kanban-mcp: Skill for Claude Code

Kanban Sprint - Ralph Wiggum Full Development Cycle

You are running a Sprint - a full development cycle that orchestrates Architect, Agent, and QA roles with iterative refinement (Ralph Wiggum pattern) and session tracking for cross-context-window continuity.

Arguments

Optional task/feature description after the command. Example: /kanban-sprint implement user authentication

Agent Naming

IMPORTANT: Use descriptive agent names based on task specialization, NOT generic names.

Good examples:

• frontend-specialist - For UI/React/CSS work
• backend-engineer - For API/database work
• test-writer - For test coverage
• security-reviewer - For security tasks
• devops-agent - For infrastructure work

Bad examples (too generic):

• agent-alpha, agent-beta, agent-gamma

Sprint Lifecycle

PLANNING -> EXECUTING -> REVIEWING -> COMPLETE/FAILED
             ^    |
             |    v (if rejections)
             +----+

A sprint can have multiple iterations. Each iteration:

1. Agents work on tasks
2. QA reviews
3. If rejections, loop back (up to maxIterations)

Execution Instructions

Phase 0: Session Start & Learning Context

Start a session and get project insights:

kanban_session_start with agentId: "sprint-orchestrator"

This returns:

• Board state and recent activity
• Last session notes for continuity
• Urgent items (escalated, blocked, critical)
• Learning context

Also get detailed learning insights:

kanban_get_learning_insights with role: "architect"

Review:

• Past project lessons
• Codebase conventions
• Common patterns to follow

Phase 1: Planning

Create the sprint first:

kanban_sprint_create:
  role: "architect"
  goal: "[User's goal description]"
  successCriteria:
    - "Criterion 1"
    - "Criterion 2"
    - "All tests pass"
  maxIterations: 5

Generate a summary for sub-agents:

kanban_generate_summary

Then spawn an Architect agent:

Task tool:
  subagent_type: "general-purpose"
  description: "Architect planning sprint"
  prompt: |
    You are the ARCHITECT for Kanban sprint [SPRINT_ID].

    GOAL: [User's goal]

    1. Start session: kanban_session_start with agentId: "architect"
    2. Get learning insights: kanban_get_learning_insights
    3. Analyze the codebase to understand patterns
    4. Create 3-8 tasks using kanban_create_task with:
       - role: "architect"
       - sprintId: "[SPRINT_ID]"
       - acceptanceCriteria: { description, verificationSteps, testCommand }
       - maxIterations: 3
    5. Set priorities and dependencies
    6. Assign tasks to descriptive agent names based on task type:
       - frontend-specialist (UI/React work)
       - backend-engineer (API/database work)
       - test-writer (test coverage)
       - Or other descriptive names matching the task domain
    7. End session: kanban_session_end with sessionNotes

    IMPORTANT: Each task must have clear acceptance criteria!
    IMPORTANT: Use descriptive agent names, NOT generic names like agent-alpha!

    Report: tasks created with their acceptance criteria and agent assignments.