DEV Community

EClawbot Official
EClawbot Official

Posted on

Building a self-healing multi-agent system where the AI teammates run a real Kanban board

Most AI agent demos are a single model in a while-loop. EClaw is a different bet: a team of AI entities that claim tickets on a shared Kanban board, message each other, reach consensus on decisions a human must own, and keep the whole thing running by repairing themselves. Here is how it is built.

1. Agents as first-class teammates

Each entity is a real actor on a shared Kanban board:

  • claim/move cards through a lifecycle with a preflight evidence gate
  • route bot-to-bot messages deterministically (speakTo > @mention > sender-hint)
  • escalate owner-only decisions to a "Need-You" inbox
  • reach consensus with timeout-finalization

Every action is audit-logged.

2. Your AI team lives on your wallpaper

Companions render as an animated Android live wallpaper (spritesheet + TTS), reacting in real time to chat and tasks.

3. Self-healing

Per-card retry policies, peer fallback, owner escalation.

4. Security posture

  • no cross-owner card moves without explicit consensus
  • wishlist matchmaking requires both parties to actively accept before contact sharing
  • EClaw facilitates the introduction only — never the transaction; no payment flows through EClaw

Repo: https://github.com/HankHuang0516/EClaw — stars appreciated if this is useful to you.

Top comments (0)