Design-Log
LiozShor/design-log
Stop & Think planejamento research-driven para Claude Code
Ver no GitHubSkill de planejamento para Claude Code que força pesquisa e design antes da codificação, usando um protocolo de 5 fases (Discovery → Research → Explore & Document → Implement → Test Handoff). Registra decisões em artefatos persistentes e exige aprovação antes da implementação.
Instalação
npx convez-ai LiozShor/design-log1 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 novas features com documentação de design antes de codificar
- Garantir que pesquisa técnica e análise de padrões aconteçam antes da implementação
- Manter histórico persistente de decisões arquiteturais por feature
- Evitar retrabalho através de exploração e aprovação gated
Veredito

Excelente para times que precisam de rigor arquitetural e rastreabilidade de decisões. Força disciplina de design antes da implementação, reduz débito técnico e evita retrabalho — particularmente valioso em projetos complexos ou colaborativos.
README
A tradução para o português deste README ainda está na fila.
design-log — a "Stop & Think" planning skill for Claude Code
A reusable Claude Code skill that forces research-driven design before coding. Instead of jumping straight to implementation, it runs a 5-phase protocol — Discovery → Research → Explore & Document → Implement → Test Handoff — gated by a single approval and recorded in a persistent design-log file.
Built on top of Claude Code's plan mode. One approval gate, one documented artifact per feature.
Install
Drop the design-log/ folder into your skills directory:
- Per-project:
<your-repo>/.claude/skills/design-log/ - Global (all projects):
~/.claude/skills/design-log/
That's it. No build step, no config. Then in Claude Code run:
/design-log add a rate limiter to the upload endpoint
What it does
- Discovery — checks existing design logs, scans the codebase for reuse, asks you 5+ clarifying questions before touching anything.
- Research — reads 3+ sources on the relevant domain (resilience, search UX, concurrency, etc.) and extracts principles/patterns/anti-patterns.
- Explore & Document — enters plan mode, explores your code, writes a
[DRAFT]design log, then asks for approval via plan mode's exit gate. - Implement — only after approval. Tracks status, records deviations.
- Test Handoff — persists outstanding tests so nothing is lost between sessions.
The artifact lives at .agent/design-logs/NNN-description.md (created automatically).
Requirements
-
Claude Code with plan mode (
EnterPlanMode/ExitPlanMode). -
A git repo (the skill checks branch state and won't implement on
main). -
A web research tool — works with any of:
- Built-in
WebSearch+WebFetch— always available, zero setup. ✅ default - context7 MCP — library/framework/SDK/API docs, if installed.
- Exa MCP — semantic/neural source discovery + deep research, if installed.
- Tavily MCP — general multi-source web search, if installed.
- Firecrawl MCP — full-page extraction / docs-site crawling, if installed.
The skill prefers whichever MCP fits the job when present and falls back to the built-ins otherwise — so it runs fine with nothing extra installed.
- Built-in
Customizing
It works with zero config. To adapt log location, research tool, git workflow, or add log-number reservation for parallel sessions, see references/customization.md.
Files
design-log/
├── SKILL.md # the protocol router + hard gates
├── README.md # this file
├── assets/
│ └── design-log-template.md # the Section 1–8 design-log template
└── references/
├── protocol-detail.md # full step-by-step for each phase
├── research-sources.md # book/article/case-study tiers by domain
└── customization.md # optional per-project adaptation
Credits
The Design-Log methodology this skill implements originates with Yoav Abrahami (Wix Engineering): "Why I Stop Prompting and Start Logging: The Design-Log Methodology".
This repo is an independent Claude Code skill packaging of that idea — research-driven, version-controlled design documentation that the AI reads before it writes code. Credit for the core concept goes to Yoav; any bugs in this implementation are mine.
License
MIT — see LICENSE. Do whatever you want with it. Share freely.