Lean Agile Team — Claude Code Plugin

Ship fast, learn faster. 7 agents, 12 skills, Agent Teams coordination.

Prerequisites

• Claude Code installed
• Required for full coordination: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 (set automatically by this plugin's settings.json)
• Works without Agent Teams in degraded sequential mode via Task() fallback

Quick Start

Install via Claude Code Marketplace

# Add the repository to the marketplace
claude plugin marketplace add bryanweaver/claude-agent-kit

# Install the team plugin
claude plugin install team@claude-agent-kit

Load Locally (Development)

# Load plugin directly from a local directory
# claude --plugin-dir /path/to/claude-agent-kit

Start Using the Team

# Ship a feature
/team-ship add user authentication

# Fix a production bug
/team-fix users can't log in

# Clean up tech debt
/team-cleanup authentication module

# Run and fix tests
/team-run-tests

# Add critical test coverage
/team-add-tests payment processing

Architecture

┌─────────────────────────────────────────────────┐
│                  Team Lead                       │
│            (Claude Code Main Agent)              │
└──────────────────┬──────────────────────────────┘
                   │
        ┌──────────┼──────────┬──────────┐
        │          │          │          │
        ▼          ▼          ▼          ▼
  ┌──────────┐ ┌────────┐ ┌──────────┐ ┌────────────┐
  │Full Stack│ │Database│ │ Reviewer │ │ Documentor │
  │Developer │ │ Admin  │ │          │ │            │
  └──────────┘ └────────┘ └──────────┘ └────────────┘
        │          │          │          │
        └──────────┼──────────┼──────────┘
                   │
                   ▼
            ┌──────────┐
            │ Shipper  │
            │ (Pipeline│
            │  Owner)  │
            └──────────┘
                   │
                   ▼
         Shared Task List
    (Agent Teams coordination)

The Lean Team

| Agent | Role | Model | Key Traits | |-------|------|-------|------------| | Full Stack Developer | Adaptive implementation across entire stack | Sonnet | Stack detection, worktree isolation, project memory | | Database Admin | Adaptive schema, queries, migrations, data integrity | Sonnet | Multi-DB support, worktree isolation, project memory | | Shipper | Git, testing, building, deployment, PRs | Sonnet | Unblocked pipeline access | | Reviewer | Security, bugs, performance review | Sonnet | Read-only (plan mode) | | Documentor | Create, maintain, organize codebase docs | Sonnet | Runs after tests pass | | Meta-Agent | Generate new custom agents | Opus | On-demand agent creation | | Meta-Skills-Agent | Generate new workflow skills | Opus | On-demand skill creation |

Adaptive Agents

The Full Stack Developer and Database Admin agents are stack-adaptive — they detect your project's tech stack on first invocation by scanning configuration files (package.json, requirements.txt, go.mod, etc.) and tailor their approach accordingly. They reference the built-in tech-stack skills for framework-specific patterns:

• nextjs-app-router — Next.js App Router patterns
• shadcn-components — shadcn/ui component patterns
• supabase-patterns — Supabase database patterns
• tanstack-query — TanStack Query data fetching
• testing-patterns — Testing patterns for React/Next.js

Skills

Workflow Skills

| Skill | Purpose | Speed | |-------|---------|-------| | /team-ship <feature> | Feature branch → implement → review → test → deploy → PR | Balanced | | /team-fix <bug> | Hotfix → diagnose → patch → deploy (no review step) | Fastest | | /team-cleanup <area> | Analyze tech debt → refactor → validate | Thorough | | /team-run-tests [scope] | Run all tests → batch-fix failures → verify | Thorough | | /team-add-tests <area> | Identify critical gaps → write minimal tests | Focused |

Utility Skills

| Skill | Purpose | |-------|---------| | /team-init-docs [area] | Generate comprehensive codebase documentation | | /team-update-docs [changes] | Update docs to reflect recent code changes | | /team-create-agent <description> | Create a new custom agent via meta-agent | | /team-create-skill <description> | Create a new workflow skill via meta-skills-agent | | /team-repo-status [focus] | Repository health report (git, todos, activity) | | /team-all-tools | List all available tools with signatures | | /team-audit [action] | Analyze audit logs (summary, report, metrics, timeline, verify, anomalies) |

Workflow Details

/team-ship — Build and Deploy Features

Shipper ──► Full Stack Dev + DB Admin (parallel) ──► Shipper ──► Reviewer ──► Shipper ──► Documentor ──► Shipper
Branch       Implement feature                       Commit       Review       Test         Update Docs    Deploy+PR

Tasks: Create branch → Implement → Commit → Review → Test → Fix regressions (if needed) → Update docs → Deploy & PR

/team-fix — Emergency Bug Fixes

Shipper ──► Full Stack Dev + DB Admin ──► Shipper ──► Shipper ──► Documentor ──► Shipper
Hotfix       Diagnose & patch              Commit      Test+Deploy  Update Docs    Merge

No reviewer step — speed is the priority for emergencies.

/team-cleanup — Technical Debt

Shipper ──► Reviewer ──► Full Stack Dev + DB Admin (parallel) ──► Shipper ──► Shipper ──► Documentor ──► Shipper
Branch       Analyze       Refactor                                Commit      Test         Update Docs     PR

Reviewer-first: analyze before refactoring.

/team-run-tests — Batch Test and Fix

Shipper ──► Shipper ──► Full Stack Dev + DB Admin ──► Shipper ──► Shipper ──► Reviewer ──► Documentor ──► Shipper
Branch       Run tests    Fix failures (parallel)      Commit      Re-test     Review       Update Docs     PR
                              ↑                                       │
                              └───────── loop if still failing ───────┘

/team-add-tests — Critical Test Coverage

Shipper ──► Reviewer ──► Full Stack Dev + DB Admin (parallel) ──► Shipper ──► Documentor ──► Shipper
Branch       Find gaps     Write minimal tests                     Run tests    Update Docs    Commit+PR

Test the 20% that prevents 80% of disasters.

Plugin Structure

claude-agent-kit/
├── .claude-plugin/
│   ├── plugin.json              # Plugin manifest
│   └── marketplace.json         # Marketplace distribution manifest
├── agents/
│   ├── full-stack-developer.md  # Adaptive full-stack implementation
│   ├── database-admin.md        # Adaptive database and data layer
│   ├── shipper.md               # Pipeline: git, test, build, deploy
│   ├── reviewer.md              # Security, bugs, performance review
│   ├── documentor.md            # Documentation creation and maintenance
│   ├── meta-agent.md            # Agent generator
│   └── meta-skills-agent.md     # Skill generator
├── skills/
│   ├── team-ship/SKILL.md        # Feature development workflow
│   ├── team-fix/SKILL.md         # Emergency hotfix workflow
│   ├── team-cleanup/SKILL.md     # Tech debt refactoring workflow
│   ├── team-run-tests/SKILL.md   # Batch test and fix workflow
│   ├── team-add-tests/SKILL.md   # Critical test coverage workflow
│   ├── team-init-docs/SKILL.md   # Full documentation generation
│   ├── team-update-docs/SKILL.md # Documentation maintenance
│   ├── team-create-agent/SKILL.md # Custom agent creation
│   ├── team-create-skill/SKILL.md # Workflow skill generation
│   ├── team-repo-status/SKILL.md # Repository health report
│   ├── team-audit/SKILL.md       # Audit log analysis
│   └── team-all-tools/SKILL.md   # List available tools
├── hooks/
│   └── hooks.json               # Quality gates and observability
├── settings.json                # Plugin settings (Agent Teams flag)
├── docs/
│   └── lean-agile-system.md     # Detailed system specification
└── .claude/
    └── settings.json            # Claude Code project settings

Configuration

Settings

The plugin automatically enables Agent Teams via settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Hooks

Three hooks provide quality gates and observability:

| Hook | Type | Purpose | |------|------|---------| | TaskCompleted | Prompt | LLM verifies task is genuinely complete before marking done | | TeammateIdle | Command | Logs idle events to <tmpdir>/lean-agile-team.log | | SubagentStop | Prompt | Validates git state is clean when shipper finishes (scoped via agent_name matcher) |

Dual-Mode Operation

Every skill supports two execution modes:

Agent Teams Mode (recommended)

With CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1:

• Spawns real teammates that work in parallel
• Uses shared task lists with dependency tracking
• Teammates communicate via mailbox protocol
• Full parallelism where workflows allow

Sequential Fallback

Without Agent Teams:

• Executes agents sequentially via Task() calls
• Same workflow steps, same quality gates
• Works on any Claude Code version
• Slightly slower due to serial execution

Creating Custom Agents

Use the meta-agent to generate new agents:

/team-create-agent Create a new agent that handles API integration testing

The meta-agent will:

1. Analyze your requirements
2. Select appropriate tools and model
3. Generate a complete agent definition
4. Save to agents/<name>.md

Key Principles

• Ship fast, learn faster — working software over perfect software
• Minimal testing — test the 20% that prevents 80% of disasters
• Speed focus — /team-fix is fastest, /team-ship is balanced, /team-cleanup and /team-run-tests are thorough
• Non-blocking reviews — reviewer suggestions don't stop deployment (except security)
• Clean pipeline — shipper owns the entire git-to-production flow
• Quality without bureaucracy — pragmatic reviews, not nitpicks