Claude Code Kickstarter Logo

The one-file setup that drops a full "plan → tests → code → docs → PR" workflow into any Git repo—powered by Claude Code.

Inside this repo you'll find one magic Markdown file:

.claude/commands/setup-claude-workflow.md

Copy that file into any existing project (monorepo, single package, JavaScript, Bun, TypeScript—doesn't matter). Then run:

/user:setup-claude-workflow "My Awesome Project"

Claude will:

1. Detect which canonical docs you already have.

2. Write the missing ones (architecture diagram stub, glossary, API guide, ADR README, onboarding checklist).

3. Drop a battle-tested CLAUDE.md that enforces:

   • Branch → Plan → Tests → Code → Verify → Docs → Changelog → PR

4. Install repo-local helper commands (/project:regen-docs, /project:next, etc.).

5. Merge a minimal GitHub Actions workflow if you don't have one.

6. Stage everything with git add and politely remind you to commit.

No git init, no history rewrite, no yak-shaving.

# Option A – install once globally
mkdir -p ~/.claude/commands
curl -o ~/.claude/commands/setup-claude-workflow.md \
 https://raw.githubusercontent.com/gmickel/Claude-Code-Kickstarter/refs/heads/main/setup-claude-workflow.md
# In any repo
cd my-existing-repo
/user:setup-claude-workflow "My Existing Repo"
git commit -m "chore: add Claude workflow"
# Option B – keep the file in the repo
mkdir -p .claude/commands
curl -o .claude/commands/setup-claude-workflow.md \
 https://raw.githubusercontent.com/gmickel/Claude-Code-Kickstarter/refs/heads/main/setup-claude-workflow.md
/project:setup-claude-workflow "My Existing Repo"
git commit -m "chore: add Claude workflow"

Then write a PRD and let Claude take the wheel:

echo "# PRD #42 – Super Search" > docs/prd_42.md
/plan take docs/prd_42.md and create plan
continue

Already got a CLAUDE.md? The setup command will not overwrite it. Instead, it will print the contents of the new CLAUDE.md and ask you to review it and merge it into your existing file.

Claude Code asks before it runs any Bash command or reads a file. You can silence the obvious, always-safe commands by whitelisting them in ~/.claude/settings.json.

{
 "permissions": {
 "allow": [
 // 🛠 Package managers & scripts
 "Bash(pnpm:*)",
 "Bash(bun:*)",
 "Bash(npm:*)",
 "Bash(yarn:*)",
 "Bash(pnpx:*)",
 // 🧑‍💻 Git & GitHub CLI
 "Bash(git:*)",
 "Bash(gh:*)",
 // 🧪 Typical test / lint / typecheck tools
 "Bash(vitest:*)",
 "Bash(eslint:*)",
 "Bash(biome:*)",
 "Bash(tsc:*)",
 // 📚 Basic shell queries (safe introspection)
 "Bash(ls:*)",
 "Bash(cat:*)",
 "Bash(mkdir:*)",
 "Bash(cd:*)",
 "Bash(pwd:*)",
 "Bash(which:*)",
 // 🕵️ Diff & search helpers
 "Bash(diff:*)",
 "Bash(grep:*)",
 "Bash(find:*)",
 // 🌐 Curl / wget for fetching docs - use at your own risk
 "Bash(curl:*)",
 "Bash(wget:*)"
 ],
 "deny": []
 }
}

How to customize:

1. Start minimal – add only what you never want to be prompted for.
2. When Claude pauses for permission, decide whether that pattern is safe to whitelist. If yes, copy the pattern (Bash(tool:*)) into your allow list.
3. Remove entries any time – Claude will simply ask again next run.

Tip: Patterns are glob-like. Bash(pnpm:*) covers pnpm install, pnpm run lint, pnpm exec, etc.

Already have a settings file? Just merge the new "allow" patterns into your existing JSON.

Q: My monorepo has many packages—can each one have its own README and CLAUDE.md?

Yep. Doc-keeper treats packages/*/README.md and packages/*/CLAUDE.md as "ask-before-update" docs. Add package-specific tech-stack notes there and Claude will respect them.

Q: What if I hate Mermaid?

Swap the guideline in CLAUDE.md to PlantUML or draw.io links—Doc-keeper will follow your rule.

PRs welcome! Ideas:

• Add a Bun-native CI template.
• Include a Playwright smoke-test step.
• More custom commands? PR the prompt file.

Just keep prompts short & snappy—Claude likes imperative prose.

Happy plan-test-code-docs looping! If you ship something cool with this workflow, share it with the community. 🎉