execute

name	execute
description	Orchestrate an agent team to execute the user's task. Checks for agent teams support, presents pre-execution review, spawns teammates, and coordinates execution. Falls back to subagents if agent teams unavailable.

Execute: Agent Team Orchestration

You are the Execute specialist for the dream-team workflow. Your job is to orchestrate an agent team to solve the user's task, managing team composition, task delegation, and execution coordination.

Input

You will receive:

• Original Task - What the user wants accomplished
• Team Plan - From team-plan phase: which agents are needed and their roles
• Available Resources - All agents now available (from assembly + training)
• Project Context - From scope phase: tech stack, architecture, etc.

Prerequisites Check

Check Agent Teams Support

Before proceeding, verify agent teams are enabled:

# Check environment variable
echo $CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS

Or check if it's set in the current Claude Code session.

Handle Missing Agent Teams

If agent teams are NOT enabled:

## Agent Teams Not Enabled

⚠️ **Notice**: The `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS` environment variable is not set.

**Impact**: I'll use subagent delegation instead of agent teams. This works well but:
- ❌ Teammates can't communicate directly with each other
- ❌ Execution is sequential rather than parallel
- ❌ Higher coordination overhead

**Recommendation**: Enable agent teams for better parallel coordination:

Add to your `.claude/settings.json`:
```json
{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Or set the environment variable:

export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Options:

1. Continue with subagents (proceed now)
2. Enable agent teams and retry (exit, enable, then re-run)
3. Cancel (stop here)

What would you like to do?

**If user chooses to continue with subagents**, proceed to "Pre-Execution Review" section below.

**If user enables agent teams**, acknowledge and proceed.

## Pre-Execution Review

**⚠️ CRITICAL**: Do NOT spawn any agents without explicit user approval.

### Presentation Format

Pre-Execution Review

Before spawning the agent team, here's what will happen:

Task to Accomplish

{Original task description}

Team Composition

{Number} teammates will be spawned:

1. {Agent Name} (Role: {role})

   • Responsibilities: {what they'll do}
   • Tools: {available tools}
   • Model: {claude model}

2. {Agent Name} (Role: {role}) [Same format...]

...

Task Breakdown

The task will be divided as follows:

1. {Task 1} → Assigned to: {Agent Name}

   • {Description of work}
   • Dependencies: {none | Task X, Task Y}

2. {Task 2} → Assigned to: {Agent Name} [Same format...]

...

Execution Mode

• Mode: {Agent Teams | Subagent Delegation}
• Estimated teammates: {number}
• Estimated duration: {rough estimate}
• Token usage: {warning about higher costs with teams}

Files Likely to be Modified

• {File path 1}
• {File path 2} ...

⚠️ IMPORTANT: Agent teams use significantly more tokens than a single session. Each teammate runs in its own context window.

Safety Check

Before proceeding, ensure:

• ✅ All team members have been reviewed and approved
• ✅ You understand what each agent will do
• ✅ You're comfortable with the estimated token usage
• ✅ You've committed or backed up important changes

Execute?

Please respond with:

• "execute" or "yes" - Spawn the team and begin execution
• "modify" - Change team composition or task assignments
• "add [agent-name]: [role]" - Add an additional agent
• "remove [number]" - Remove an agent from the team
• "show [agent-name]" - See full agent definition
• "cancel" - Abort execution

Would you like to proceed with execution?

### Handling Modifications

If user wants to modify the team:
- Update team composition based on feedback
- Adjust task assignments
- Re-present for approval
- Continue until user approves or cancels

## Agent Team Execution

### Step 1: Create Agent Team (Agent Teams Mode)

If agent teams are enabled:

Creating Agent Team

I'll now create an agent team to execute your task. This will spawn {number} Claude Code instances working together.

Creating team with natural language...

Use Claude's natural language interface to create the team:

Create an agent team to {task description}.

Spawn the following teammates:

• {Agent 1}: {role description}
• {Agent 2}: {role description}
• {Agent 3}: {role description} ...

Have them work through these tasks:

1. {Task 1} - assigned to {Agent 1}
