Teams

UI Teams Mode v5.0 - Positive Edition

Mission

Execute UI generation with multiple specialist agents working in parallel using Claude Code's native Agent Teams feature for maximum speed and quality on large, complex projects.

Core Philosophy: Parallel excellence through coordinated collaboration - each agent contributes their expertise while maintaining clear communication channels.

---

⚙️ Active Baseline Configuration

Team Dials (adapted from ui parent):

Dial	Default	Range	Purpose
TEAM_SIZE	4	1-4	Number of parallel agents
PARALLEL_STRATEGY	8	1-10	1=Sequential, 10=Maximum parallelism
COMMUNICATION_FREQ	7	1-10	1=Minimal, 10=Chatty
QUALITY_OVERSIGHT	9	1-10	1=Fast, 10=Thorough review

AI Instruction:

• Adapt team size based on project complexity (simple→2, complex→4)
• High parallel strategy for time-critical projects
• Communication freq 7: Balanced updates (not spammy, not silent)
• Quality oversight 9: Full adversarial + reviewer for production

Justification:

• Team size 4: Maximum parallelism without coordination overhead
• Parallel 8: Aggressive parallelization for speed
• Communication 7: Keep team informed without overwhelming
• Quality 9: Always validate for production (non-negotiable)

---

What is Teams Mode?

Teams mode uses Claude Code's Agent Teams feature to spawn multiple specialist agents that work together on UI generation:

• Orchestrator Agent — Analyzes prompt, selects paradigms, chooses constraints
• Build Agent — Creates the UI with paradigm validation
• Adversarial Agent — Challenges decisions and proposes alternatives
• Reviewer Agent — Validates accessibility and quality

Each agent has its own context and can communicate with others via the team messaging system.

---

🎯 When to Use Teams Mode

Optimal Use Cases

USE Teams Mode when:

• ✅ Large/complex projects (3+ features, multi-domain)
• ✅ Time-critical deadlines (need 1.5-2x speedup)
• ✅ Enterprise scale (full systems, design platforms)
• ✅ Multi-domain work (frontend + backend + docs)

Sequential Mode Preferred When:

Sequential mode (without --team) works best when:

• ✅ Simple interfaces (single page, minimal UI)
• ✅ Quick prototypes (faster to do sequentially)
• ✅ Learning mode (step mode better for teaching)
• ✅ Resource constraints (teams use more tokens)

Specific Trigger Phrases

'parallel agents for UI' → ACTIVATE
'team build interface' → ACTIVATE
'multi-agent UI generation' → ACTIVATE
'split UI work' → ACTIVATE
'concurrent UI design' → ACTIVATE
'agent team for design system' → ACTIVATE
'full SaaS platform' → ACTIVATE (complex)

---

🔧 Activation

Enable Agent Teams

# Enable experiment flag
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS="1"

Use Teams Mode

# Auto-detect (large/complex prompts)
/ui "Full SaaS dashboard with user management, billing, and real-time analytics"

# Explicit teams mode
/ui --team "Multi-feature application"

# Teams with step mode
/ui --step --team "Enterprise dashboard"

# Teams with explanation
/ui --team --explain "Complex design system"

---

👥 Team Composition

Default Team (2 Agents)

┌─────────────────────────────────────────────────────┐
│  STRIKE TEAM                                         │
├─────────────────────────────────────────────────────┤
│                                                       │
│  ┌─────────────────┐      ┌─────────────────────┐  │
│  │ ORCHESTRATOR     │      │ BUILD               │  │
│  │                 │      │                     │  │
│  │ • Analyze       │      │ • Build UI          │  │
│  │ • Detect        │      │ • Validate a11y     │  │
│  │ • Select        │      │ • Paradigm compliance│  │
│  │ • Enrich        │      │ • Metrics           │  │
│  │                 │      │                     │  │
│  └─────────────────┘      └─────────────────────┘  │
│                                                       │
└─────────────────────────────────────────────────────┘

Full Team (4 Agents)

