⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
kanbu
Manage projects, tasks, wikis, and teams in Kanbu — a self-hosted project management platform built for AI agents. Use when the user asks about project management, task tracking, kanban boards, sprints, wikis, team workload, or anything related to Kanbu. Also triggers on task creation, updates, comments, subtasks, search, analytics, GitHub integration, permissions (ACL), and audit logs.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
Kanbu
Self-hosted project management for people who work with AI agents.
Built for people who work with AI agents
I built Kanbu because I use OpenClaw every day. My projects regularly cross 300,000 lines of code — and at that scale, you need somewhere to track what your agents are doing, what's been decided, and what comes next. No tool did that. So I built one.
Kanbu connects directly to your local OpenClaw gateway. Dispatch any task to an agent from the task view, read its response, steer it with follow-up instructions, and let it update the board as it works. Your task list becomes the shared memory between you and your agents.
Works with any MCP-compatible agent — OpenClaw, Claude Code, Cursor, or any tool that speaks MCP.
It's MIT. It's self-hosted. It's here because I needed it, and maybe you do too.
Why Kanbu?
Most project management tools are either too simple (Trello) or too complex (Jira). And none of them understand AI-native workflows.
Kanbu is different:
| Problem | Kanbu Solution |
|---|---|
| "I need fine-grained permissions but Trello doesn't have them" | Scoped ACL with inheritance, deny rules, and security groups |
| "I want AI to help but it can't access my tasks" | 154 MCP tools - Claude Code works directly in your board |
| "I want to dispatch tasks to AI agents and see what they do" | OpenClaw agent dispatch — one click, agent runs, response logged |
| "GitHub issues and my PM tool are never in sync" | Bi-directional GitHub sync with webhook integration |
| "I want to self-host but lose features" | No seat fees, no feature gates — everything included, always |
| "My backups are not secure or automated" | Automated backups with AES-256-GCM encryption & scheduling |
| "My team speaks different languages" | Each user talks to their own AI assistant in their own language |
Key Features
🎯 Project Management
- Kanban boards with drag-and-drop, swimlanes, and WIP limits
- Multiple views: Board, List, Calendar, Timeline
- Sprints & Milestones with burndown charts
- Time tracking with estimates vs. actuals
- Real-time collaboration: Live cursors, typing indicators, and instant sync via Socket.io
🔐 Permissions & Access Control
Kanbu implements a permission system inspired by filesystem ACLs (think Windows NTFS / Active Directory, but for a web app).
- Granular permissions: Read, Write, Execute, Delete, Permissions — each independently grantable or deniable.
- Inheritance: Workspace → Project → Task permissions flow down automatically.
- Deny-first logic: Explicit deny overrides any grant (e.g., ban a specific user from a project even if they are an admin).
- Audit logs: Complete security trail with export to CSV/JSON.
🧠 Built-in Knowledge Graph Engine
Kanbu includes a wiki intelligence system built in TypeScript (33,000+ lines).
The base wiki runs without any extra services and includes:
- BM25 full-text search: Fast keyword search across all wiki pages
- Contradiction detection: Automatically flags conflicting information across your documentation
- Community clustering: Groups related knowledge using label propagation (own algorithm)
- Version history: Up to 20 versions per page, with restore
The full graph engine (optional, requires Python/FalkorDB via docker-compose.openclaw.yml) adds:
- Semantic search: Find knowledge using vector embeddings, not just keywords
- Temporal queries: Ask "What did we know about this feature last month?"
- Duplicate deduplication: Detects and merges overlapping knowledge nodes across pages
- RAG: Retrieval-Augmented Generation — AI answers grounded in your actual wiki content
- D3.js graph visualization: Interactive visual map of your knowledge graph, live in the browser
Inspired by Graphiti — thanks for the foundation.
🤖 AI Agent Dispatch (OpenClaw)
If you run OpenClaw, Kanbu connects directly to your local agent gateway. Assign any task to an AI agent with one click — no copy-pasting, no context switching.
- One-click dispatch: Open any task → Agent tab → select agent → Dispatch
- Task context included automatically: Title, description, project and workspace name are sent as a structured prompt
- Custom instructions per run: Add extra steering without editing the task
- Run history with response: Every run is logged with status, duration, and the agent's full response
- Iterative control loop: Dispatch → read response → dispatch again with new instructions
- Fire-and-forget: Returns immediately, status updates async as the agent works
Setup (2 env vars):
OPENCLAW_GATEWAY_URL=http://127.0.0.1:18789
OPENCLAW_GATEWAY_TOKEN=your-token-here
When not configured, the Agent tab shows a friendly "not configured" state instead of an error.
🐾 OpenClaw Skill
Kanbu ships with a ready-made OpenClaw skill. Install it once, and your OpenClaw agent knows how to manage projects, tasks, wikis, and teams — no manual configuration needed.
# Copy the skill into your OpenClaw workspace
cp -r packages/openclaw-skill ~/.openclaw/workspace/skills/kanbu
# Restart your gateway
openclaw gateway restart
That's it. Your agent can now create tasks, move cards, add comments, search across projects, run analytics, and manage your entire board — all through natural language.
🔌 AI Agent Integration (MCP)
Connect any MCP-compatible AI agent to manage your Kanbu projects with 154 available tools.
Tool Categories:
| Phase | Category | Tools | Examples |
|---|---|---|---|
| 1 | Pairing | 3 | `kanbu_connec |
// HOW IT'S BUILT
KEY FILES