mission-control（任务控制）

Mission Control — Task Management for AI Assistants

A Kanban-style task board that you (the AI assistant) manage. Your human creates and prioritizes tasks via the web dashboard; you execute them automatically when they're moved to "In Progress".

🚀 Quick Start

Just say: *"Set up Mission Control for my workspace"*

The agent will:

1. Check prerequisites (Tailscale, gh CLI)
2. Copy dashboard files to your workspace
3. Create the config file (~/.clawdbot/mission-control.json)
4. Install the webhook transform
5. Set up GitHub webhook
6. Push to GitHub and enable Pages

That's it. The agent handles everything.

Prerequisites

Before setup, you need:

If any are missing, tell the agent — it will guide you through installation.

How It Works

1. Dashboard — Web UI hosted on GitHub Pages where humans manage tasks
2. Webhook — GitHub sends push events to Clawdbot when tasks change
3. Transform — Compares old vs new tasks.json, detects status changes
4. Auto-Processing — When a task moves to "In Progress", the agent starts working

The Flow

Human moves task → GitHub push → Webhook → Transform → Agent receives work order
      ↓                                                         ↓
   Dashboard                                              Executes task
      ↓                                                         ↓
Agent updates status ← Commits changes ← Marks subtasks done ←─┘

Task Structure

Tasks live in <workspace>/data/tasks.json:

{
  "id": "task_001",
  "title": "Implement feature X",
  "description": "Detailed context for the agent",
  "status": "backlog",
  "subtasks": [
    { "id": "sub_001", "title": "Research approach", "done": false },
    { "id": "sub_002", "title": "Write code", "done": false }
  ],
  "priority": "high",
  "dod": "Definition of Done - what success looks like",
  "comments": []
}

Status Values

CLI Commands

Use <skill>/scripts/mc-update.sh for task updates:

# Status changes
mc-update.sh status <task_id> review
mc-update.sh status <task_id> done

# Comments
mc-update.sh comment <task_id> "Progress update..."

# Subtasks
mc-update.sh subtask <task_id> sub_1 done

# Complete (moves to review + adds summary)
mc-update.sh complete <task_id> "Summary of what was done"

# Push to GitHub
mc-update.sh push "Commit message"

Agent Workflow

When you receive a task (moved to "In Progress"):

1. Read — Check title, description, subtasks, dod
2. Mark started — mc-update.sh start <task_id>
3. Execute — Work through subtasks, mark each done
4. Document — Add progress comments
5. Complete — mc-update.sh complete <task_id> "Summary"

Handling Rework

If a completed task is moved back to "In Progress" with a new comment:

1. Read the feedback comment
2. Address the issues
3. Add a comment explaining your changes
4. Move back to Review

EPICs

EPICs are parent tasks with multiple child tickets. When you receive an EPIC:

1. Child tickets are listed in the subtasks (format: MC-XXX-001: Title)
2. Work through them sequentially (1 → 2 → 3...)
3. After each child: comment result, set to "review", mark EPIC subtask done
4. After last child: set EPIC to "review"

Heartbeat Integration

Add to your HEARTBEAT.md:

## Task Check

1. Check `data/tasks.json` for tasks in "in_progress"
2. Flag tasks with `processingStartedAt` but no recent activity
3. Check "review" tasks for new feedback comments

Configuration

Config lives in ~/.clawdbot/mission-control.json. See assets/examples/CONFIG-REFERENCE.md for all options.

Minimal config (set by agent during setup):

{
  "gateway": { "hookToken": "your-token" },
  "workspace": { "path": "/path/to/workspace" },
  "slack": { "botToken": "xoxb-...", "channel": "C0123456789" }
}

Troubleshooting

See docs/TROUBLESHOOTING.md for common issues:

• Dashboard shows sample data → Connect GitHub token
• Webhook not triggering → Check Tailscale Funnel
• Changes not appearing → GitHub Pages cache (wait 1-2 min)

Security

Mission Control is a task management system for AI agents — its core purpose is to pass human-authored task descriptions to an agent for execution. This is by design, not a vulnerability.

Trust Model

• Single-user / trusted-user setup: Task authors are the same people who control the agent. The trust boundary is identical to typing a message directly to your assistant.
• Multi-user setups: If multiple users can create tasks on the dashboard, treat task content as untrusted input. Use Clawdbot's agent sandbox and permission model to limit what the agent can do.

Mitigations

• Input sanitization: mc-update.sh validates all inputs against injection patterns before passing them to Python or git.
• No credential storage: The dashboard stores no tokens or secrets — all auth is handled by Clawdbot's config.
• Webhook HMAC verification: The transform module validates webhook signatures using timingSafeEqual to prevent tampering.
• Security scan on sync: The sync-to-opensource.sh script scans for leaked credentials before publishing.

Recommendations

• Keep your dashboard repository private if you don't want others to see your task data.
• Review task descriptions before moving them to "In Progress" if the task was created by someone else.
• Use Clawdbot's groupPolicy and allowFrom settings to restrict who can interact with the agent.

Files