The land where AI agent teams forge production software.
A package ecosystem for curated teams of AI agents.
AgentTeamLand is a package manager + registry for teams of AI agents. Each team bundles a set of specialized agents (plus their skills and rules) for a specific kind of work — full-stack web apps, data pipelines, mobile-first products, game servers, whatever you need.
Install a team once, and every agent it contains is symlinked into your project's Claude Code directory. Teams can extend other teams, override agents by name, and exclude ones you don't need — so building on someone else's work doesn't mean maintaining a fork.
| Platform | Command |
|---|---|
| macOS / Linux | brew install agentteamland/tap/atl |
| Windows (Scoop) | scoop bucket add agentteamland https://github.com/agentteamland/scoop-bucketscoop install atl |
| Windows (winget) | winget install agentteamland.atl (awaiting Microsoft review) |
| Manual | Download from Releases |
atl install software-project-teamThat's it. The team clones into a shared cache, and every agent / skill / rule it provides symlinks into your project's .claude/. You can now start using /create-new-project, /verify-system, or invoke any of the 13 specialized agents from Claude Code.
Teams are listed in the public registry. Install any of them by short name.
| Team | Status | Stack | Install |
|---|---|---|---|
| software-project-team | ✅ verified | .NET 9 + Flutter + React + Docker | atl install software-project-team |
| starter-extended | example | extends software-project-team | atl install starter-extended |
More teams incoming. Submit yours.
If you're building tooling on top of AgentTeamLand, these are the pieces you work with:
| Package | What it does |
|---|---|
| cli | atl binary — the package manager CLI (Go) |
| registry | Public teams.json — the catalog you see above |
| team-manager | /team skill + bootstrap install.sh (inside Claude Code) |
| core | /save-learnings, /wiki, /create-code-diagram, global rules, JSON schemas |
| brainstorm | /brainstorm skill for structured decision capture |
| rule | /rule + /rule-wizard |
- API is the brain. 99% of business logic lives in the API. Frontend and workers are thin bridges.
- One architecture, every project. Vertical Slice + Clean Architecture + martinothamar/Mediator. No alternatives.
- Docker-only. No local SDK installs. Flutter is the single exception.
- Agents have children.
agent.mdholds identity;children/holds deep, topical knowledge. - Verify everything.
/verify-systemproves the system works end-to-end — containers, ports, pipelines, auth — before any deploy.
If your work needs a different stack, build a team:
- Create a git repo with a
team.jsonat its root (validates against the team schema) - Add agents under
agents/<name>/agent.md, skills underskills/<name>/skill.md, rules underrules/<name>.md - (optional) Extend an existing team with
extends: "software-project-team@^1.0.0", override agents by name, exclude ones you don't want - Open a PR against agentteamland/registry — CI validates schema + URL reachability, maintainers review, your team goes live
Full guide: scaffolder-spec.md.
MIT — each repo in the org has its own LICENSE file.