Coding Crew
ypxing/coding-crew
Agentes paralelos para automação do ciclo ideia → código → review
Ver no GitHubConjunto distribúvel de agentes IA que automatiza o ciclo completo de desenvolvimento: da ideia ao código, passando por planejamento, implementação paralela em TDD e revisão de código. Integra-se nativamente com Claude e GitHub Copilot para orquestrar múltiplos coders em worktrees isoladas.
Instalação
npx convez-ai ypxing/coding-crew1 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
- Acelerar sprints com múltiplos agentes codificadores trabalhando em paralelo
- Garantir qualidade através de TDD automático e revisão de código integrada
- Transformar ideias em issues refinadas e prontas para implementação
Veredito

Solução ambiciosa para automação end-to-end do ciclo de desenvolvimento. Requer shell/git profundo e é voltada a equipes que já dominam DevOps e CI/CD; o modelo paralelo de coders é inovador mas requer clareza nas issues para evitar conflitos.
README
A tradução para o português deste README ainda está na fila.
Coding Crew
A distributable collection of AI agents and skills that automate the issue → implementation → review cycle.
The flow
you have an idea
│
▼
┌─────────────────────────────────────────────────────┐
│ Create issues │
│ │
│ auto: /plan-sprint │
│ │
│ manual: /grill-me (or /grill-with-docs) │
│ → /to-prd → /to-issues │
│ (or write .scratch/.../issues/*.md) │
└─────────────────┬───────────────────────────────────┘
│
▼
/afk-sprint
│
┌────────────┼────────────┐
▼ ▼ ▼
coder coder coder (parallel, isolated worktrees)
issue 1 issue 2 issue 3
└────────────┼────────────┘
│ committed branches merged
▼
code-reviewer
│
▼
/address-code-review
1. Install
Prerequisites: git, jq
Quick (no clone required)
Install everything to $HOME (user-level, works in any project):
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/bootstrap.sh | bash
Claude only:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/bootstrap.sh | bash -s -- claude
Copilot only:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/bootstrap.sh | bash -s -- copilot
Specific skills only:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/bootstrap.sh | bash -s -- claude --skills tdd,caveman,grill-me
Into the current project instead of $HOME:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/bootstrap.sh | bash -s -- --project
From a local clone
# Into the current project
./install.sh claude
# Into a specific project
TARGET_REPO=/path/to/your/project ./install.sh claude
# Into $HOME for all projects (user-level)
./install.sh --user claude
2. Create issues
Auto — from an idea
/plan-sprint ← grill → PRD → issues in one flow
Manual — step by step
/grill-me ← stress-test your idea interactively
(or /grill-with-docs — challenges against your domain model, creates one if you don't have it)
/to-prd ← turn the refined idea into a PRD
/to-issues ← break the PRD into ready-for-agent issues
Or write issues directly. Create .scratch/<feature>/issues/01-slug.md:
Status: ready-for-agent
## What to build
One paragraph describing the end-to-end behavior.
## Acceptance criteria
- [ ] Criterion one
- [ ] Criterion two
## Blocked by
- None - can start immediately
3. Run the sprint
/afk-sprint
Picks up every ready-for-agent issue, spawns coder agents in parallel, commits, loops until done. Runs a code review pass on exit.
4. Address the review
/address-code-review
Opens the review report, triages findings, implements fixes with TDD.
More install options
./install.sh # everything, all platforms, into project
./install.sh --user # everything into $HOME (user-level)
./install.sh claude # all agents + skills for Claude Code
./install.sh copilot afk-sprint # afk-sprint for GitHub Copilot
./install.sh claude --skill grill-me # a standalone skill only
./install.sh claude --skills tdd,caveman,grill-me # multiple skills at once
./install.sh --user claude --skill tdd # one skill into $HOME/.claude/skills/
./install.sh --update # update changed agents/skills in place
Set TARGET_REPO=/path/to/other/repo to install into a repo other than the current directory.
Uninstall
Quick (no clone required)
Remove everything from $HOME:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/unbootstrap.sh | bash
Remove specific skills only:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/unbootstrap.sh | bash -s -- --skills tdd,caveman
Remove from the current project instead of $HOME:
curl -fsSL https://raw.githubusercontent.com/ypxing/coding-crew/main/unbootstrap.sh | bash -s -- --project
From a local clone
./uninstall.sh --user # remove all from $HOME
./uninstall.sh --user --skills tdd,caveman # remove specific skills from $HOME
./uninstall.sh --user --agent coder # remove a specific agent
./uninstall.sh # remove all from current project
Skills
In order of use:
| Skill | When |
|---|---|
/plan-sprint | Full design pipeline — grill → PRD → issues in one automated flow |
/grill-me | Stress-test your idea interactively (step-by-step alternative to /plan-sprint) |
/grill-with-docs | Same, but challenges against your domain model — creates one if you don't have it |
/to-prd | Turn the refined idea into a PRD |
/to-issues | Break the PRD into ready-for-agent issues |
/solve-issue | Implement one issue manually (what coder uses internally) |
/address-pr-comments | After a PR review — implement sensible comments with TDD |
/improve-codebase-architecture | Ongoing — find refactoring opportunities |
/caveman | Switch to ultra-compressed communication to reduce token usage ~75% |
Guides
- Consumer guide — full setup, issue lifecycle, troubleshooting
- Contributor guide — adding agents/skills, registry schema, security rules
Acknowledgements
Several skills in this repo are borrowed directly from Matt Pocock's skills collection (MIT License, Copyright © 2026 Matt Pocock). See LICENSE for the full copyright notice. Thanks Matt.
The karpathy-guidelines skill is derived from multica-ai/andrej-karpathy-skills, itself based on Andrej Karpathy's coding observations.