Why OFFICINA

OFFICINA is a workshop for working with AI tools in product development. At its core is the method: how to structure work with an AI assistant so it stays predictable and reproducible. Around it — skills, subagents, CLI, and configs, ready to reuse.

Method first. Reusable rules and guides — writing style, CLI authoring, skill & subagent authoring, loop engineering — each readable on its own.
Portable skills, Claude Code at the core. Skills aren't tied to one tool — they run in Claude Code and compatible environments. Subagents and settings are tuned for Claude Code. Both are hand-picked rather than dumped wholesale.
Environment around the core. VS Code as an applied section that supports the core rather than being the headline.

Who it's for

You work with an AI assistant and want a system — a method, not one-off prompts.
You write your own skills, subagents, CLI — take battle-tested patterns and guides.
You set up your working environment for AI development on macOS.

If you're after a plug-and-play, install-and-forget distribution — this isn't it: OFFICINA is about method and deliberate selection, not a ready-made turnkey build.

What's inside

Section	What it holds
`skills/`	Portable skills — reusable modes and procedures (Claude Code and compatible environments)
`claude/`	Claude Code layer — subagents and settings
`methodology/`	Rules and guides — how to work
`cli/`	Personal CLI tools and shared libraries
`vscode/`	VS Code settings, keybindings, curated extensions
`install/`	Install and layout scripts

How the method works

The starting point is one question — what to solve the task with:

A deterministic action, no model needed — CLI (e.g. jig scaffolds a new repository in one command)
Need to read the context and decide by situation — skill
An isolated subtask or a persona role — subagent
A recurring cycle on a schedule — loop

Then comes the matching guide in methodology/: structure, conventions, a pre-deploy checklist. The method is shared; the artifact carries over between projects.

jig is the method turned into a tool: in one command it scaffolds the very doc structure and Claude setup that the guides in methodology/ describe.

Quick start

Requires macOS. Clone the repo and open the section that fits your need — each ships its own README and, where relevant, a one-command installer under install/:

git clone https://github.com/KirKruglov/OFFICINA.git
cd OFFICINA

The first ready installer sets up the VS Code environment:

./install/vscode.sh   # settings, keybindings, MesloLGS Nerd Font, curated extensions

Philosophy

The right tool for the task. Deterministic work goes into a script, context-driven decisions to the model. An extra model in the loop adds cost and uncertainty.
Reliability rests on constraints. Boundaries, budgets, and checks hold the system together. An autonomous loop needs something able to say "no".
Knowledge lives in the artifact. A guide, skill, or subagent gets reused; what's worked out once isn't worked out again every time.
Hand-picked. What lands in the repo is proven in practice and deliberately chosen.

Contributing

Issues, questions, and small fixes are welcome — see CONTRIBUTING.md. OFFICINA is an open project.

Useful? Star it — that's how the next developer finds the project. Fork it, take what fits, and adapt it to your own setup.

Officina

README