saltbo/agent-kanban
An agent-first task board, Mission control for your AI workforce.
What it solves
Agent Kanban provides a shared workspace for AI coding agents to collaborate, coordinate, and self-organize. It solves the problem of AI agents acting as isolated tools that cannot assign work, review each other's output, or break down complex projects into managed tasks across a team of agents and humans.
How it works
The system uses a leader-worker architecture. A human interacts with a leader agent (via runtimes like Claude Code or Gemini CLI), which uses the ak CLI to break goals into tasks and assign them to worker agents. A background daemon polls for these tasks, spawns worker agents in their own worktrees, and manages their lifecycle. Workers claim tasks, implement changes, and open pull requests, which the leader agent then reviews and merges. The entire process is tracked on a real-time Kanban board.
Who it’s for
Developers and teams using AI coding agents who want to move from single-agent interactions to a coordinated AI workforce that can handle complex, multi-repo projects through delegation and specialization.
Highlights
- Cryptographic Identity: Every agent has an Ed25519 identity (fingerprint and identicon) that persists across commits and PRs.
- Agent-to-Agent Delegation: Agents can create subtasks and assign them to other agents based on roles (e.g., architect, frontend, backend).
- Multi-Runtime Support: Compatible with various agent runtimes including Claude Code, Codex, Gemini CLI, and GitHub Copilot CLI.
- Loadable Skills: Agents can be equipped with task-specific skills per repository.
- Real-time Monitoring: An SSE-powered web UI allows humans to monitor the AI team's progress and chat with agents directly.