bradygaster/squad

What it solves

Squad addresses the challenge of managing AI-assisted development by providing a structured, human-led team of AI agents that persist within your repository. It solves the problem of fragmented AI interactions by ensuring that agents learn from your codebase, share decisions, and maintain context across sessions, all while keeping a human in the loop for oversight and approvals.

How it works

Squad acts as an orchestration layer for GitHub Copilot. You initialize a team in your project directory, which creates a .squad/ folder containing agent charters, history, and decision logs. When you assign a task, a coordinator routes the work to the appropriate specialists (e.g., frontend, backend, tester). These agents operate in parallel, record their decisions in shared files, and update their individual knowledge bases so they improve over time. The system also includes a "Watch" mode (Ralph) that can automatically poll for GitHub issues, triage them, and dispatch agents to handle tasks.

Who it’s for

It is designed for software developers and teams who want to accelerate their coding workflow using AI without losing control. It is particularly useful for those who want to delegate repetitive tasks, documentation, or boilerplate generation to a persistent, specialized AI team that understands their specific project conventions.

Highlights

• Persistent Team Memory: Agents maintain history and learn from your codebase, reducing the need to repeat instructions.
• Human-in-the-loop: Designed to amplify human operators rather than replace them, with clear gates for review and approval.
• Parallel Execution: The coordinator can launch multiple specialized agents simultaneously to handle different aspects of a task.
• Git-Native: All team state, decisions, and configurations are stored in your repository, making the team portable and version-controlled.
• Automated Triage: Includes a "Watch" mode to monitor issues and automatically assign them to agents for execution.