Agent Tree

A simplified hierarchical problem-solving system using Claude CLI. Breaks down complex tasks into manageable subtasks with human review between planning and execution.

Features

• Two-Phase Approach: Separate decomposition and solving phases
• Human-in-the-Loop: Review and modify task plans before execution
• Claude CLI Integration: Uses claude for all LLM interactions
• Dependency Resolution: Handles task dependencies intelligently
• Context-Aware Solving: Each task knows its place in the hierarchy
• Clean Architecture: Simple, modular design

Installation

Requires Claude CLI to be installed and configured.

Usage

Phase 1: Decompose a Complex Task

python agent_tree.py decompose your_task.md

This creates:

• your_task_plan.md - Analysis and decomposition plan
• your_task_children/ - Folder containing subtask files

Phase 2: Human Review

Review the generated files and modify as needed:

• Adjust task descriptions
• Add/remove dependencies
• Change complexity markers
• Refine the plan

Phase 3: Solve the Task Tree

python agent_tree.py solve your_task.md

This:

• Processes tasks in dependency order (leaves first)
• Updates plan files with progress and solutions
• Handles both simple and complex tasks

How It Works

Decomposition

1. Claude analyzes the task and creates a plan
2. Generates subtask files marked as "simple" or "complex"
3. Recursively decomposes complex subtasks (max 5 Claude calls)

Solving

1. Builds a tree view of all tasks with one-line summaries
2. Processes dependencies and children before parents
3. Each task sees where it fits in the overall system
4. Updates plan files with implementation details

Example

# 1. Create your task file
echo "# Build a URL shortener service" > url_shortener.md

# 2. Decompose it
python agent_tree.py decompose url_shortener.md

# 3. Review and edit the generated files
# ... make any changes you want ...

# 4. Solve it
python agent_tree.py solve url_shortener.md

File Structure

url_shortener.md                    # Original task
url_shortener_plan.md               # Decomposition and progress
url_shortener_children/
├── create_api.md                   # Subtask (might be complex)
├── create_api_plan.md              # Its plan and progress
├── implement_storage.md            # Subtask (simple)
└── implement_storage_plan.md       # Its progress

Task File Format

# Task Title

## Type
[simple or complex]

## Summary
[One line description]

## Task
[Detailed description]

### Dependents
- [Other Task](path/to/task.md)

Development

The system is designed for simplicity:

• agent_tree.py - Entry point with subcommands
• decompose.py - Handles task decomposition
• solve.py - Handles task execution

During development, errors are allowed to crash for easier debugging.