Claude Dev Skill — Testing skill for Claude Code

/dev — AI-Assisted Multi-Agent Development SOP for Claude Code

[中文版](./README.zh.md)

[](LICENSE) [](https://claude.com/claude-code) [](https://docs.anthropic.com/en/docs/claude-code/slash-commands)

A custom skill for [Claude Code](https://claude.com/claude-code) that turns Claude into a **Tech Lead** coordinating multiple AI Worker Agents through a complete software development workflow — from requirements alignment to PR merge.

---

Why /dev?

Most AI coding tools give you a smart autocomplete. `/dev` gives you an **engineering process**.

Without /dev	With /dev
Claude writes code directly in chat	Claude acts as Tech Lead — never writes code itself
No structure, easy to lose track	6-phase SOP from PRD to merge
Single-threaded, one thing at a time	Multiple Worker Agents develop in parallel worktrees
You catch conflicts at merge time	Pre-coding conflict scan catches overlaps before they start
Security review is manual	bandit + pip-audit / npm audit run mandatorily before every Review

---

How It Works

You: /dev I want to build a Todo app with user auth and task management

Claude (Tech Lead)
  │
  ├─ Phase 0 ── Classify request → New Project
  ├─ Phase 1 ── Align on PRD (2-round confirmation)
  ├─ Phase 2 ── Architecture decisions + GitHub Issues
  │
  ├─ Phase 3 ── Spawn Worker Agents (parallel worktrees)
  │              ├─ Worker A: Auth module
  │              ├─ Worker B: Task CRUD API
  │              └─ Worker C: Frontend components
  │
  ├─ Phase 3.5 ── QA Agent static verification
  ├─ Phase 4 ── Code Review + merge (7-item checklist)
  └─ Phase 5 ── Retro + next iteration loop

---

What It Does

`/dev` is a **multi-agent software development SOP** covering:

• **Phase