@long.dg/le-restaurant
v0.2.0
Published
Install the brigade of agent skills into your project, translated for your coding agent.
Readme
le-restaurant 🍽️
A brigade of agent skills, installed into your project and translated for your coding agent.
le-restaurant ships four kitchen-themed skills that take an idea from a napkin
sketch all the way to shipped code — and installs them for the agent you actually
use: Claude Code, Codex, or Gemini. One command, native format.
npx @long.dg/le-restaurantThe brigade
Software, like a restaurant, runs on a brigade — each station owns one job and hands off cleanly to the next. The four skills mirror that line, in order:
| Course | Skill | What it does | |--------|-------|--------------| | 1 | sous-chef | Your thinking partner for a raw idea. Asks short, focused questions, taste-tests the concept, and plates a structured recipe card (problem / solution / inspiration / risks) once the idea is "cooked." | | 2 | mise-en-place | Turns a confirmed idea into a buildable plan — a small cross-linked wiki of scope, architecture, phases, and a development plan. Re-run it to harden the plan over time. | | 3 | maitre-d | Front-of-house orchestrator. Reads the plan, cuts it into per-feature "order tickets," and runs the build one course at a time — firing each ticket, then checking the plate at the pass before the next goes out. | | 4 | chef-de-rang | The station waiter who owns one ticket end-to-end. Builds a single feature test-first (red → green → refactor), writes progress back to the ticket, and hands the plate to the pass. Resumable by design. |
The natural flow is sous-chef → mise-en-place → maître-d → chef-de-rang: cook the idea, plan the build, run service, cook each course. You can also reach for any one of them on its own.
Install
Run it in the root of the project you want the skills in:
# Interactive — pick your agent, tick the skills you want
npx @long.dg/le-restaurant
# Non-interactive — scriptable, no prompts
npx @long.dg/le-restaurant --agent codex --skills all --yes
npx @long.dg/le-restaurant --agent claude --skills sous-chef,mise-en-placeFlags
| Flag | Values | Meaning |
|------|--------|---------|
| --agent | claude | codex | gemini | Target coding agent. Omit it to launch the interactive wizard. |
| --skills | all or a comma list | Which skills to install (e.g. sous-chef,maitre-d). Defaults to a prompt. |
| --yes | — | Skip confirmations; install straight away. |
| [targetDir] | path | Where to install. Defaults to the current directory. |
After install, the CLI prints a summary of every file written, the target agent, and how to use the skills.
What lands, per agent
Each agent has a different idea of what a "skill" is, so each gets a native translation:
Claude Code — passthrough. Each skill is copied verbatim to
.claude/skills/<name>/SKILL.md. Claude Code triggers them on demand — it picks the right skill from its description when a matching task comes up. Full fidelity.Codex — merged into
AGENTS.md. Codex reads one monolithicAGENTS.mdper directory, so each selected skill becomes a delimited section betweenle-restaurantmanaged markers. Re-runs update in place without touching the content you wrote outside the markers.Gemini — imported into
GEMINI.md. Each skill is written to.gemini/skills/<name>.mdand pulled in via an@importline inGEMINI.md, again inside managed markers.
⚠️ Always-on caveat for Codex & Gemini
Codex and Gemini have no skill system — there is no on-demand triggering. The translation degrades to always-on guidance: every installed skill sits in the agent's context all the time, rather than activating only when relevant (as it does in Claude Code). We do this honestly rather than fake a skill system. For the on-demand experience the skills were designed for, use Claude Code.
Re-running the installer is always safe: managed markers and atomic writes mean your own content outside the markers is never clobbered.
Development
npm install
npm run build # bundle the CLI with tsup
npm test # run the vitest suite
npm run typecheck # tsc --noEmit
npm run check:sync # verify vendored skills/ match canonical .claude/skills/The four skills live canonically in .claude/skills/ and are vendored into
skills/ for distribution. npm run check:sync (also a CI gate) fails if the
two ever drift.
License
MIT
