section | content",
      "title": "Slide Title",
      "content_spec": "What this slide covers and how it is structured (see below)",
      "local_theme": "Optional — describe special visual treatment for this slide only (e.g. cover uses a contrasting color, data slide uses dark background)"
    }
  ]
}

• The order of the slides array defines slide order. The export tool processes slides in array order.
• file is a stable identifier — never rename a file after it is created. To add or remove slides, modify only the JSON array.
• Naming rule: slide_.html (e.g. slide_001.html). Indices are monotonically increasing and are never reused after deletion.
• If content_spec or other field values contain ASCII double-quotes ("), escape them to prevent JSON parse failures.

content_spec: Slide Content Summary

content_spec describes what specific points this slide presents — list the topical directions, but leave the actual copy and data to be drawn from research.md at generation time. Core principle: say clearly "what this slide is about and how it is divided", without hard-coding copy or specifying layout.

See references/ppt-design.md for writing conventions and examples.

Workflow

> [!CAUTION]

> ## 🚨 Global Execution Discipline (MANDATORY)

>

> The following rules have the highest priority. Violating any one of them constitutes execution failure:

>

> 1. SERIAL EXECUTION — Steps MUST be executed in order. Non-BLOCKING adjacent steps may proceed continuously once prerequisites are met, without waiting for the user to say "continue".

> 2. ⛔ BLOCKING = HARD STOP — Steps marked ⛔ BLOCKING require a full stop. The AI MUST wait for an explicit user response and MUST NOT make any decisions on the user's behalf.

> 3. NO SPECULATIVE EXECUTION — Pre-generating content for a later step while still executing an earlier step is FORBIDDEN (e.g. writing slide HTML during outline planning).

> 4. GATE BEFORE ENTRY — Each step lists its prerequisites (🚧 GATE). These MUST be verified before starting that step.

> 5. DECLARE SUCCESS ONLY AFTER VERIFICATION — Do not declare a step complete until its checkpoint conditions are confirmed.

Core Workflow (Full Generation)

Step 1: Gather Requirements

🚧 GATE: User has initiated a PPT creation request.

Ask the user about: topic, purpose, target audience, desired page count, and style preference. Find out whether the user already has source material.

✅ Checkpoint — Requirements understood. Proceed to Step 2.

Step 2: Generate research.md

🚧 GATE: Step 1 complete; requirements are clear.

Read references/research-collection.md. Based on the user's situation, choose the appropriate mode and write all content material to /research.md:

✅ Checkpoint — research.md written. Proceed to Step 3.

Step 3: Plan Outline

🚧 GATE: Step 2 complete; research.md exists and contains content.

Read references/ppt-design.md. Using content from research.md, design each slide's type / title / content_spec. Write content_spec as topical direction only — no finalized copy, no layout prescriptions.

✅ Checkpoint — Outline planned. Proceed to Step 4.

Step 4: Present Plan to User

🚧 GATE: Step 3 complete; outline and theme are ready.

⛔ BLOCKING: Present the full outline (slide list + theme) to the user and wait for explicit confirmation.

> ❌ NEVER proceed to Step 5 or any subsequent step before receiving explicit user confirmation.

✅ Checkpoint — User confirmed. Proceed to Step 5.

Step 5: Initialize Project

🚧 GATE: Step 4 complete; user has confirmed the outline.

• Create /slides/ directory.
• Write /presentation.json using a file write tool. Follow the format defined in the presentation.json Format section above.

✅ Checkpoint — presentation.json written, slides/ directory created. Proceed to Step 6.

Step 6: Generate Slides (HTML)

🚧 GATE: Step 5 complete; presentation.json is valid and slides/ directory exists.

Read references/slide-generation.md and select serial or parallel mode based on the current platform. Each slide MUST be generated with simultaneous reference to both research.md and references/html-spec.md.

✅ Checkpoint — All slide HTML files generated. Proceed to Step 7.

Step 7: Validate & Fix Icons

🚧 GATE: Step 6 complete; all HTML files exist in slides/.

Run the icon validation script to automatically fix invalid Lucide icon names:

node "$SKILL_DIR/scripts/validate-icons.mjs" "<project_dir>"

The script fetches the latest Lucide icon list from CDN, scans all data-lucide references across all HTML files, and fuzzy-replaces invalid icon names with the closest valid match.

> ⚠️ If the output contains items flagged "⚠ recommend manual review" (large edit distance), inspect those icons for semantic correctness and manually replace with a more appropriate icon if needed.

✅ Checkpoint — Icon validation complete, all invalid icons fixed. Proceed to Step 8.

Step 8: Export .pptx

🚧 GATE: Step 7 complete; all HTML files are icon-validated.

See the Export to .pptx section below.

✅ Checkpoint — .pptx exported successfully. Workflow complete.

Incremental Edit Rules

When making partial changes to an existing project (modifying a slide, adding or removing pages, changing the theme, etc.), select the relevant steps from the core workflow based on user intent and follow these constraints:

Filenames are immutable: HTML filenames on disk are never renamed after creation. Indices are monotonically increasing and never reused. When adding a new slide, take the current maximum index + 1. To remove a slide, delete its entry from the slides array; the disk file may be kept or deleted.

Minimum-change principle: Choose the least-invasive operation for the type of change:

Theme change: Update the theme field in presentation.json, then rewrite all slides in "keep copy, change visuals" mode (use the same execution mode as Step 6 — serial by default).

After any modification: sync content_spec in presentation.json for affected slides, re-run icon validation, then re-export the .pptx.

Export to .pptx

Before first export, check and install fonts (one-time only — missing fonts cause layout calculation errors):

bash "$SKILL_DIR/scripts/check-fonts.sh"           # check
bash "$SKILL_DIR/scripts/check-fonts.sh" --install  # install if not found

Run export (must run outside sandbox — export depends on Playwright headless Chromium):

node "$SKILL_DIR/dist/export-pptx.mjs" <project_dir> --output <output_path>

The export tool automatically validates presentation.json. If validation fails, it outputs specific errors and fix suggestions.

If export fails due to missing Playwright, install it first:

npm init -y && npm install playwright && npx playwright install chromium

Export after every operation. Output filename format: