LocalAgentCrew

A multi-agent orchestration framework for Claude Code that enables specialized AI agents to collaborate on complex tasks.

Overview

LocalAgentCrew provides 15 specialized agents across three domains:

• Technical Development: Research, Implementation, Debug, Testing, Security, Performance, Documentation
• Productivity: Writing Assistant, Task Management, Research Assistant
• Business Analysis: Market Researcher, Competitor Analyst, Customer Insights, Trend Forecaster, SWOT Analyst

Architecture

User Query → Hook Interception → Orchestrator → Agent Selection → Claude Execution
                                      │
                    ┌─────────────────┼─────────────────┐
                    ▼                 ▼                 ▼
              Single Agent      Multi-Agent        Workflow
                                (parallel)       (multi-stage)

The system uses a hook-based architecture that intercepts user prompts, analyzes them against agent triggers, and routes to the appropriate specialized agent(s).

Quick Start

1. Clone this repository
2. The .claudecode/settings.json hooks are automatically detected by Claude Code
3. Start asking questions - the orchestrator handles routing

Example queries:

"How does the authentication system work?"  → Research Agent
"Implement a new login feature"             → Implementation Agent
"Debug the checkout error"                  → Debug Agent
"Run a complete market analysis"            → Market Analysis Workflow

Documentation

Document	Description
Architecture	Detailed system architecture and components
Gap Analysis	Comparison with latest Claude Code features
Migration Roadmap	Plan to modernize the system
Claude Code Evolution	Research on recent Claude Code changes
Agent System	Comprehensive agent documentation

Project Status

Current Implementation

• ✅ 15 specialized agents with keyword triggers
• ✅ 6 predefined workflows (feature-complete, bug-fix, code-review, market analysis)
• ✅ Parallel and sequential execution modes
• ✅ Context-aware routing (git, files)
• ✅ ANSI color-coded output
• ✅ Usage tracking with [USAGE] prefix

Migration Progress (Phases 1-3 Complete)

• ✅ Skills System: All 15 agents converted to Skills format
• ✅ Native Subagents: 11 subagent definitions with model tiering
• ✅ Model Tiering: Haiku for fast tasks, Sonnet for complex reasoning
• 🔄 REST API for results (pending)
• 🔄 Web dashboard for visualization (pending)
• 🔄 Database persistence (pending)

Model Tiering

Model	Agents	Use Case
Haiku	research, testing, documentation, writing-assistant, task-planner, research-assistant	Fast, cost-effective
Sonnet	implementation, debug, security, performance, market-analyst	Complex reasoning

Directory Structure

LocalAgentCrew/
├── .claude/                   # Modern Claude Code configuration
│   ├── settings.json          # Settings with Skills enabled
│   ├── agents/                # Native subagent definitions (11)
│   │   ├── research.md        # haiku
│   │   ├── implementation.md  # sonnet
│   │   ├── debug.md           # sonnet
│   │   └── ...
│   └── skills/                # Auto-discovered Skills (15)
│       ├── research/SKILL.md
│       ├── implementation/SKILL.md
│       └── ...
├── .claudecode/               # Legacy system (still functional)
│   └── agents/
│       ├── config.json        # Agent & workflow definitions
│       ├── prompts/           # Original prompt templates
│       ├── scripts/           # Orchestration engine (JS)
│       └── workflows/         # Workflow definitions
├── docs/                      # Documentation
│   ├── ARCHITECTURE.md
│   ├── GAP_ANALYSIS.md
│   ├── MIGRATION_ROADMAP.md
│   └── CLAUDE_CODE_EVOLUTION.md
├── CLAUDE.md                  # Project conventions for Claude
├── AGENT_SYSTEM.md            # Detailed agent documentation
└── README.md

Agent Categories

Technical Agents

Agent	Triggers	Color
Research	how does, what is, explain, analyze	Cyan
Implementation	implement, create, add feature	Green
Debug	debug, fix bug, error	Red
Testing	test, unit test, coverage	Yellow
Security	security, vulnerability, auth	Bright Red
Performance	optimize, performance, slow	Magenta
Documentation	document, write docs, readme	Blue

Productivity Agents

Agent	Triggers	Color
Writing Assistant	write email, draft, compose	Purple
Task Management	organize tasks, create plan	Orange
Research Assistant	research, summarize, compare	Teal

Business Agents

Agent	Triggers	Color
Market Researcher	market research, market size	Bright Cyan
Competitor Analyst	competitor analysis, benchmark	Bright Yellow
Customer Insights	customer analysis, buyer persona	Bright Magenta
Trend Forecaster	forecast, predict, trends	Bright Green
SWOT Analyst	swot analysis, strategic analysis	Bright Blue

Workflows

Workflow	Stages	Use Case
Feature Complete	research → implementation → QA → finalization	Full feature development
Bug Fix	debug → testing	Debug and validate
Code Review	security + performance + testing → documentation	Comprehensive review
Market Analysis	research → intelligence → strategic	Complete market assessment

Configuration

Hook Configuration (.claudecode/settings.json)

{
  "hooks": {
    "userPromptSubmit": "node .claudecode/agents/scripts/hook-handler.js"
  },
  "agents": {
    "enabled": true,
    "orchestrationMode": "auto"
  }
}

Key Settings (.claudecode/agents/config.json)

• orchestration.max_concurrent_agents: Max parallel agents (default: 3)
• routing.fallback_agent: Default when no match (research)
• routing.multi_agent_threshold: Min score for multi-agent

Usage Tracking

Prefix your query with [USAGE] to enable session tracking:

[USAGE] Analyze the payment system architecture

Creates timestamped folders:

• questions/ - Stored prompts
• interactions/ - Conversation logs
• results/ - Output artifacts

Future Vision

This project aims to become a comprehensive platform for:

1. Multi-agent collaboration - Specialized agents working together
2. Result visualization - Web dashboard for viewing outputs
3. Data persistence - API and database integration
4. Enterprise integration - Sync results to external systems

See Migration Roadmap for detailed plans.

Recent Changes

The Claude Code ecosystem has evolved significantly since this project was started. Key updates include:

• Skills (Dec 2025) - New extensibility paradigm replacing slash commands
• Native Subagents - Built-in multi-agent support
• Model Tiering - Use Opus/Sonnet/Haiku for different tasks
• Claude Haiku 4.5 (Oct 2025) - 90% capability at 3x cost savings

See Gap Analysis for detailed comparison.

Contributing

This is a personal project for local agent crews. Feel free to fork and adapt for your own use.

License

See LICENSE file.

---

Last updated: January 2026