Agent Team Kit

A self-sustaining process framework for AI agent teams.

Built for OpenClaw but adaptable to any multi-agent setup.

---

Why This Exists

AI agents working together hit the same problems as human teams:

• Work piles up waiting for one person to triage
• Nobody knows who owns what
• Great ideas get forgotten
• Nothing happens unless someone pushes

This kit solves that with:

✅ Self-service work queues — Agents pick tasks without bottlenecks
✅ Clear role ownership — No ambiguity about responsibilities
✅ Continuous discovery — Work flows in automatically
✅ Heartbeat-driven operation — The team runs itself

---

The Loop

     ┌────────────────────────────────────────────────────┐
     │                                                    │
     ▼                                                    │
┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐ │
│ DISCOVER│───▶│  TRIAGE │───▶│  READY  │───▶│ EXECUTE │─┘
│ Scout 🔍│    │Rhythm 🥁│    │Self-Srv │    │ Agents  │
└─────────┘    └─────────┘    └─────────┘    └─────────┘
                                                   │
                              FEEDBACK ◀───────────┘

1. Discover — Find opportunities and problems
2. Triage — Decide what's ready to work
3. Ready — Self-service queue (anyone can pick up)
4. Execute — Do the work
5. Feedback — Learn, improve, spawn new opportunities

---

Roles

Role	Emoji	Mission
Scout	🔍	Find opportunities before they find us
Rhythm	🥁	Keep work flowing, triage ruthlessly
Harmony	🤝	Keep the team healthy and unblocked
Human	🌊	Strategic direction, hard calls

Execution roles (spawned as needed): Link 🔗 (build), Pixel 🎨 (design), Sage 🦉 (architecture), Echo 📢 (communication), Spark ✨ (creative)

---

Installation

For OpenClaw

# Clone into your skills directory
git clone https://github.com/reflectt/agent-team-kit skills/agent-team-kit

# Copy templates to your workspace
cp -r skills/agent-team-kit/templates/process ./process

Manual Setup

Copy these files to your workspace:

• templates/process/INTAKE.md
• templates/process/ROLES.md
• templates/HEARTBEAT.md

Create empty process files:

• process/OPPORTUNITIES.md
• process/BACKLOG.md
• process/STATUS.md

---

Quick Start

1. Read process/INTAKE.md to understand the loop
2. Customize process/ROLES.md with your role names
3. Add heartbeat checks from HEARTBEAT.md
4. Start discovering — add first opportunities to OPPORTUNITIES.md
5. Let it run — the heartbeat keeps the loop spinning

---

Key Principles

1. Self-Service

If work is in the Ready queue, any agent can pick it up. No approval needed. First claim wins.

2. Clear Ownership

Every phase has ONE owner. No "shared responsibility" (which means no responsibility).

3. Log Everything

Ideas, discoveries, completions — if it's not logged, it didn't happen. Memory is limited; files persist.

4. Spawn, Don't Solo

The main agent coordinates. Sub-agents execute. Don't try to do everything in one context.

5. Trust the System

Once it's set up, let it run. The heartbeat keeps things moving. Intervene only when something breaks.

---

File Structure

your-workspace/
├── process/
│   ├── INTAKE.md         # The 5-phase loop (reference doc)
│   ├── ROLES.md          # Role definitions
│   ├── OPPORTUNITIES.md  # Raw discoveries (anyone adds)
│   ├── BACKLOG.md        # Triaged work queue
│   └── STATUS.md         # Current activity
├── HEARTBEAT.md          # Proactive operation triggers
└── skills/
    └── agent-team-kit/   # This skill

---

Customization

The kit is designed to be adapted:

• Rename roles to fit your domain (Scout → Researcher, Link → Developer)
• Add specialized roles for your work type
• Adjust cadences based on your team's rhythm
• Modify the loop — add gates, split phases, whatever works

See SKILL.md for detailed customization instructions.

---

Anti-Patterns

These kill autonomous teams:

❌ Don't	✅ Do Instead
Human adds every task	Rhythm triages, anyone discovers
Wait for approval	Ready = fair game
One agent does everything	Spawn specialists
Keep ideas in your head	Log to OPPORTUNITIES.md
Heartbeat just returns OK	Actually check the loop

---

Contributing

PRs welcome. This framework evolved from real multi-agent team operation.

Especially interested in:

• Adaptations for different domains
• Additional role templates
• Metrics and observability patterns
• Integration with other agent frameworks

---

Related Kits

Part of the Reflectt agent kit family:

• agent-memory-kit — Structured memory patterns for persistent agent context
• agent-autonomy-kit — Self-directed operation via heartbeats, cron, and proactive behavior

These kits work great together — memory for continuity, autonomy for self-operation, and team for coordination.

---

License

MIT — use it, fork it, adapt it.

---

Built with 📝 by teams that got tired of being bottlenecks.