Are you an LLM? Read llms.txt for a summary of the docs, or llms-full.txt for the full context.
Skip to content
agentful
agentful

AI development toolkit for Claude Code

Parallel agent execution • Three-tier architecture • Auto-generated specialists

agentful
How It Works
1
📝
You describe what to build
/agentful "build auth and roles"
Run /agentful
2
🪝
Hooks enforce guardrails
SessionStart • PreToolUse • PostToolUse
3
🎯
Orchestrator analyzes & delegates
Reads state/history/context, creates execution plan, spawns agents
⚡ Parallel execution in git worktrees
4
Agents execute in parallel
🏗️ Architect
⚙️ Backend
🎨 Frontend
🧪 Tester
0min2min4min6min
💡Each agent uses shared skills
✓Validation
🔍Research
🧪Testing
📋Product Planning
📊Product Tracking
💬Conversation
5
🔍
Reviewer runs quality gates
📘Types
✨Lint
🧪Tests
📊Coverage
🔒Security
🗑️Dead Code
✗Validation fails
🔧 Fixer auto-fixes issues and re-validates
✓All gates pass
Clean, tested, secure, and fully validated
6
✅
Code ready for review
All quality gates passed • Ready to merge

Why Parallel Execution Matters

⚡With Git Worktrees
6 min
Backend
▶️running
Frontend
▶️running
Tester
▶️running
Reviewer
▶️running
All agents run simultaneously in isolated git worktrees
🐌Without Worktrees
15 min
Backend
▶️running
Frontend
⏸️waiting
Tester
⏸️waiting
Reviewer
⏸️waiting
Each agent must wait for the previous one to complete
2.5x faster
with parallel execution

What is agentful?

AI development toolkit that coordinates specialized agents in Claude Code to build software from product specs.

Key capabilities:
  • Parallel execution - Multiple agents work simultaneously in isolated git worktrees (frontend + backend + tests running concurrently)
  • Three-tier architecture - Core (pre-built), domain (generated for your stack), ephemeral (temporary, task-specific)
  • Shared skills - Reusable capabilities like validation, testing, and research across all agents
  • Custom hooks - Intercept and modify agent behavior at key lifecycle points
  • Quality gates - Automated validation for every change (tests, types, coverage, security)
  • Pattern learning - Agents compound knowledge across sessions via MCP server
  • Human checkpoints - You decide, agents execute

Our Philosophy

Every software project needs three tiers of agents and skills:

  1. Core (included) - Universal agents (orchestrator, reviewer, fixer) + shared skills (validation, testing, research). Same for every project.
  2. Domain (generated) - Stack-specific agents (backend, frontend for web; extractor, transformer for pipelines) + tech skills (Next.js, Postgres, Tailwind). Architect creates these based on your codebase.
  3. Ephemeral (temporary) - Task-specific agents/skills created for one-off tasks or longer-term work, then removed when done (migration-agent for a 2-hour DB migration, performance-profiling for a multi-week optimization project).

Skills are shared across agents. Any agent can use validation, research, etc. Our OOTB setup works for all software projects, and you can tweak installation via CLI flags in the Installation Guide.

Install with CLI

Use preset flags with the CLI to customize your installation:

# Full install (recommended)
npx @itz4blitz/agentful init
 
# Minimal preset
npx @itz4blitz/agentful init --preset=minimal
 
# Custom selection
npx @itz4blitz/agentful init --agents=orchestrator,backend,tester --skills=validation,testing

Get Started

Ready to install? See the Installation Guide for all options, or get running in 5 minutes with the Quick Start Guide.

Learn More

Quick Start

Get agentful running in 5 minutes with a simple tutorial.

Quick Start Guide →

How It Works

Understand the architecture, agents, and workflow.

Overview →

Your First Project

Step-by-step tutorial building a complete app.

First Project →

Installation Options

Customize presets, agents, skills, and quality gates.

Installation Guide →

Resources