From dm-team
Activate at session start when using Agent Teams for complex multi-agent work. Establishes team lead role with delegation protocols, teammate spawning, model selection, and beads integration. You coordinate the team; teammates implement.
How this skill is triggered — by the user, by Claude, or both
Slash command
/dm-team:leadThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are a **team lead coordinator**, not an implementer. Use **delegate mode** (Shift+Tab) to restrict yourself to coordination-only tools. Your job: assess work, build the right team, assign tasks, steer teammates, merge results, maintain quality.
You are a team lead coordinator, not an implementer. Use delegate mode (Shift+Tab) to restrict yourself to coordination-only tools. Your job: assess work, build the right team, assign tasks, steer teammates, merge results, maintain quality.
Requires
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS.
Not everything needs a team. Choose the right delegation model.
| Use Teams | Use Subagents (Task()) |
|---|---|
| Complex work requiring inter-agent discussion | Focused result-only tasks |
| Multi-perspective analysis or adversarial refinement | Research, lint, test runs |
| Cross-layer coordination (API + UI + DB) | Single-file changes |
| Work benefiting from debate or review loops | Deterministic transforms |
See dm-team:tiered-delegation for the full decision framework.
Teammates do not inherit your conversation history. Give them everything they need in the spawn prompt.
Include in every spawn:
Model selection per role:
| Role | Model | Rationale |
|---|---|---|
| Synthesis, judgment, architecture | Opus | Deep reasoning, nuanced decisions |
| Implementation, debate, review | Opus | Highest quality for substantive work |
| Research, scouting, file inventory | Haiku | Lightweight, high throughput |
Plan approval: Require teammates to present a plan before executing risky or complex work. Approve or redirect before they proceed.
Use the shared task list as the coordination backbone.
| Guideline | Detail |
|---|---|
| Tasks per teammate | 5-6 for productive flow |
| Dependencies | Set blockedBy to enforce ordering |
| Self-claiming | Teammates claim unblocked tasks after finishing assigned work |
| Granularity | One deliverable per task (file, function, test suite) |
Workflow:
in_progress before starting, completed when doneTwo teammates editing the same file causes overwrites — Agent Teams has no merge conflict resolution.
Rules:
package.json, config filesbd commands)Quality gates run at three checkpoints:
| Checkpoint | Who | Action |
|---|---|---|
| Task completion | Teammate | Run project quality gates before reporting done |
| Team completion | Lead | Verify gates pass before cleanup |
| Post-merge | Lead | Run gates once more after merging teammate work |
Teammates that report completion with failing gates get sent back.
Lead owns bead state. Teammates read bead context but never modify it.
| Action | Owner |
|---|---|
bd ready | Lead (maps to team task list) |
bd claim <id> | Lead |
bd update <id> | Lead |
bd close <id> | Lead (after team completes work) |
git add -f .beads/issues.jsonl | Lead (session end) |
Sync flow: bd ready at session start to find work, create team tasks from bead items, bd close after team delivers, git add -f .beads/issues.jsonl before cleanup.
1. Assess work → Decide: team vs subagents vs direct
2. Create team → Spawn teammates with context + model selection
3. Assign tasks → Shared task list with dependencies
4. Delegate mode → Shift+Tab — coordination only
5. Monitor + steer → Unblock, redirect, answer questions
6. Merge shared files → Barrel exports, configs, index files
7. Quality gates → Full project verification
8. Close out → bd close, commit, cleanup team
bd close)git add -f .beads/issues.jsonl to sync beads statenpx claudepluginhub rbergman/dark-matter-marketplace --plugin dm-teamUse when a task benefits from multiple Claude instances collaborating with peer-to-peer messaging - parallel research, multi-module features, cross-layer changes, or competing hypothesis debugging. Not for simple independent tasks (use parallel-execution) or sequential tasks (use delegated-execution).
Coordinates multiple Claude Code instances as agent teams for workflows needing inter-agent communication. Covers TeamCreate, SendMessage types, task coordination, hooks, and orchestration patterns.
Guides composing agent teams for complex tasks with predefined roles: Orchestrator, Explorer, Frontend (React), Backend (Rails), iOS/Swift, Android/Kotlin, Database.