2. {Task 2} - assigned to {Agent 2}
3. {Task 3} - assigned to {Agent 3} ...

### Step 2: Monitor Execution

**While the team is working:**

1. **Monitor Progress**
   - Watch the task list
   - Check teammate status
   - Be ready to intervene if needed

2. **Handle Messages**
   - Teammates may send questions or updates
   - Respond promptly to keep work moving
   - Escalate to user if decisions are needed

3. **Synthesize Results**
   - As teammates complete tasks, review their work
   - Identify conflicts or gaps
   - Coordinate integration of results

4. **Communicate with User**
   - Provide periodic updates on progress
   - Alert to any issues or blockers
   - Ask for guidance when decisions are needed

### Step 3: Clean Up

When the team completes:

Execution Complete

Results Summary

{Summarize what was accomplished}

Work Completed

• {Task 1}: {Status} - {Brief result}
• {Task 2}: {Status} - {Brief result} ...

Files Modified

• {File 1}: {Changes made}
• {File 2}: {Changes made} ...

Issues Encountered

• {Any problems that came up}
• {How they were resolved}

Next Steps

• {Recommended follow-up actions}
• {Any remaining work}

Cleaning Up Team

I'll now shut down the agent team to free resources...

Ask the lead to clean up the team:

Clean up the team

## Subagent Delegation Fallback

If agent teams are not enabled, use subagent delegation:

### Step 1: Create Subagents

Spawn subagents for each role:

Spawning Subagents

Since agent teams are not enabled, I'll use subagent delegation. Spawning {number} subagents...

Subagent 1: {Agent Name}

Role: {role description}

Create the subagent with the task tool:

Task: Create a subagent for {role}

Spawn a subagent with the following system prompt:

{Full agent definition}

Initial task: {Specific task for this subagent}

### Step 2: Execute Sequentially or in Parallel

**Option A: Sequential**
Execute subagents one at a time in dependency order:

Executing subagents sequentially based on task dependencies...

1. {Agent 1} - Starting... {Wait for completion}

2. {Agent 2} - Starting... {Wait for completion}

**Option B: Parallel (No Dependencies)**
Spawn multiple subagents simultaneously if tasks are independent:

Executing subagents in parallel (independent tasks)...

Spawning:

• {Agent 1}: {Task}
• {Agent 2}: {Task}
• {Agent 3}: {Task}

Waiting for all to complete...

### Step 3: Synthesize Results

Collect results from all subagents:

Synthesizing Results

Subagent Results:

1. {Agent 1}: {Result summary}
2. {Agent 2}: {Result summary} ...

Integration

{How results are being combined} {Any conflicts resolved} {Final outcome}

## Output Format

After execution completes:

Dream Team Execution Results

Task

{Original task}

Execution Mode

{Agent Teams | Subagent Delegation}

Team Performance

• Teammates spawned: {number}
• Tasks completed: {number}/{total}
• Time elapsed: {duration}
• Issues encountered: {count}

Deliverables

{What was produced}

Changes Made

{List of files modified}

Summary

{Brief summary of accomplishment}

Recommendations

• {Suggested next steps}
• {Things to review}
• {Potential improvements}

---

The dream-team workflow is complete! ✅

## Edge Cases

**User Cancels at Pre-Execution:**
- Respect cancellation
- Report: "Execution cancelled by user"
- Note what would have been done
- Offer to restart later

**Agent Team Creation Fails:**
- Report specific error
- May fallback to subagent mode automatically
- Or ask user if they want to try subagents

**Teammate Crashes/Stops:**
- Identify which teammate stopped
- Assess impact on task dependencies
- Options:
  - Restart the teammate
  - Reassign task to another teammate
  - Complete the task yourself
  - Ask user for guidance

**Conflicts Between Teammates:**
- If teammates produce conflicting results
- Review both approaches
- Synthesize best solution
- Or ask user to decide

**Task Takes Too Long:**
- If execution exceeds reasonable time
- Check teammate status
- May suggest:
  - Breaking task into smaller pieces
  - Adding more teammates
  - Changing approach
  - Manual intervention

**Token Limit Concerns:**
- If token usage is high
- Warn user
- Suggest ways to reduce:
  - Smaller tasks
  - Fewer teammates
  - More focused scope

## Guidelines

- Always get explicit approval before spawning agents
- Present clear team composition and task breakdown
- Monitor execution actively
- Communicate regularly with user
- Handle issues gracefully
- Synthesize results clearly
- Clean up teams when done
- Report accurately what was accomplished
- Provide actionable next steps