Lean Coding Workspace

Turn your coding agent into a teammate that follows a process.

A ready-made workspace for your AI coding agent: one lean instruction file plus a set of skills that take every change from idea to shipped code the same way — shape, spec, build, test, document, release.

Quick start

Open your coding agent anywhere and paste:

Fetch https://raw.githubusercontent.com/voxlo-dev/Lean-Coding-Workspace/main/INSTALL.md and follow it.

The agent installs everything, asks a few questions about you and your machine, and tells you when to restart it. You need Git and an agent tool with a folder in adapters/. Prefer doing it by hand? Follow INSTALL.md. Verified on Windows; Linux and macOS reports welcome.

It replaces your agent's global setup and is not compatible with other workspace frameworks or project doc layouts (details). Read SECURITY.md first.

Why you want it

• Plain Markdown, no scripts. Nothing to build, nothing running in the background — every file is an instruction your agent reads.
• A curated toolkit, not a zoo. Two helpers, picked for the job: codegraph answers "how does this work" from a code graph instead of reading half the repo, and context7 brings current library docs instead of stale recall.
• Your workflow, your choice. Use the built-in workflows, or add others from a curated skill catalog — superpowers, OpenSpec, BMAD and more install straight into the workspace. Every skill is yours to change.
• Agile, the way you know it from work. Backlog, sprints, a board, reviews and releases — the agent keeps the rhythm, you make the calls.
• Nothing gets forgotten. Knowledge moves up in stages — a short handout between chats, tickets for open work, a folder per sprint, lasting docs for what shipped, memory for the rest.
• Domains on demand. Specialist skills and knowledge for Unity, Android, web or academic writing load only in the projects that need them. Build your own with /domain-init.
• Light on context — made for the small plan. About 2.3k tokens are always loaded; everything else loads only when used, so a small subscription goes a long way.
• The right model for each step. Hand single steps — a build package, an end-to-end test — to another agent tool or a local model, while your main chat keeps the overview. /dispatch-configurator finds what your machine can run.
• Runs in any agent tool. New tool? The harness-onboard skill lets your agent probe it and work out the differences by itself.
• Clean split between repo and install. This repo is the source you fork and edit; the installed workspace holds only what your agent actually reads. No dead weight — every file has a job.

How a session goes

Where everything lives

All skills

Alphabetical within each section.

Set up

Skill	What it does
dispatch-configurator	records which agent tool and model each dispatched step runs on, per machine
domain-init	builds a domain for one kind of work
project-init	onboards a new or existing project: scaffold, docs, test framework, gitignore

Plan

Skill	What it does
open-sprint	plans the next sprint, pulls tickets onto the board, opens the branch
shape	turns a fuzzy idea or brainstorm into concrete tickets
spec-design	brainstorms a feature, picks its test strategy, writes the spec
ui-design	look and feel: colors, typography, layout, styleguide, mockups

Build

Skill	What it does
dynamic-workflow	takes a feature from approved spec to tested, documented code
e2e	end-to-end tests that grow into a reusable script
minimal-workflow	one small change or bugfix, no spec

Close and ship

Skill	What it does
close-sprint	clears the board, distils the sprint into docs, merges to main
release	changelog, version bump, security gate, release PR, ship and tag

Keep current

Skill	What it does
checkpoint	ends a chat with a short handout for the next one
maintain-docs	keeps docs in line with the code
maintain-memory	curates memory and prunes what's no longer true

Only in this repo

Run from your clone; they work on the workspace itself and are not installed into projects.

Skill	What it does
compress	keeps the Markdown you edit here short
harness-onboard	adds an agent tool the workspace has no adapter for
workspace-sync	installs or repairs the workspace in your agent tools

Make it yours

Fork the repo, edit workspace_TEMPLATE/, and re-run the workspace-sync skill from your clone — it also repairs a broken install. Your personal rules in the installed AGENTS.md survive every sync.

Contributing & license

Fixes, adapters and generally useful improvements are welcome — personalised or niche features are not; see CONTRIBUTING.md. Working on the repo starts at AGENTS.md.

MIT © Leonard Müller