The professional Claude Code setup you'd spend 3 months discovering on your own — working in 30 minutes.
Fork → open Claude Code → run /start-project "your idea" → 30 minutes later you have:
- A structured project with multi-agent running in parallel
- Complete feature lifecycle (start → ship → close → repeat)
- Opinionated CLAUDE.md with best practices baked in
- CI/CD on GitHub Actions (lint + validation + branch protection)
- Institutional memory system (HANDOVER + LEARNINGS + MEMORY)
The /start-project research phase with 4 parallel agents is itself a demo of the multi-agent power — you see it work on the first interaction, without configuring anything.
# 1. Fork this repo on GitHub ("Use this template" button) # 2. Clone your fork git clone https://github.com/YOUR_USERNAME/YOUR_PROJECT.git cd YOUR_PROJECT # 3. Open Claude Code and kick off your project claudeThen in Claude Code:
/start-project "your idea here" That's it. Claude will research your market, validate assumptions, plan the architecture, and bootstrap the project — asking only what it can't infer.
| Skill | What it does |
|---|---|
/start-project | Discovery → research → plan → bootstrap (this skill) |
/start-feature | Intake + research → plan → worktree + execution |
/ship-feature | Commit + rebase + PR + CI + smoke test |
/close-feature | Documentation (HANDOVER, MEMORY, LEARNINGS) + cleanup |
/handover | Summarize session and prepend to HANDOVER.md |
/sync-skills | Pull latest skills from upstream template |
.claude/settings.json— Hooks: lint Markdown before every write.claude/rules/— Modular rules imported by CLAUDE.md (git, coding style, security).github/workflows/ci.yml— Lint + JSON validation + structure check on every PR.github/workflows/bootstrap.yml— Auto-applies branch protection on first push (fork only).github/workflows/template-sync.yml— Weekly PRs with upstream skill updatesMakefile—make check,make lint,make validate,make sync-skills,make clean
memory/MEMORY.md— Persistent context loaded each session (architectural decisions, key files)HANDOVER.md— Session history (newest at top)LEARNINGS.md— Technical gotchas and non-obvious behaviors
- Fill in
CLAUDE.md— replace the<!-- TODO -->sections with your project specifics - Update
CODEOWNERS— replace@rmolineswith your GitHub username - Set GitHub secrets — add any secrets your project needs (document in
.env.example) - Run
make check— validate that everything is wired up correctly
/start-feature "feature name" # Research + plan → /clear # Clean context → /start-feature "feature name" # Execute in worktree /ship-feature # PR + CI + merge /close-feature # Docs + cleanup Each /clear between phases prevents hallucination — outputs are saved to .claude/feature-plans/<name>/ so phases can read them without conversation memory.
Your fork gets weekly PRs from this template via template-sync.yml. Review and merge to stay current. Or run manually:
make sync-skills- Zero deps in core — no CLI to install, no runtime to maintain
- Fork-native — "Use this template" is the install command
- The skills are the product — everything else is scaffolding
- Opinionated but not prescriptive — the template suggests, your CLAUDE.md decides
Found a better pattern? Discovered a pitfall we should document? PRs welcome.
See LEARNINGS.md for the kind of content that belongs in this project.
MIT — fork freely, ship fast.