AWaC — Agent Workspace as Code

Declare an AI-agent workspace once. Compose it from versioned stack repos. Same .agents/, same CLAUDE.md, same tooling on every machine.

Get started View on GitHub

Why AWaC

Anyone using AI coding agents seriously hits two walls: drift (rules that diverge across projects within weeks) and bloat (every workspace ending up as a 200-file .agents/ no one owns). AWaC solves both with versioned stacks + a declarative manifest + deterministic composition.

One file describes the workspace

A workspace.yml lists the stacks. The CLI resolves them, clones the repos they declare, composes .agents/{rules,skills,workflows}/, and writes CLAUDE.md + AGENTS.md. Lockfile makes it reproducible.

Agent-first by design

Every command emits machine-readable output. Errors are code + category + cause + remediation so AI agents can react programmatically. No interactive prompts in core commands.

100% local, no SaaS

Stacks live in plain GitHub repos. The CLI is pipx-installable. No backend, no telemetry. MIT licensed — fork it, vendor it, embed it.

Quick try

# Install
TAG=$(gh release view --repo getGanemo/workspace-cli --json tagName -q .tagName)
gh release download "$TAG" --repo getGanemo/workspace-cli --pattern '*.whl' --dir /tmp/wsp
pipx install /tmp/wsp/wsp-*.whl

# Compose a workspace
mkdir my-feature && cd my-feature
wsp init my-feature --template blank
wsp bootstrap

Where next

Get started — install + first workspace.
Architecture — manifest, stacks, composition order, lockfile.
CLI reference — every wsp command, every flag.
FAQ — what AWaC is, what it isn’t, when not to use it.
Decisions (ADRs) — recorded design choices.