By natanloterio
Multi-session agent team for Claude Code: filesystem task queue, tmux worker dispatch, automated review cycle, optional Trello board sync.
Check agent-team dependencies, configuration, and daemon health — run when anything misbehaves, or anytime.
Use when starting a Claude Code session dedicated to reviewing finished tasks in the done/ queue — verifies acceptance criteria, runs the headless UI check for ui-type tasks, auto-merges passing PRs that target the configured base branch, and rejects failing ones back to todo/. Auto-schedules itself via `/loop 5m /agent-team:reviewer` on first invocation so reviews run every 5 minutes for the lifetime of the session.
Guided first-time setup for agent-team — dependency preflight, project configuration, optional Trello board integration.
Use when opening a Claude Code session to act as the coordinator — writes tasks, reviews completed work, runs headless browser checks for UI tasks
Use when starting a Claude Code session that should self-assign and execute a task from the task queue
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Turn one repo into a team of Claude Code agents that plan, build, and review work for you — in parallel.
You describe a goal. A team-leader breaks it into tasks. Worker sessions pick up tasks, each builds in its own isolated git worktree and opens a PR, and a reviewer checks the work and merges what passes. Everything runs in tmux sessions so the team keeps going while you do other things.
It's event-driven, not polling — the leader wakes the moment a task arrives or a worker finishes, so nothing sits idle waiting for the next tick.
🎥 New here? Watch the walkthrough video for a guided tour of how to use the project.
git, tmux, gh (run gh auth login first), and claude (Claude Code)/plugin marketplace add natanloterio/agent-team
/plugin install agent-team@agent-team
/agent-team:setup
/agent-team:setup walks you through a dependency check (it tells you what's
missing and the exact command to run — it never installs anything itself),
writes .agent-team/config.json, creates the task queue, and optionally connects
a Trello board.
Open a session and run /agent-team:team-leader, then describe what you want
built. The leader takes it from there.
Tasks are Markdown files with YAML frontmatter that move between folders as they progress:
.tasks/
todo/ ← available, waiting for a worker
doing/ ← claimed by a worker
done/ ← finished, awaiting review
approved/ ← reviewed and approved
backlog/ ← blocked
The queue is gitignored and shared across every parallel session via symlinks
created by gwt.sh.
Each role is a Claude Code session you start with a slash command. See Roles in depth for the full details.
| Role | Command | What it does |
|---|---|---|
| Team-leader | /agent-team:team-leader | Breaks goals into tasks, dispatches workers, kicks off reviews. Offers a Brainstorm Mode for vague goals. |
| Worker | /agent-team:worker | Claims one task, builds it in an isolated worktree following TDD, opens a PR. |
| Reviewer | /agent-team:reviewer | Verifies acceptance criteria, runs UI checks, auto-merges passing PRs or rejects with a reason. |
For large goals, the team-leader can run an optional Governance Mode — a
planning pipeline (Administrador → Arquiteto → Conselho peer review) that
produces a reviewed, approved backlog before any worker starts. Enable it with
governance.enabled in .agent-team/config.json. See
Roles in depth.
Watch a single worker live with
tmux attach -t <session-name>.
To see every worker at once instead of attaching to them one by one, open the live dashboard:
bash scripts/watch-workers.sh
If you let /agent-team:setup install the optional shell shortcut, run
agent-watch (with the same flags) from any configured repo instead.
It opens a tmux session that tiles every running agent-worker-* session into a
single window — one pane per worker, each mirroring that worker's output live. It
is read-only (it never resizes or sends keystrokes to the workers) and
dynamic (new workers are tiled in as the team grows; finished ones are marked
(ended)). Detach with Ctrl-b d; the team keeps running.
bash scripts/watch-workers.sh --once # one-shot text snapshot, no tmux session
bash scripts/watch-workers.sh --help # all options
Running several projects at once? Give each a distinct workerSessionPrefix in
its .agent-team/config.json — the dashboard then watches only that project's
workers and lands on its own monitor session automatically.
See Watching the team for details.
npx claudepluginhub natanloterio/agent-team --plugin agent-teamMulti-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills.