shard-orchestrator 1.0.0rc1

A TDD-Driven, Parallelized AI Coding Orchestrator

Shard

A TDD-driven, parallelized AI coding orchestrator that breaks down complex tasks and executes them concurrently using git worktrees.

Installation

Homebrew (macOS)

brew tap nihalgunu/shard
brew install shard

pip

pip install shard-ai

From source

git clone https://github.com/nihalgunu/Shard.git
cd Shard
pip install -e ".[dev]"

Quick Start

# Run a full pipeline
shard run -p "Add user authentication with JWT tokens"

# Preview the execution plan without running
shard plan -p "Refactor database layer to async"

# Check status
shard status

# Resume an interrupted run
shard resume <run-id>

How It Works

Shard takes a natural language prompt and:

1. Plans - Generates a DAG of sub-tasks with file ownership boundaries
2. Partitions - Provisions isolated git worktrees for parallel execution
3. Dispatches - Runs AI agents concurrently on each task
4. Aggregates - Merges all branches and resolves conflicts
5. Self-heals - Runs tests and automatically fixes failures

Configuration

Create shard.toml in your repository:

[agent]
backend = "claude-code"  # or "aider", "cursor-cli"
max_concurrent = 4

[cost]
max_usd = 10.0

[retries]
max_per_task = 3
max_global = 5

Commands

Command	Description
shard run	Execute a full pipeline
shard plan	Preview execution DAG
shard resume	Resume interrupted run
shard status	Show current status
shard logs	View agent output
shard abort	Stop all agents
shard clean	Remove artifacts
shard config	Show configuration

Requirements

• Python 3.11+
• Git 2.20+
• AI coding agent (Claude Code, Aider, or Cursor CLI)

License

Apache 2.0