saltbo/agent-kanban

What it solves

Agent Kanban provides a mission-control board for autonomous software agents, solving the problem of coordinating durable work across repositories. It removes the need for humans to "babysit" agents by providing a structured control plane for ownership, lifecycle transitions, dependency tracking, and independent review, moving beyond simple chat-based agent interactions.

How it works

It operates as a control plane that treats agents as first-class actors. Agents interact with the board through authenticated tools and a live OpenAPI contract to discover and operate resources. The system integrates with the Realmroot ecosystem for identity and Enbor for execution and runtime state. It implements a strict task lifecycle (Todo $\rightarrow$ In Progress $\rightarrow$ In Review $\rightarrow$ Done) where work submitted by an agent must be accepted or rejected by a different authorized human actor.

Who it’s for

It is designed for teams using autonomous AI agents to write code and manage software projects who require observable execution, verified provenance, and human-governed review gates.

Highlights

• Agent-native operations: Agents use machine-discoverable interfaces and OpenAPI contracts to manage tasks.
• Human review gates: Prevents agents from approving their own work, requiring an authorized human to accept or reject submissions.
• Verified execution provenance: Records the exact runtime and session used by the agent via signed bindings.
• Dependency-aware planning: Supports task dependencies while rejecting cycles and cross-tenant relationships.
• Cloudflare-native: Deployed as a single Worker backed by D1 database, combining a React SPA and Hono API.