This skill should be used when a user asks about orchestrating Claude Code with vibe-kanban, running parallel Claude Code sessions, designing tasks for AI coding agents, or managing a kanban board connected to Claude Code. It also applies when the user says "vibe-kanbanのワークフロー", "並列でClaude Codeを使いたい", "タスク設計のコツ", "how to run multiple Claude Code agents", "connect Claude Code to a kanban board", "worktree-based task management", or "AIエージェントのタスクを管理したい".
From vibe-kanbannpx claudepluginhub utakatakyosui/c2lab --plugin vibe-kanbanThis skill uses the workspace's default tool permissions.
references/mcp-configuration.mdreferences/parallel-workflow.mdreferences/task-design-patterns.mdDesigns and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
vibe-kanban is a local-first kanban board designed to orchestrate AI coding agents (Claude Code, Codex, Gemini CLI). Each task runs in an isolated git worktree, enabling parallel development with multiple Claude Code sessions.
Install and start vibe-kanban:
npx vibe-kanban
The UI opens at http://localhost:3000 by default. Connect to a git repository by pointing to the project directory.
Configure Claude Code as an agent in vibe-kanban's agent settings, selecting:
For interactive setup guidance, run /vibe-kanban:setup (a slash command provided by this plugin).
Write issues that Claude Code can execute autonomously without interactive clarification. Good tasks are specific, bounded, and self-contained.
Effective task structure:
Title: [Action verb] [specific component] [measurable outcome]
Description:
- Context: What already exists, what the current behavior is
- Goal: Exact expected result or behavior
- Acceptance criteria: How to verify completion
- Files affected: Key files/directories to look at (optional)
- Do NOT do: Explicit out-of-scope items
Good examples:
Avoid:
See references/task-design-patterns.md for detailed templates and antipatterns.
Run multiple Claude Code sessions simultaneously for independent work:
Critical constraint: Avoid parallel tasks that modify the same files. Merge conflicts require manual resolution outside vibe-kanban.
See references/parallel-workflow.md for patterns like feature slicing and independent module strategy.
Before starting a task:
http://localhost:3000During a session:
After session completion:
Status section in the review view)This plugin includes a .mcp.json that connects Claude Code directly to vibe-kanban's MCP server:
npx vibe-kanban@latest --mcp
When the plugin is enabled, the vibe_kanban MCP server starts automatically. Use /mcp to verify available tools (e.g., listing tasks, updating status, creating issues).
Using MCP tools to manage tasks:
vibe-kanban also centralizes MCP server configuration for all connected agents. Configure additional MCP servers once in vibe-kanban's agent settings under MCP Servers and they propagate to all Claude Code sessions automatically.
See references/mcp-configuration.md for common MCP server configurations to add to vibe-kanban.
| Status | Meaning | Next Actions |
|---|---|---|
| Backlog | Not yet started | Prioritize, refine description |
| In Progress | Agent running | Monitor, wait for completion |
| Review | Agent done, awaiting review | Inspect diff, run tests |
| Done | Reviewed and merged | Close issue |
| Cancelled | Dropped or superseded | Archive |
Move tasks through statuses manually after reviewing completed work. vibe-kanban does not auto-close tasks.
Session hangs or produces no output:
~/.claude/credentials)Worktree conflict on task start:
git worktree pruneDiff is empty after session: