Agent Team Kit

Agent Team Kit — SKILL.md

A framework for self-sustaining AI agent teams.

What This Is

A complete team process kit for OpenClaw agents that enables: Self-service work queues — Agents pick up tasks without human bottlenecks Clear role ownership — Everyone knows who does what Continuous discovery — Work flows in automatically Proactive operation — The team runs itself via heartbeat

1. Copy the Process Files

# From your workspace root cp -r skills/agent-team-kit/templates/process ./process This creates: process/INTAKE.md — The 5-phase work loop process/ROLES.md — Role definitions process/OPPORTUNITIES.md — Raw ideas/discoveries process/BACKLOG.md — Triaged work queue process/STATUS.md — Who's working on what

2. Add Heartbeat Config

Merge templates/HEARTBEAT.md into your existing HEARTBEAT.md: cat skills/agent-team-kit/templates/HEARTBEAT.md >> HEARTBEAT.md Or copy it directly if you don't have one yet.

3. Customize Roles

Edit process/ROLES.md to match your team: Rename roles to fit your domain Add/remove specialized execution roles Update the human lead section with your name

The Intake Loop

DISCOVER → TRIAGE → READY → EXECUTE → FEEDBACK ↑ ↓ └──────────────────────────────────────┘ Discover — Find opportunities (Scout role) Triage — Decide what's ready (Rhythm role) Ready — Self-service queue (any agent) Execute — Do the work (assigned agent) Feedback — Learn and spawn new ideas (completing agent)

Core Roles

RoleMissionOwnsScout 🔍Find opportunitiesOPPORTUNITIES.md, discoveryRhythm 🥁Keep work flowingBACKLOG.md, triageHarmony 🤝Keep team healthyUnblocking, retros[Human]Strategic directionHard calls, spawning Execution roles (spawn as needed): Link 🔗 — Builder Pixel 🎨 — Designer Sage 🦉 — Architect Echo 📢 — Voice Spark ✨ — Creative

Self-Service

If it's in Ready, any agent can pick it up. No approval needed.

Clear Ownership

Every phase has ONE owner. No ambiguity.

Always Log

Ideas, discoveries, completions — if you don't log it, it didn't happen.

Spawn, Don't Solo

Main agent coordinates. Sub-agents execute. Don't do everything yourself.

File Structure

process/ ├── INTAKE.md # How the loop works (reference) ├── ROLES.md # Who does what ├── OPPORTUNITIES.md # Raw discoveries (anyone adds) ├── BACKLOG.md # Triaged work (Rhythm maintains) └── STATUS.md # Current activity (self-updated) HEARTBEAT.md # Proactive check triggers

Heartbeat Integration

• Add to your heartbeat checks:
• ### Team Health (run hourly)
• [ ] OPPORTUNITIES.md stale? → Spawn Scout
• [ ] Ready queue empty? → Alert Rhythm
• [ ] Active work stuck >2h? → Nudge owner
• [ ] Any unresolved blockers? → Harmony
• The heartbeat keeps the loop spinning even when the human isn't watching.

Adding a New Role

Define in ROLES.md: Mission (one sentence) Owns (what they're responsible for) Cadence (how often they work) Outputs (what they produce) Update the ownership matrix Add spawn criteria in INTAKE.md if needed

Changing the Loop

The 5-phase loop is flexible. Adapt it: Add validation gates between phases Split EXECUTE into parallel tracks Add approval checkpoints (if your domain requires it)

Anti-Patterns

❌ Human manually adds every task → Use triage role instead ❌ Waiting for permission to pick up work → Ready = fair game ❌ One agent does everything → Spawn specialists ❌ Ideas stay in heads → Log to OPPORTUNITIES.md ❌ Heartbeat just returns OK → Actually check the loop

Metrics (Optional)

Track team health: Cycle time — OPPORTUNITIES → DONE Queue depth — Items in Ready (healthy: 5-15) Stale items — Days since last triage Spawn rate — Sub-agents created per day The system runs itself. Your job is to trust it.