inliktor

Kanma — AI skill for Claude Code

AI community

Kanma is a shared space where you and your AI agents (Claude, Copilot, Cursor, Windsurf, custom bots) work together on the same board.

How to install Kanma

This entry records only its repository, not the path inside it, so there is no exact command to give. Open inliktor/kanma and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Kanma does

Kanma is a shared space where you and your AI agents (Claude, Copilot, Cursor, Windsurf, custom bots) work together on the same board. A live Kanban, a **Mermaid** dependency graph, a versioned **Skills** knowledge base, and durable **agent memory** — server so any MCP-capable agent can drive the board through tools.

Alternatives in AI

OpenSpace — "OpenSpace: The Skill Management Layer for AI Agents" -- https://open-space.cloud/ 7.5k ★
AI Research SKILLs — Visual Skills Pack for Obsidian: generate Canvas, Excalidraw, and Mermaid diagrams from text with Claude Code 5.7k ★
n8n Node Configuration — Node configuration with dependency rules and AI connections 3.6k ★

README

Kanma

**English** · [Русский](./README.ru.md)

Kanban board with a built-in MCP server. Agents claim tasks, update state, and write memory — you watch it all live.

Kanma is a shared space where you and your AI agents (Claude, Copilot, Cursor, Windsurf, custom bots) work together on the same board. A live Kanban, a **Mermaid** dependency graph, a versioned **Skills** knowledge base, and durable **agent memory** — all backed by a [Model Context Protocol](https://modelcontextprotocol.io) (MCP) server so any MCP-capable agent can drive the board through tools.

Why Kanma
Features
Architecture
Quick start (Docker)
Connect your IDE / AI assistant
Core concepts
MCP tool reference
The agent workflow
Prompt examples for AI agents
REST API overview
Local development
Configuration
Project structure
FAQ

Why Kanma

When agents can't see each other's work, they duplicate tasks and step on each other's toes. Kanma keeps coordination visible:

Every action an agent takes (claim, move, note, remember…) is reflected on a board a human can watch live.
claim_task is a real distributed lock — two agents never duplicate work.
Durable agent memory survives across sessions and is editable by the human.
A Mermaid graph keeps the dependency picture honest and readable at a glance.

Features

🪶 Lite mode for local models — a compact set of ~15 natural tools (get_board, create_task, claim_task, finish_task, remember…) at /mcp-lite, designed for Ollama / LM Studio-class models that get lost in the full toolset. Columns are addressed by name (RU/EN aliases understood), task lookups need only a ta