Walkthrough Plan
GKjohns/walkthrough-plan
Converta um vídeo de walkthrough narrado em plano de reforma executável
Ver no GitHubSkill que transforma um vídeo narrado de um ambiente em um plano de limpeza e redecioração organizado por cômodos. Analisa áudio transcrito e frames do vídeo para gerar uma checklist markdown e uma página HTML interativa com lista de compras ajustada ao orçamento.
Instalação
npx convez-ai GKjohns/walkthrough-plan1 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
Casos de uso
- Planejar renovação de ambiente (sala, quarto, apartamento)
- Organizar listas de compras com restrição de orçamento
- Gerar checklists interativas para compartilhar com familiares
Veredito

Skill bem executada que prova real compreensão de vídeo: transcricão + análise visual de frames + síntese estruturada em plano acionável. Excelente para quem precisa executar reforma ou limpeza grande e quer organizar trabalho por prioridade e orçamento.
README
A tradução para o português deste README ainda está na fila.
walkthrough-plan
A Claude skill that turns a narrated video walkthrough of a space into a room-by-room cleanup and redecoration plan — a markdown checklist plus a calm, shareable HTML one-pager with a budget-tuned shopping list.
You record yourself walking through your place saying "throw this out," "move this here," "I want art on this wall." The skill transcribes the audio, visually analyzes the frames to figure out what you were pointing at, and produces a plan you can actually execute — phone in hand, ticking tasks off as you go.
What it produces
PLAN.md— the canonical plan: a summary line, the decisions to make first, an effort-sorted checklist (Tier 1 = zero purchases this weekend, Tier 2 = after one shopping run), a "leave it alone" list, and a budget-tuned shopping list.index.html— a shareable one-pager (AirDrop it to your phone). Soft, calm design; checkboxes persist on the device as you tick them off.
It's built to prove it understood the video — the "leave it alone" list, the inline ⚠️ warnings ("measure the overhang first," "not a crate"), and the decisions pulled to the top are the signal that it listened, not just transcribed.
Inputs
- A video — point the skill at a
.mov/.mp4walkthrough. - A budget — exact (
$300) or general ("on a budget","whatever it takes"). The skill asks if you don't say.
Requirements
ffmpeg+ffprobe,curl,python3OPENAI_API_KEYin the environment (used for Whisper transcription)
Install
Clone into your skills directory (or wherever your agent loads skills from):
git clone https://github.com/<you>/walkthrough-plan.git ~/.claude/skills/walkthrough-plan
Then just ask: "turn this walkthrough video into a cleanup plan — budget's $300."
Example
Point it at a 7-minute phone video of someone walking through their apartment. It comes back with:
- a summary line —
25 things to do · ~$300 · 1 weekend - the decisions to make first (where does the heirloom chair go? approve the art style?)
- a Tier 1 list of everything doable this weekend with bare hands, then Tier 2 after one shopping run
- a "leave it alone" list of everything you said to keep — so nothing gets touched that shouldn't
- inline ⚠️ constraints it caught from the narration ("measure the overhang first," "not a crate")
- a shopping list tuned to the budget, one pick each, stable links with search-term fallbacks
The screenshot above is the actual rendered one-pager. Checkboxes persist on the device, so you can AirDrop it to your phone and tick things off as you walk the space.
How it works
scripts/prep.shextracts compressed audio, samples a frame every 5s, transcribes with Whisper, and writes a timestamped transcript (frames align to transcript timestamps).- A subagent views the frames against the transcript and returns text-only notes on each room's state and exactly what was being pointed at.
- The skill synthesizes
PLAN.md, then fills theDATAobject inassets/onepager_template.htmlto renderindex.html.
Repo layout
walkthrough-plan/
├── SKILL.md # the skill: instructions for the agent
├── scripts/
│ └── prep.sh # audio + frames + Whisper transcription
├── assets/
│ └── onepager_template.html # the shareable one-pager (DATA-driven)
└── README.md
License
MIT