Competition PPT Template-First Skill

Start by deciding what the deck must communicate and what must visibly appear on every page, then make the important parts editable.

Chinese documentation: README.zh-CN.md

competition-ppt-template-first is a reusable Agent Skill for competition, product, and technical PPTs. It replaces the usual "dark background + generic components + tiny text" workflow with a content-to-template sequence: analyze the material, choose the story, define each page's layout and visual ingredients, generate a bespoke whole-slide background template for that exact page, then build the editable deck.

flowchart LR
    A[Report, brief,<br/>screenshots and assets] --> B[Content analysis<br/>and visual inventory]
    B --> C[Story route, asset roles,<br/>and deck style tokens]
    C --> D[Speaker map, evidence plan,<br/>and page rhythm]
    D --> E[Layout registry and per-page<br/>template contracts]
    E --> F[Template images<br/>and composition review]
    F --> G[Editable PPT<br/>with real evidence]
    G --> H[Rendered visual critique<br/>and fix verification]
    H --> I[Approved files]

The template image carries composition, atmosphere, material, light, and visual detail. The editable PPT layer carries the factual claim, real screenshots, charts, certificates, and the text the user may need to update. Before page production, the Skill locks a deck token system, assigns every image a truth-aware role, plans a speaker map, records data evidence for quantitative claims, and registers anchor / dense / breathing page rhythm across the deck. Every page is planned first: content grouping, background scene or quiet surface, primary visual, supporting visual ingredients, real-image slots, and text zones. Then every content page receives a bespoke whole-slide template image generated from that plan. Dense technical pages still receive a quieter generated template, not a flat or card-only exception. Cover and closing pages use separate hero/conclusion template rules.

Two Practical Levels

Start with the level that matches the material on hand:

Both levels follow the same visual method. The report-grounded level adds traceability where numbers, model comparisons, awards, and technical claims need to withstand questions. Neither level defaults to a dark-blue cyber style: palette, background scene, product/material cues, and light/dark distribution are selected from the subject.

Report-Grounded Deck Mode

Feed the project package, not just a topic. The skill converts source documents into a traceable defense route before it designs a page:

report / proposal / requirements / data / screenshots
  -> source manifest
  -> content analysis: deck type, audience, story, visual ingredients, gaps
  -> usable facts and source links for key claims
  -> judge-facing narrative
  -> speaker map: what the judge sees versus what the presenter explains
  -> data-evidence plan for comparisons, metrics, and test results
  -> page rhythm and page/template contracts
  -> representative template
  -> editable defense PPT

Use it when the deck must stay faithful to a .docx, PDF, technical report, research paper, data workbook, or competition brief. Keep a precise source locator for metrics, comparisons, awards, and other claims likely to be challenged. Read content-to-template-analysis.md and source-ingestion.md for the complete option.

Why Template-First

Install

Use any Skill-compatible agent installer, or copy skills/competition-ppt-template-first/ into the agent's skills directory. For installers based on the community skills CLI:

npx -y skills@latest add che626/competition-ppt-template-first-skill \
  --skill competition-ppt-template-first \
  --agent codex \
  --global

The skill entry point is skills/competition-ppt-template-first/SKILL.md. It is designed for Codex, Claude Code, Cursor, and other agents that recognize Agent Skills.

Maintainers can validate installer discovery before a release with:

.\skills\competition-ppt-template-first\scripts\validate-distribution.ps1

Use -Remote after publishing to test the GitHub clone path as well.

Use It

Attach the report, evidence images, and any reference deck, then use a prompt such as:

$competition-ppt-template-first
Read the project report and create an 11-page AI-vision competition defense deck.
Use the supplied screenshots as real evidence. First produce a content-to-deck analysis,
the story route, a speaker map, and a template contract for every page: background scene,
primary visual, real-image slots, text zones, supporting visual ingredients, page rhythm,
and avoid list. Make a data-evidence plan for model comparisons or metrics. Then make one
representative template-first page for approval.

More copy-ready prompts are in prompt-recipes.md.

To initialize the standard report-grounded workspace:

python skills/competition-ppt-template-first/scripts/init-report-grounded-deck.py ./my-competition-deck --source-folder ./project-materials

What the Skill Produces

competition-ppt/
  00_intake/     source manifest, extraction notes, image asset roles, original-asset references
  00_plan/       content analysis, fact registry, deck brief, style tokens, speaker map, data-evidence plan, layout registry, slide-source map, page blueprints
  01_templates/  approved 16:9 template images and prompt records
  02_build/      editable PPTX work files
  03_renders/    exported previews, rendered visual critique, and QA notes
  04_approved/   explicitly confirmed pages
  99_retired/    rejected variants kept for traceability

The full folder convention is documented in project-conventions.md.

Included Playbook

Design Position

Clear before ornate. Use scenes, materials, real visual evidence, product context, light, and typographic scale when they help the page. Keep dense pages calm and readable; keep low-text pages more visual. Avoid generic cyberpunk frames, repeated rounded-card walls, fake UI screenshots, darkened stock photos behind cards, illegible generated text, and body copy reduced to microscopic size.

Scope and Limits

• Best for innovation, AI, robotics, engineering, research, and product-defense decks.
• It is not a generic slide-template library or an automatic corporate-branding system.
• It does not invent metrics, achievements, or product capabilities.
• It intentionally keeps complex visual underlays as images when that is the best way to preserve quality; this is not the same as dropping a finished slide screenshot into PowerPoint.
• It is a working method, not a promise that every project needs cinematic visuals, source tables, or a complicated production pipeline.
• It does not treat the word "AI" as a reason to make every page black, navy, cyan, or HUD-like. The visual system must be justified by the project subject and the page role.

Contributing

Read CONTRIBUTING.md before opening an issue or proposing an archetype. The repository accepts generalized, rights-cleared workflows and examples only; do not submit private project evidence.

License

MIT. This repository contains no project screenshots, certificates, or proprietary competition material.