Agent Teams

SKILL: Agent Teams

Purpose: Native Claude Code Agent Teams coordination with team templates and quality gates Target: Team Leads coordinating execution-team, review-team, planning-team

Quick Start

When to Use

• Execute SCs with specialized agents (coder, frontend-engineer, backend-engineer)
• Multi-angle review with cross-discussion
• Parallel exploration and research

Quick Reference

# Spawn teammate for SC implementation
Spawn teammate "sc1-impl" with prompt:
  "You are a coder (see @.claude/agents/coder.md).
  Execute SC-1 from $PLAN_PATH. Skills: tdd, ralph-loop, vibe-coding.
  When complete, mark task as done. If blocked, message the team lead."

# Delegate mode (Shift+Tab): coordinate only, never implement
# TaskCompleted hook auto-verifies tests/coverage/todos

What This Skill Covers

In Scope

• Agent Teams coordination patterns
• Team templates (execution, review, planning)
• Teammate spawning with role-specific prompts
• TaskCompleted and TeammateIdle hook integration

Out of Scope

• Hook implementation details → @.claude/hooks/verify-task-completion.sh
• Individual agent definitions → @.claude/agents/
• Test frameworks and coverage → @.claude/skills/tdd/SKILL.md

Core Concepts

Architecture

• Team Lead: Coordinates in delegate mode (Shift+Tab), never implements directly
• Teammates: Independent agents, full context, direct communication, shared Task List
• Quality Gates: TaskCompleted hook (auto-verify), TeammateIdle hook (prevent premature idle)
• Communication: Multi-directional (teammate ↔ teammate ↔ lead)

Delegate Mode (MANDATORY)

Team Lead MUST coordinate only (spawn teammates, monitor, aggregate) and NEVER implement (edit files, write tests).

Pattern: Spawn → Monitor → Handle messages → Aggregate

Team Templates

1. execution-team: SC implementation (coder, frontend-engineer, backend-engineer)
2. review-team: Multi-angle review (tester, validator, code-reviewer, security-analyst)
3. planning-team: Exploration/research (explorer, researcher)

Team Templates

execution-team

Purpose: Implement SCs with specialized agents (coder, frontend-engineer, backend-engineer)

Spawn Pattern:

Spawn teammate "sc1-impl" with prompt:
  "You are a [role] (see @.claude/agents/[role].md).
  Execute SC-1: [description] from $PLAN_PATH. Skills: tdd, ralph-loop, vibe-coding.
  When complete, mark done. If blocked, message team lead."

Coordination: Parallel (independent SCs) / Sequential (dependent SCs) / Blocked → GPT delegation

Quality Gates: Tests pass, coverage ≥80%, todos checked (TaskCompleted hook)

review-team

Purpose: Multi-angle review with cross-discussion (tester, validator, code-reviewer, security-analyst, design-reviewer)

Spawn Pattern:

Spawn teammate "test-reviewer" with prompt:
  "You are a tester (see @.claude/agents/tester.md).
  Review test coverage for $PLAN_PATH.
  Evaluate: completeness, coverage ≥80%, verification commands.
  Discuss findings with other reviewers via Message.
  Output: PASS or FAIL with details."

Key Enhancement: Reviewers cross-reference and debate findings via direct messaging

Quality Gates: All PASS required, any FAIL blocks completion

planning-team

Purpose: Parallel exploration/research for plan creation (explorer, researcher)

Spawn Pattern:

Spawn teammate "codebase-explorer" with prompt:
  "You are an explorer (see @.claude/agents/explorer.md).
  Explore codebase for [domain] related to [task].
  Find: files, patterns, config, tests. Share findings via Message.
  Output: File list with descriptions."

Spawn teammate "docs-researcher" with prompt:
  "You are a researcher (see @.claude/agents/researcher.md).
  Research external docs for [task].
  Find: official docs, best practices, security. Share findings via Message.
  Output: Research summary with links."

Coordination: Teammates share findings, lead synthesizes for plan

Quality Gates

TaskCompleted Hook

Trigger: Teammate marks task done Verification: .claude/hooks/verify-task-completion.sh checks tests pass, coverage ≥80%, todos checked Result: Exit 0 (allow) / Exit 2 (reject with feedback)

TeammateIdle Hook

Trigger: Teammate attempts idle Check: .claude/hooks/teammate-idle-check.sh verifies SC complete and quality gates pass Purpose: Prevent premature idle

Environment Setup

Required: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in settings.json env Hooks: Register TaskCompleted and TeammateIdle in settings.json (see REFERENCE.md) Terminal: tmux/iTerm2 for split-pane (recommended), in-process works in all terminals

Further Reading

Internal: @.claude/skills/agent-teams/REFERENCE.md - Full teammate spawn templates, hook examples, before/after comparison | @.claude/agents/CONTEXT.md - Agent ecosystem | @.claude/skills/parallel-subagents/SKILL.md - Dependency analysis patterns | @.claude/skills/tdd/SKILL.md - Test-driven development | @.claude/skills/ralph-loop/SKILL.md - Autonomous iteration

External: Claude Code Agent Teams Documentation

Version: claude-pilot 5.0.0