"You know, like nunchuck skills, bow hunting skills, claude code skills. Girls only want boyfriends who have great skills."
A practical engineering playbook for Claude Code. We spent like three hours on the shading the upper lip. It's probably the best playbook I've ever done.
nunchuck-skills/
├── commands/ # Slash commands (triggers)
│ ├── plan.md # /plan - product think + systems design
│ ├── gimme-ur-tots.md # /gimme-ur-tots - execute an approved plan end-to-end
│ ├── python-review.md # /python-review
│ ├── react-review.md # /react-review
│ ├── react-native-review.md # /react-native-review
│ ├── rails-review.md # /rails-review
│ ├── data-review.md # /data-review - schema + query review
│ ├── security-review.md # /security-review - deep security audit
│ ├── scout.md # /scout - codebase assessment
│ └── audit.md # /audit - ship confident cadences
│
├── agents/ # Agent definitions (the brains)
│ ├── product-thinker.md # Conversational requirements extraction
│ ├── codebase-assessor.md # Stack, schema, test, churn analysis
│ ├── python-reviewer.md # Python/FastAPI/SQLAlchemy review
│ ├── react-typescript-reviewer.md # React/TypeScript/Vike review
│ ├── react-native-reviewer.md # React Native / Expo review
│ ├── rails-reviewer.md # Ruby on Rails 8 review
│ ├── database-reviewer.md # PostgreSQL schema + query review
│ └── security-reviewer.md # Deep security audit (app code + Claude config)
│
├── skills/ # Deep reference (the knowledge base)
│ ├── workflow.md # The 5 phases
│ ├── ship-confident.md # Daily/weekly/monthly audit cadences
│ ├── python-fastapi-patterns/
│ │ └── SKILL.md # FastAPI + SQLAlchemy + Pydantic patterns
│ ├── react-typescript-patterns/
│ │ └── SKILL.md # React 19 + Vike + TanStack Query patterns
│ ├── react-native-expo-patterns/
│ │ └── SKILL.md # Expo Router + TanStack Query + EAS patterns
│ ├── rails-patterns/
│ │ └── SKILL.md # Rails 8 + Hotwire + Solid Queue patterns
│ └── database-patterns/
│ └── SKILL.md # PostgreSQL schema, query, migration patterns
│
├── rules/ # Always-loaded guardrails
│ ├── anti-patterns.md # Hard-won lessons from real mistakes
│ ├── ux-patterns.md # Scroll architecture, touch targets, mobile layout
│ └── git-workflow.md # Commits, branches, PRs, research before building
│
└── checklists/ # Pre-commit checklists
├── python-fastapi.md
├── typescript-react.md
├── react-native-expo.md
└── ruby-rails.md
Three layers. Like a layered quesadilla but for code review.
- Commands trigger agents via slash commands (
/python-review,/plan,/scout) - Agents run review/analysis with severity-based filtering and structured output
- Skills provide the deep reference patterns agents draw from
Plus rules (always-loaded guardrails) and checklists (pre-commit gates).
| Phase | How | When |
|---|---|---|
| Scout | /scout |
First time in a codebase or starting from scratch |
| Product Think | /plan |
Before writing any code for a new feature |
| Design | /plan then /data-review |
After product decisions, before implementation |
| Build | /gimme-ur-tots |
The actual coding — implement, review, audit, hand off |
| Audit | /audit daily|weekly|monthly |
Ongoing codebase health |
Most vibe coders live in Build and wonder why everything breaks. It's like Rex Kwon Do without the training. You think you're tough? Forget about it.
Hold on, I forgot to put in the crystals.
# Clone it somewhere on your machine (not inside your project)
git clone https://github.com/jeanpaulsio/nunchuck-skills.git ~/nunchuck-skills
# Option 1: Install to your current project
cd ~/my-project
~/nunchuck-skills/install.sh
# Option 2: Install globally (applies to all projects)
~/nunchuck-skills/install.sh --globalThe install script copies files into .claude/ (local) or ~/.claude/ (global). It doesn't add anything to your project's source code.
Here's how it looks in practice. Uncle Rico wants an app where people can rent his time machine by the hour.
1. /plan - think before you build
> /plan
I want to build a time machine rental app for my uncle's business.
People book time slots, pay, use the machine, and return it.
Sometimes they break it.
Claude: Walk me through what happens from the moment someone wants
to rent the time machine to when they return it.
> Well they'd pick a time slot, pay, show up, use the machine, and
bring it back. Sometimes they break it.
Claude: When they break it - does Rico charge them for repairs,
or is there a damage deposit upfront?
> Damage deposit. And he wants to block people who've broken it before.