┌─────────────────────────────────────────────────────────────────────────────┐
│  STRIKE FULL TEAM                                                            │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  ┌─────────┐  │
│  │ORCHESTRATOR  │  │BUILD         │  │ADVERSARIAL   │  │REVIEWER │  │
│  │              │  │              │  │              │  │         │  │
│  │• Analysis    │  │• Build       │  │• Challenge   │  │• A11y   │  │
│  │• Paradigms   │  │• Components  │  │• Alternatives│  │• Quality │  │
│  │• Constraints │  │• Validation  │  │• Stress test │  │• Docs   │  │
│  │              │  │              │  │              │  │         │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  └─────────┘  │
│                                                                      │
└─────────────────────────────────────────────────────────────────────────────┘

---

🔄 Workflow

Phase 1: Team Creation

## Team Initialization

Creating ui team for: "<user_prompt>"

Team: ui-<session-id>
Members:
1. orchestrator — Prompt analysis and constraint selection
2. build — UI building and validation
3. adversarial — Decision challenge (optional)
4. reviewer — Quality and accessibility review (optional)

Shared task list: ~/.claude/tasks/ui-<session-id>/

Phase 2: Parallel Execution

Orchestrator Agent:

1. Analyzes prompt for trend keywords
2. Applies PIVOT MATRIX for paradigm mapping
3. Selects creative constraints with scoring
4. Generates enriched specification
5. Writes to .claude/.strike/enriched-spec.json

Build Agent (parallel):

1. Reads enriched specification
2. Loads component registry
3. Plans implementation with constraints
4. Builds UI (react-tailwind or vanilla)
5. Validates against paradigms
6. Creates build-result.json

Adversarial Agent (parallel):

1. Reviews constraint selection
2. Proposes alternatives
3. Stress tests decisions
4. Reports findings to orchestrator

Reviewer Agent (parallel):

1. Runs accessibility checklist
2. Validates bundle size
3. Checks code quality
4. Documents findings

Phase 3: Coordination

## Team Coordination

Waiting for all agents to complete...

[========] Orchestrator: 100% (spec ready)
[========] Build: 100% (build complete)
[====] Adversarial: 50% (reviewing constraints)
[==] Reviewer: 25% (a11y check)

All agents complete! Synthesizing results...

Phase 4: Synthesis

## Team Results

Orchestrator delivered:
- analysis.md: Keywords and risk assessment
- constraints.md: 3 constraints selected
- enriched-spec.json: Validated specification

Build delivered:
- output/react-tailwind/: Complete UI
- build-result.json: Metrics and validation

Adversarial delivered:
- challenge-review.md: Alternative options considered
- Final decision: APPROVED with safeguards

Reviewer delivered:
- accessibility-report.md: WCAG AA compliant
- quality-report.md: Code quality: 95%

**Output ready at ./output/**

---

📋 Task List Structure

The team shares a task list at ~/.claude/tasks/ui-<session-id>/:

~/.claude/tasks/ui-abc123/
├── task-001-analyze.prompt      # Orchestrator: Analyze prompt
├── task-002-apply-pivot         # Orchestrator: Apply PIVOT MATRIX
├── task-003-select-paradigms   # Orchestrator: Select paradigms
├── task-004-select.constraints  # Orchestrator: Select constraints
├── task-005-enrich.spec         # Orchestrator: Create spec
├── task-006-read.registry       # Build: Load components
├── task-007-plan.ui             # Build: Plan structure
├── task-008-build.ui            # Build: Build
├── task-009-validate.a11y       # Build: Check accessibility
├── task-010-challenge.decision  # Adversarial: Review
├── task-011.review.quality      # Reviewer: Quality check

Agents claim tasks as they work, blocking dependent tasks.

---

💬 Inter-Agent Communication

Agents communicate via the team messaging system:

# Orchestrator → Build
@build Spec ready at .claude/.strike/enriched-spec.json
Paradigms: Warm Brutalism, List Narrative
Constraints: paper_ink (score: 35), architectural (score: 78)

# Build → Orchestrator
@orchestrator Build complete. Paradigm compliance: 100%
Used Room components instead of Grid (architectural constraint)
Output: ./output/react-tailwind/

# Adversarial → Orchestrator
@orchestrator Challenge: architectural constraint has edge cases
Alternative: warm_only + mechanical (score: 68)
Recommendation: Proceed with safeguards

# Reviewer → All
@team Accessibility: PASS (WCAG AA compliant)
Quality: 95/100 (minor: add more comments)
Bundle: 8KB gzipped (excellent)

