convez.ai
SkillProdutividadeIntermediário
SpikeMontana

Claude Decks Skill

SpikeMontana/claude-decks-skill

Gera pitch decks e tutoriais em PDF a partir de qualquer projeto Python

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill para Claude que gera automaticamente dois PDFs polidos a partir de qualquer projeto: um pitch deck B2B e um tutorial para usuários finais. Analisa o repositório, cria persona de demonstração com dados realistas, captura screenshots autenticadas via Playwright e renderiza em 4 estilos estéticos diferentes.

Instalação

$npx convez-ai SpikeMontana/claude-decks-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Criar materiais de vendas B2B automaticamente
  • Gerar documentação visual com screenshots capturados em tempo real
  • Produzir tutoriais para usuários finais sem edição manual
  • Demonstrar produtos com dados demo seguros (sem exposição de dados reais)

Veredito

Solução completa e bem pensada para automação de materiais de apresentação. Excelente para founders e product managers que precisam gerar decks profissionais rapidamente sem design manual. Requer Python e Playwright instalados localmente, mas o setup é guiado.

README

A tradução para o português deste README ainda está na fila.

claude-decks-skill

A Claude Code skill that generates two polished PDFs for any project: a B2B pitch deck and an end-user tutorial. Analyzes your repo, creates a demo persona with realistic fake data, captures authenticated screenshots via Playwright, and renders the deck in your choice of 4 aesthetic styles.

What it does

  1. Reads your project — README, package manifests, route structure — to understand the product.
  2. Asks you the aesthetic — Warm AI Concierge, Linear Frost, Dark Concierge, Editorial, or custom.
  3. Seeds a demo user with fake-but-coherent data so screenshots don't leak your real info.
  4. Captures screenshots of key screens (1440×900 @2x retina) with the demo session.
  5. Writes the markdowns for both decks with slide-type markers.
  6. Renders the PDFs via Playwright print-to-PDF using the chosen CSS recipe.

Output:

decks/
├── 01-pitch.pdf      # 15–22 slides B2B pitch
├── 02-tutorial.pdf   # 20–28 slides user manual
├── 01-pitch.md       # source — pega en Gamma para editar
├── 02-tutorial.md    # source — pega en Gamma para editar
└── screenshots/      # 6–10 PNGs from your live app

Aesthetic styles

VibeBest for
Warm AI ConciergeNotion + Things 3 — cream + coralB2B friendly, non-technical users, productivity SaaS
Linear FrostLinear + Vercel docs — frost white + electric blueDeveloper tools, dense dashboards, technical B2B
Dark ConciergeArc + Granola — dark glass + violet→cyan glowPremium AI products, futuristic, demos to tech/investors
EditorialApartamento + Mast — italic serif + generous whitespaceLifestyle, agencies, portfolios, brand storytelling
CustomWhatever you describeWhen none fit

Install

Option A — As a Claude Code skill (auto-discovered)

cd ~/.claude/skills
git clone https://github.com/SpikeMontana/claude-decks-skill.git decks

The skill becomes available in any Claude Code session. The model will detect it when relevant.

Option B — As a slash command

mkdir -p ~/.claude/commands
curl -fsSL https://raw.githubusercontent.com/SpikeMontana/claude-decks-skill/main/SKILL.md \
  -o ~/.claude/commands/decks.md

Then in any project just type /decks.

Option C — Manual

Copy SKILL.md content into your AI assistant of choice (Claude.ai, Cursor, Continue, etc.) as a prompt.

Pre-requisites

The skill is mostly self-contained, but you need Python + Playwright locally:

python3 -m venv /tmp/deck-shot
/tmp/deck-shot/bin/pip install playwright markdown
/tmp/deck-shot/bin/playwright install chromium

(The AI will set this up automatically when first run if missing.)

Usage

In a Claude Code conversation inside your project:

/decks

The AI will:

  1. Summarize what it understood about your product (5–6 bullets).
  2. Ask you to pick an aesthetic style (with recommendation based on product type).
  3. Optionally ask about light/dark default and A4 vs 16:9.
  4. Run the pipeline end-to-end (~30–90 minutes first run, 5–15 minutes subsequent regenerations).

You can also pass overrides as free text:

/decks usa paleta dark Linear-like en vez de warm
/decks la persona es un médico, no founder
/decks solo el tutorial, omite el pitch

Bundled scripts

Templates in scripts/ that the AI adapts (rather than rewriting from scratch):

ScriptRole
seed-demo.pyExample seeder for a calendar/tasks/contacts schema (SQLAlchemy). Adapt to your ORM.
shoot.pyPlaywright auth + screenshot capture. Works as-is for most web apps.
md2pdf.pyMarkdown → PDF with 8 slide types and full CSS for the 4 aesthetic styles.

Known gotchas (handled by the skill)

  • SQLite timezone loss: seeded datetimes get stripped of tz info. The skill normalizes everything to UTC explicitly before insert.
  • Naive datetimes in API responses: JS interprets them as browser-local. The skill patches the frontend to add Z suffix when missing.
  • Playwright timezone_id doesn't always override toLocaleTimeString without explicit timeZone option. The skill fixes this in the product code (real bug fix, not a screenshot workaround).
  • Single-use magic tokens: the skill re-mints if a Playwright run fails mid-way.
  • macOS AppleDouble files in tar deploys break Docker builds. The skill includes --exclude='._*' + COPYFILE_DISABLE=1.

Versioning

Aesthetic recipes and slide-type markers may evolve. Pin to a tag if you need stable output:

cd ~/.claude/skills/decks
git checkout v1.0.0

License

MIT — see LICENSE.

Origin

Extracted from internal tooling used to generate decks for Asistente Iris demos.

Ver o original no GitHub

Bom para

Relacionados