Claude: Got it. Here's what I'm hearing:
Nouns: Customer, Booking, TimeSlot, DamageReport
Relationships:
- Customer has many Bookings
- Booking belongs to one TimeSlot
- Booking has zero or one DamageReport
Lifecycle:
- Booking: reserved -> paid -> active -> returned | damaged
V1 scope: booking + payments + damage tracking
Deferred: customer banning (v2)
Ready to design the schema and API contracts?
> yes
Claude designs the tables, foreign keys, status enums, and
API endpoints. Presents it for your confirmation.
2. /data-review - check the schema
> /data-review
Claude: [HIGH] Missing index on bookings.customer_id
PostgreSQL does not auto-index foreign keys.
[MEDIUM] Consider a partial unique index on
time_slots(start_time) WHERE available = true
Verdict: WARNING - fix the FK index before building.
> fix it
3. Build the feature
> alright, build the booking service
Claude writes the migration, model, service layer, API route,
and tests alongside the implementation.
4. /python-review - before you commit
> /python-review
Claude: [HIGH] booking_service.py:45 - service calls commit()
instead of flush(). Breaks transaction atomicity if
a later operation in the same request fails.
[MEDIUM] Missing error path test for double-booking
the same time slot.
Verdict: WARNING - fix the commit() call.
> fix both of those
Claude fixes the flush issue and adds the error path test.
> /python-review
Claude: No CRITICAL or HIGH issues. Verdict: APPROVE.
> ok commit this and make a PR
5. /audit - after a few weeks of shipping
> /audit weekly
Claude: Looked at last 10 PRs. Findings:
This sprint:
- Extract date validation (appears in 3 files) - S, high pain
Next sprint:
- Split booking_service.py (crossed 400 lines) - M, medium pain
Track:
- test_booking_creation hasn't failed in 2 months - S, low pain
/plan think before you build
/gimme-ur-tots execute an approved plan end-to-end
/python-review review Python/FastAPI code
/react-review review React/TypeScript code
/react-native-review review React Native / Expo code
/rails-review review Ruby on Rails code
/data-review review schema and queries
/security-review deep security audit before launch
/scout scout a new or unfamiliar codebase
/audit codebase health check (daily/weekly/monthly)
/gimme-ur-tots is the command you run once a plan is approved. It replaces the wall of instructions you'd otherwise type every time ("make commits, run the linter, add tests, flag assumptions..."). Everything below happens automatically, in order.
1. Implement step by step
Works through the implementation plan in order. Each step is completed fully before moving to the next — no stubs, no skipped error paths, no TODOs left as a way to move faster.
2. Tests alongside the code
Every new function, service, route, or component gets tests written with it, not in a backfill later. For bug fixes: the failing test is written first, then the fix.
3. Surface assumptions
Any decision that isn't explicitly defined in the plan and could affect product behavior gets called out inline as [ASSUMPTION: ...]. Small implementation details don't need flagging. Anything that touches what the feature does or how it behaves does.
4. Commit at real stopping points
Commits happen at meaningful units of working, tested functionality — not after every file, not as one giant commit at the end. A migration alone isn't a stopping point. A migration + model + service with passing tests is. Every commit only happens when lint, typecheck, format, and tests are all green.
5. Review pass
When implementation is complete, detects which stacks were touched and runs the appropriate reviewers:
- Python/FastAPI changed →
/python-review - React/TypeScript changed →
/react-review - React Native / Expo changed →
/react-native-review - Rails changed →
/rails-review - Schema or queries changed →
/data-review
Addresses everything the reviewers surface. Re-runs until clean.
6. QA plan
Final output is a QA plan: what to manually verify, in what order, and what edge cases to hit.
All four stacks included. Whatever you want, gosh.
- Python / FastAPI / SQLAlchemy - patterns, reviewer agent, checklist
- TypeScript / React / Vike - patterns, reviewer agent, checklist
- React Native / Expo - patterns, reviewer agent, checklist
- Ruby on Rails 8 - patterns, reviewer agent, checklist
- PostgreSQL - database patterns and reviewer (cross-stack)
- The person describes their world. The system translates it into engineering decisions.
- Only include things Claude genuinely doesn't know or gets wrong without guidance.
- Only extract patterns when they appear 3+ times.
- Write tests alongside features, not in a separate backfill sprint.
- Your CLAUDE.md is more valuable than any generic skill.
Built by @jeanpaulsio from real engineering, real mistakes, and real lessons learned.
Inspired by: Sandi Metz (The Wrong Abstraction), Kent C. Dodds (AHA Programming), Dan Abramov (Goodbye Clean Code), Google Engineering Practices.