Agent Threads - Thread-Based Engineering Plugin
A comprehensive Claude Code plugin that teaches and demonstrates all 7 thread types
from the Thread-Based Engineering framework. Learn to work more effectively with
AI coding assistants through practical, hands-on examples.
What is Thread-Based Engineering?
Thread-Based Engineering is a mental framework for understanding how to work with
AI coding assistants like Claude. Created by IndyDevDan, it defines 7 patterns
("threads") for human-AI collaboration:
┌─────────┐ ┌─────────────────┐ ┌─────────┐
│ YOU │ ──▶ │ AI AGENT │ ──▶ │ YOU │
│ Prompt │ │ (tool calls) │ │ Review │
└─────────┘ └─────────────────┘ └─────────┘
START MIDDLE END
The Core Idea: You show up at the beginning (prompt) and end (review).
The AI does the work in the middle. Getting better means running MORE threads,
LONGER threads, and THICKER threads (agents using agents).
Quick Navigation
For New Users (Start Here!)
| Document | Description | Time |
|---|
| Quick Start Guide | Step-by-step setup with 10 hands-on exercises | 15 min |
| User Guide | Complete guide to all features | 30-60 min |
| FAQ | Common questions answered | As needed |
For Developers
Reference
Installation
Prerequisites
Install the Plugin
# Navigate to the plugin directory
cd /path/to/agent-threads
# Install
claude /plugin install .
Verify Installation
# Start Claude Code
claude
# Test a command
/agent-threads:base
If you see the Base Thread explanation, you're ready to go!
The 7 Thread Types
| Thread | Name | What It Does | Difficulty |
|---|
| Base | Base Thread | Single prompt → agent work → review | Beginner |
| P | Parallel | Multiple agents working simultaneously | Beginner |
| C | Chained | Multi-phase work with human checkpoints | Intermediate |
| F | Fusion | Best-of-N - multiple agents, aggregate results | Intermediate |
| B | Big/Meta | Agents orchestrating other agents | Advanced |
| L | Long | Extended autonomous work with validation loops | Advanced |
| Z | Zero-Touch | Fully automated with comprehensive validation | Expert |
Try Each Thread Type
/agent-threads:base # The fundamental pattern
/agent-threads:parallel # Scale with multiple agents
/agent-threads:chained # Safe multi-phase workflows
/agent-threads:fusion # Get the best of multiple attempts
/agent-threads:big # Orchestrate agent teams
/agent-threads:long # Extended autonomous work
/agent-threads:zero-touch # Full automation (the goal)
What's Included
Slash Commands (7)
One command per thread type, with explanations and live demos.
Custom Agents (4)
Specialized AI helpers with different capabilities:
| Agent | Role | Access |
|---|
@researcher | Fast codebase exploration | Read-only |
@reviewer | Code quality analysis | Read-only |
@builder | Implementation | Read/Write |
@validator | Test verification | Bash access |
Helper Scripts (5)
Standalone bash scripts for advanced patterns:
| Script | Purpose |
|---|
parallel-runner.sh | Spawn N parallel Claude instances |
fusion-aggregator.sh | Aggregate results from multiple agents |
chained-workflow.sh | Multi-phase workflow with checkpoints |
ralph-loop.sh | Stop hook for validation loops |
zero-touch-validator.sh | Comprehensive automated validation |
Hooks
Stop hook implementing the "Ralph Wiggum" pattern - code that controls
when AI continues or stops working.
Quick Examples
Basic Usage
# Start Claude Code
claude
# Ask about thread types
/agent-threads:base
# Analyze code
/agent-threads:base src/
Run Parallel Agents
./scripts/parallel-runner.sh "Review this code" 3
Run Validation Suite
./scripts/zero-touch-validator.sh
How to Improve
Thread-Based Engineering defines 4 metrics for improvement:
