Yarstack
yarlson/yarstack
Workflow disciplinado para agentes de IA codificarem com qualidade e segurança
Ver no GitHubStack de workflow para agentes de IA disciplinados em codificação. Estrutura o trabalho em passes repetíveis: framing de decisões, planejamento seguro, implementação com controle rigoroso de escopo, revisão de qualidade, shipping via CI e documentação de contexto. Mantém engenheiros no controle total do código e da arquitetura.
Instalação
npx convez-ai yarlson/yarstack1 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
- Estruturar requisitos ambíguos em planos de implementação claros
- Revisar código e riscos de infraestrutura antes do merge
- Documentar jornadas críticas e contexto do repositório automaticamente
- Implementar com strict scope control e CI feedback
- Coordenar múltiplos agentes de IA em workflows de engenharia
Veredito

Yarstack é um framework robusto para disciplinar agentes de IA em fluxos de desenvolvimento real. Bem adequado para times que querem automação sem abdicar controle de qualidade ou segurança.
README
A tradução para o português deste README ainda está na fila.
Yarstack
Yarstack is a workflow stack for disciplined coding agents. It helps an agent frame ambiguous work, plan safely, implement with strict maintainability standards, review quality and risk, ship through CI, and leave useful project context behind.
Built for engineers who want agent help without losing control of scope, code quality, or dirty worktrees.
Shipping skills check dirty-worktree scope and stage only confirmed paths by default.
Workflow Promise
Yarstack turns broad requests into repeatable engineering passes:
- Frame product and architecture decisions before code.
- Plan work in reviewable phases.
- Implement with strict scope control.
- Review code, infrastructure, and maintainability risks.
- Ship with CI feedback.
- Document critical journeys and repo context.
Install Matrix
| Surface | Status | Install path |
|---|---|---|
| Individual skills | Available now | npx skills add https://github.com/yarlson/skills/tree/main/<skill-name> |
| Full Yarstack core plugin | Available now | claude plugin install yarstack-core@yarstack |
| Claude plugin marketplace | Available now | claude plugin marketplace add yarlson/skills |
| Direct repo clone or vendoring | Available now | Clone this repo and copy the skill folders your agent host supports |
Install Claude Plugin Marketplace
claude plugin marketplace add yarlson/skills
claude plugin install yarstack-core@yarstack
Install Individual Skills
Core:
npx skills add https://github.com/yarlson/skills/tree/main/frame
npx skills add https://github.com/yarlson/skills/tree/main/plan
npx skills add https://github.com/yarlson/skills/tree/main/implement
npx skills add https://github.com/yarlson/skills/tree/main/review
npx skills add https://github.com/yarlson/skills/tree/main/infra-review
npx skills add https://github.com/yarlson/skills/tree/main/ship
npx skills add https://github.com/yarlson/skills/tree/main/journey-docs
npx skills add https://github.com/yarlson/skills/tree/main/repo-context-docs
Integrations:
npx skills add https://github.com/yarlson/skills/tree/main/coderabbit
Personal utilities:
npx skills add https://github.com/yarlson/skills/tree/main/create-todo
Deprecated:
npx skills add https://github.com/yarlson/skills/tree/main/ci
ci is deprecated from the public flagship path because it stages all changes and pushes the current branch directly. Use ship for PR-based shipping; keep ci only for intentional direct-push flows.
Current Skill Map
Core
| Skill | Workflow role | What it does |
|---|---|---|
frame | Frame | Force clear product and system design decisions one question at a time |
plan | Plan | Write implementation plans that prevent structural regressions, wrong-layer logic, file sprawl, and vague phases |
implement | Implement | Implement code under an extremely strict maintainability bar so strict review finds no structural blockers |
review | Review | Review code for security, bugs, performance, regressions, and missing tests |
infra-review | Review infra | Review IaC for network exposure, IAM, destructive changes, cost, and deploy risk |
ship | Ship with PR | Check scope, branch, stage confirmed paths, commit, push, open a draft PR, wait for checks, and fix red builds |
journey-docs | Document journeys | Create concise Markdown journey, flow, acceptance, telemetry, and coverage artifacts |
repo-context-docs | Document repo | Create or update current-state docs under docs/context/ |
Integrations
| Skill | Workflow role | What it does |
|---|---|---|
coderabbit | Integration | Triage CodeRabbit review comments, fix legitimate issues, push, and resolve threads |
Personal Utilities
| Skill | Workflow role | What it does |
|---|---|---|
create-todo | Personal utility | Capture concrete work as committed Markdown todos grouped by project |
Deprecated
| Skill | Workflow role | What it does |
|---|---|---|
ci | Ship direct | Deprecated public flagship path. Stages all changes and pushes the current branch directly. Use ship for PR-based shipping. |
How Skills Work
Each skill is a folder with a SKILL.md file: YAML frontmatter plus a prompt that defines the agent's workflow. Some skills also include scripts or templates for deterministic operations.
Install the specific skills you want, then invoke them by describing the work. Examples: "review my changes", "write an implementation plan", "review my infra code", or "document this critical user journey".
Validate Repo
scripts/validate-skills.sh
Run this before release-oriented skill changes. It checks skill directory shape, SKILL.md frontmatter names, per-skill README titles and install links, root README references, and core plugin skill paths.