Squad provides a framework for creating and managing AI agent teams that collaborate on software development projects within a GitHub repository. It aims to offer a persistent, specialized AI workforce that learns and improves over time, reducing the need for repetitive human context-setting. This is beneficial for developers seeking to augment their workflow with AI specialists.

How It Works

Squad deploys specialized AI agents (e.g., Frontend, Backend, Tester) that operate in parallel, each with its own context window. A coordinator manages task distribution and agent communication. Agents learn from their individual history.md files and a shared decisions.md file, storing accumulated knowledge directly in the .squad/ directory within Git. This approach allows the entire team's knowledge and state to be cloned with the repository, and optimizations focus on maximizing LLM context window utilization for agent tasks.

Quick Start & Requirements

• Install: npx github:bradygaster/squad
• Prerequisites: Node.js 22.0.0+, GitHub CLI (gh auth login), GitHub Copilot (CLI or VS Code), and a running SSH agent for installation (ssh-add).
• Setup: Requires authentication with GitHub and Copilot.

Highlighted Details

• Parallel Agent Execution: Agents work concurrently, with the coordinator managing dependencies and chaining follow-up tasks.
• Persistent, Git-Stored Knowledge: Agent learnings and team decisions are saved in .squad/ and versioned, allowing any clone to inherit the team's accumulated expertise.
• Context Window Optimization: Significant engineering effort is dedicated to managing LLM context budgets, ensuring agents have substantial room for task execution (78-83% of 200K tokens).
• GitHub Integration: Leverages GitHub Issues for automated triage and assignment via squad: labels and uses GitHub Actions for workflow automation.

Maintenance & Community

Conceived by @bradygaster. While a "Squad Community" is mentioned, specific links to Discord, Slack, or a roadmap are not provided in the README. Planned features include automated pruning of decisions.md for context management.

Licensing & Compatibility

The license type is not specified in the provided README. Compatibility requires Node.js 22+, GitHub CLI, and Copilot. The project is marked as experimental.

Limitations & Caveats

The project is experimental, with potential for API and file format changes. Installation requires a running SSH agent to avoid apparent hangs. Known platform-level issues originate from the Copilot CLI runtime, including cosmetic --no-warnings errors and potential server error retry loops, though workarounds are provided. Agent capability increases with accumulated usage history.