team-orchestrator
If you like it, please support us with a simple share on your social networks — especially on x.com.
Orchestrate agent teams across multiple AI coding CLIs. Define an agent team once, run it on Claude Code (slash commands) and Codex (subagents) — with one source of truth and a contributor-friendly path to add more runtimes.
Languages: English · Português (BR) · 简体中文 · 繁體中文 · 日本語 · 한국어 · Türkçe
The pain
Claude Code agent teams are powerful — but every team means writing YAML by hand, dropping it in the right folder, reloading plugins. Get a team that works? It's stuck in that one repo. Next project: blank page. Want the same team on Codex? Different format, different folder, do it all again.
The fix
Write the team once. Use it anywhere.
Create — start the instruction with One teammate on (Claude) or Use agents on (Codex). The prefix tells the runtime how to spawn the team:
# Claude Code team
/team:new --name reviewers \
--instructions "One teammate on code review with agents focused on correctness, security, and missing tests." \
--model sonnet --size 3
# Claude Code team
/team:new --name designers \
--instructions "One teammate on product design with agents focused on UX, visual hierarchy, and component reuse." \
--model sonnet --size 3
# Codex team (mirrored)
/team:new --name auditors \
--instructions "Use agents on security auditing focused on auth flows, input validation, and dependency risk." \
--model gpt-5.4 --size 3 --target codex
Use in Claude Code — slash or natural language:
/team:reviewers check the auth refactor in PR #42
/team:designers redesign the dashboard empty state
# or just:
spawn the reviewers team on the current diff
have designers propose 3 variants for the pricing page
Use in Codex — same teams, mirrored automatically:
Use the reviewers subagent on this branch.
Spawn designers and have them critique the new onboarding screens.
One definition. Both runtimes. Saved forever.
Why
Every modern AI coding CLI is shipping its own multi-agent / sub-agent system: Claude Code has agent teams, Codex has subagents, and more are coming (Gemini CLI, Qwen Code, opencode, etc).
The definitions are conceptually similar — a name, a role/instructions, optionally a model and team size — but every CLI uses a different file format, location, and invocation pattern. Keep two of them in sync by hand and you're already losing.
team-orchestrator fixes that: write the team definition once, the plugin emits the right artifacts for every supported runtime.
What it does
| Runtime | Format | Invocation |
|---|
| Claude Code | commands/<name>.md + teams/<name>.md (markdown + YAML) | /team:<name> slash command |
| Codex | ~/.codex/agents/<name>.toml (TOML) | natural-language ("spawn the <name> subagent") |
| Your CLI | your format | your invocation — PRs welcome |
A single command — /team:new --name designers --instructions "..." --model sonnet --size 3 — writes to all enabled runtimes.
Quick Start
Get up and running in under 2 minutes.
Step 1 — Install the plugin
Recommended: Claude Code plugin marketplace (one-line install, auto-updates):
# Inside Claude Code:
/plugin marketplace add https://github.com/kellvembarbosa/team-orchestrator
/plugin install team@team-orchestrator
After install, the /team:* commands are available. If they don't appear, run /reload-plugins.
Alternative: local development (--plugin-dir) — best for hacking on the plugin or running an unreleased branch:
git clone https://github.com/kellvembarbosa/team-orchestrator.git
claude --plugin-dir ./team-orchestrator
Alternative: fully manual install — clone into your user Claude config:
git clone https://github.com/kellvembarbosa/team-orchestrator.git ~/.claude/plugins/team-orchestrator
# Restart Claude Code, then verify with /plugin list.
# Windows PowerShell — manual install
New-Item -ItemType Directory -Force -Path "$HOME/.claude/plugins" | Out-Null
git clone https://github.com/kellvembarbosa/team-orchestrator.git "$HOME/.claude/plugins/team-orchestrator"
# Restart Claude Code, then verify with /plugin list.
