agent-kanban
A shared Kanban board where AI coding agents and humans coordinate as a team — “mission control for an AI workforce.” Its premise: “AI coding agents can write code, but they can’t collaborate. There’s no shared workspace where agents and humans coordinate.” A concrete instance of the agentic-coding-harness pattern (the scaffolding around the model) and of agent-orchestration.
What it provides
- Real-time board (SSE) with Todo → In Progress → In Review → Done flow.
- Cryptographic agent identity — each agent gets an Ed25519 keypair + fingerprint that persists across tasks and commits (accountability for an autonomous workforce).
- Agent autonomy — agents create subtasks, assign work to teammates, and review each other’s PRs; a leader agent decomposes goals and delegates to specialists.
- Skill system — task-specific agent-skills loaded per repository.
- Multi-runtime — Claude Code, Codex, Gemini CLI, GitHub Copilot, and ACP-compliant agents; human↔agent direct messaging; task dependencies with cycle detection.
Why it belongs here
It is agent-orchestration made collaborative and supervised: not just orchestrator → fan-out, but a persistent shared workspace with identity and human-in-the-loop review — the augment→automate axis (synthesis) held deliberately in the middle (humans supervise a visible board). Shares the multi-runtime, skills-loaded form factor seen in agentsys and the agentskills-spec sources. (Caveat: README claims; not independently evaluated.)
Related
agentic-coding-harness · agent-orchestration · agent-skills · agentsys