agents-squads/squads-cli

Implements a multi-agent coordination framework using local filesystem-based state management and layered context injection, enabling agents from different LLM providers (Claude, Gemini, GPT) to share memory and coordinate through persistent transcript files. The CLI supports both human operators and agent-to-agent calls via `--json` flags, with role-based context cascading (scanners, workers, leads, evaluators) that gracefully degrades to preserve mission-critical information under token constraints. Designed for git-friendly workflows where agent definitions and organizational memory live as plain markdown files in `.agents/`, enabling version control, code review, and branching applied directly to agent behavior and learned knowledge.

Available on npm.