---

🚀 Flag Combinations

Command	Team Size	Features	Speed
/ui --team "..."	2	Orchestrator + Build	1.6x faster
/ui --team --explain "..."	2	With diagrams	1.6x faster
/ui --team --full "..."	4	Everything	1.85x faster

---

📊 Performance

Sequential vs Teams

Sequential (baseline):
Orchestrator: 45s → Build: 60s → Review: 15s = 120s

Teams (parallel):
Orchestrator + Build: 60s → Review: 15s = 75s (1.6x faster)

Full Team Speedup

Sequential: 120s
Full Team (4 agents, parallel): 65s (1.85x faster)

---

✅ Teams Mode Quality Gates

Before claiming team build complete:

Team Coordination

• All agents completed successfully
• Inter-agent communication working
• No agent failures or timeouts
• Task list properly synchronized

Output Quality

• All deliverables created
• Orchestrator spec valid
• Build paradigm-compliant
• Adversarial challenges addressed
• Reviewer checks passed

State Preservation

• Task list archived
• Communication logs saved
• Results synthesized
• Session reproducible

---

🚨 Error Handling

Orchestrator Fails

If Orchestrator fails:
1. Build stops (waiting for spec)
2. Adversarial reports error
3. Team lead (you) decides: retry or fallback

Fallback: Run without constraints using /ui --build directly

Build Fails

If Build fails:
1. Orchestrator completes (spec reusable)
2. Reviewer logs the failure
3. Team lead decides: retry with different stack or debug

Debug: Check build log in .claude/.strike/build.log

Communication Failure

If an agent stops responding:
1. Wait 30 seconds for timeout
2. Check agent status in task list
3. Restart failed agent if needed
4. Resume from last completed task

---

🔗 Integration with Other Modes

Step + Teams

/ui --step --team "Complex dashboard"

In step + teams mode:

• Each phase executed by appropriate agent
• Team lead (you) reviews and approves
• Agents pause and wait for input
• Parallel work within phases, sequential between phases

Demo + Teams

/ui --demo --team "Large app"

In demo + teams mode:

• Simplified agent outputs (less verbose)
• Faster phase execution
• Still parallel, still faster
• Good for large but simple projects

---

🎯 Best Practices

1. Use for complex prompts — Multiple features, enterprise scale
2. Direct mode for simple prompts — Single page, minimal UI
3. Monitor task list — Check ~/.claude/tasks/ui-*/ for progress
4. Read agent messages — Inter-agent communication contains decisions
5. Review outputs — Each agent creates documentation
6. Clean up — Archive task list after completion

---

🔧 Configuration

Team Settings

{
  "ui": {
    "teams": {
      "enabled": true,
      "default_size": 2,
      "max_agents": 4,
      "display_mode": "in-process",
      "timeout_seconds": 300
    }
  }
}

Agent Roles

{
  "agents": {
    "orchestrator": {
      "tasks": ["analyze", "pivot", "paradigms", "constraints", "enrich"],
      "skills": ["ui"],
      "dependencies": []
    },
    "build": {
      "tasks": ["build", "validate", "metrics"],
      "skills": ["build"],
      "dependencies": ["orchestrator"]
    },
    "adversarial": {
      "tasks": ["challenge", "stress-test"],
      "dependencies": ["orchestrator"],
      "optional": true
    },
    "reviewer": {
      "tasks": ["a11y", "quality", "docs"],
      "dependencies": ["build"],
      "optional": true
    }
  }
}

---

🔗 Integration with Other Skills

Requires:

• ui (parent orchestrator) - Teams mode is a ui sub-skill

Complements:

• step - Step mode can be combined with teams mode
• verification-before-completion - Final verification before done

---

📋 Quick Reference

Team Activation

export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS="1"
/ui --team "complex project"

Team Composition

2 agents: Orchestrator + Build (default)
4 agents: + Adversarial + Reviewer (full)

Speedup

2 agents: 1.6x faster
4 agents: 1.85x faster

Task List

~/.claude/tasks/ui-<session-id>/

---

📊 Migration from Sequential

Before (sequential v3.0):

/ui "Create dashboard"
# Orchestrator runs (45s)
# Build runs (60s)
# Total: 105s

After (teams v5.0):

/ui --team "Create dashboard"
# Orchestrator + Build run together (60s)
# Total: 60s (1.75x faster)

---

UI Teams Mode v5.0 - Positive Edition: Parallel multi-agent orchestration for complex UI projects