A Straw-Hat crew of Claude Code agent skills — each one a named crewmate with a job to do. Hand the crew a plan and they'll build it, navigate it to a green PR, plate the design, and chart the codebase.
Every skill is a self-contained SKILL.md. Claude Code loads them automatically and invokes them by name (e.g. /luffy, /nami) or when a task matches the skill's description.
Asking an agent — every time you start your dev workflow — to "explore a codebase" or "fix a PR comment" or X, Y, or Z is time-consuming and error-prone. Abstractions are easier to remember, quicker to invoke, and more consistent, and you can chain abstractions together to form an agentic dev workflow end to end.
And abstractions can be anything you like or enjoy — your crew could be the Ninja Turtles, Marvel characters, or the cast of Friends — injecting some fun into development. Fork and make your own crew or characters, then PR to link it in Make your own crew. 🏴☠️
| Skill | Role | What it does |
|---|---|---|
| 🧢 luffy | Captain — Builder | Autonomously implements a planned feature end-to-end, then drives it through a QA + Design + Engineer review loop until clean, verifies it runs, and ships a green PR. |
| 🍊 nami | Navigator — Shipper | Takes a branch or PR to a green, reviewed state: pushes, opens/updates the PR, and loops on CI + review comments — fixing real issues and pushing back on wrong ones. |
| 🍳 sanji | Cook — Design Reviewer | Reviews UI work as a senior product designer against the project's design system, returning severity-ranked findings and a ship/iterate/block verdict. |
| 🎯 usopp | Sniper — Architect | Deeply explores a codebase, understands its mechanics, and produces or updates MECHANICS.md as a living architectural reference. |
Hand luffy a plan or spec (a file path or an inline description) and it builds the feature like a senior engineer, self-reviews through an iterative QA + Design/Product + Engineer loop, proves it actually runs with /verify, takes it to a green reviewed PR via /nami, and finishes with a standup-style session log.
Triggers: /luffy <plan> · "run luffy on this plan" · "autonomously implement this feature" · "build this and review it until clean"
nami takes the helm on a branch or PR. It pushes, opens or updates the PR, then loops on the two external signals — CI results and review comments (bots + humans) — critically evaluating each: fixing real problems and pushing back on wrong suggestions, until checks are green and nothing blocking is unresolved. Also invoked in-context by luffy's ship phase.
Triggers: /nami · "take this to a green PR" · "get CI green" · "address the PR comments" · "babysit this PR till it's green"
sanji reviews UI work — a PR, a diff, or a component — as a senior product (UX/UI) designer. It judges interaction quality, visual consistency, accessibility, content, and states against the repo's own DESIGN.md (binding source of truth; falls back to general craft when none exists), and returns severity-ranked findings with a ship / iterate / block verdict. Design only — it defers correctness bugs to /code-review and luffy.
Triggers: /sanji <pr#|path> · "review the design" · "sanji review this PR"
usopp is a senior staff engineer who deeply explores a codebase to understand its architecture, design decisions, and mechanics, then produces or updates MECHANICS.md as a living architectural reference. First run generates the doc; subsequent runs read it first and update only what has changed.
Triggers: /usopp · "document the architecture" · "update MECHANICS.md"
usopp → chart the codebase (MECHANICS.md)
luffy → build the feature from a plan
sanji → review the design of the UI changes
nami → drive the PR to green and ship it
luffy already calls sanji, nami, and /verify internally during its review and ship phases — so for a full feature you can often just set sail with /luffy <plan>.
These are user-level Claude Code skills. Drop each skill folder into your skills directory:
# clone, then symlink (or copy) the crew into your Claude Code skills dir
git clone https://github.com/<you>/agentic-onepiece.git
cd agentic-onepiece
for skill in luffy nami sanji usopp; do
ln -s "$PWD/skills/$skill" "$HOME/.claude/skills/$skill"
doneOr copy them instead of symlinking:
cp -R skills/* "$HOME/.claude/skills/"Restart Claude Code (or start a new session) and the crew will be available by name.
agentic-onepiece/
├── README.md
├── assets/
│ └── banner.svg
└── skills/
├── luffy/SKILL.md
├── nami/SKILL.md
├── sanji/SKILL.md
└── usopp/SKILL.md
The crew is just an abstraction — the names are yours to choose. Prefer the Ninja Turtles, the Avengers, or the Friends gang? Fork the repo and rename the skills to a crew you'll actually enjoy invoking.
- Fork this repo.
- Copy a skill folder under
skills/and rename it to your character (e.g.skills/leonardo/SKILL.md). - Edit the
SKILL.md— update thename,description, and triggers so Claude Code invokes it by your character's name. - Keep the job, swap the costume — a skill's role (builder, shipper, design reviewer, architect) is what matters; the persona is flavour.
- Open a PR to add your crew to the table below.
| Crew | Theme | Author |
|---|---|---|
| 🏴☠️ Straw Hats | One Piece | this repo |
| your crew here | — | open a PR |
“I'm gonna be King of the Pirates!” 🏴☠️