Mission Control

Mission Control Integration Skill

You are connected to a Mission Control dashboard that tracks your work, manages projects and tasks, handles approvals, and stores documents in a Library. This skill tells you how to communicate with it.

Overview

Mission Control is a database-backed dashboard running at $MISSION_CONTROL_URL (default: http://localhost:8000). It provides:

• Pipeline: Tasks flow through stages: backlog → todo → doing → review → done
• Inbox: A unified queue where humans review requests, approve work, and handle workflow gates
• Library: A curated document store for research, reports, documentation, and reference material
• Costs: Token spend tracking per session
• Activity: A chronological feed of all events

Your job is to report your progress so the dashboard stays accurate.

Reporting Task Progress

When you are assigned a task (you'll see a task ID like t_abc123), report your progress at each stage by sending HTTP POST requests to the hooks endpoint.

Endpoint

POST $MISSION_CONTROL_URL/api/hooks/event
Content-Type: application/json
X-Hook-Secret: $MISSION_CONTROL_HOOK_SECRET

When Starting a Task

{
  "event": "task:started",
  "agentId": "YOUR_AGENT_ID",
  "taskId": "TASK_ID",
  "data": { "message": "Beginning work on [description]" }
}

When Making Progress

{
  "event": "task:progress",
  "agentId": "YOUR_AGENT_ID",
  "taskId": "TASK_ID",
  "data": { "progress": 50, "message": "Completed first draft" }
}

When Submitting for Review

Use this when your work is done and ready for human review. This automatically creates a review record in the Inbox with your deliverables and checklist.

{
  "event": "task:review",
  "agentId": "YOUR_AGENT_ID",
  "taskId": "TASK_ID",
  "data": {
    "work_summary": "Describe what you did and what the human should look at",
    "deliverables": [
      { "name": "report.md", "type": "doc", "path": "/workspace/docs/report.md", "summary": "The main report" },
      { "name": "data.json", "type": "file", "path": "/workspace/data/output.json", "summary": "Raw analysis data" }
    ],
    "checklist": [
      { "label": "Meets requirements", "checked": true },
      { "label": "No errors", "checked": true },
      { "label": "Well documented", "checked": false }
    ]
  }
}

When a Task is Complete (after human approval)

{
  "event": "task:completed",
  "agentId": "YOUR_AGENT_ID",
  "taskId": "TASK_ID",
  "data": { "message": "Task completed successfully" }
}

When a Task Fails

{
  "event": "task:failed",
  "agentId": "YOUR_AGENT_ID",
  "taskId": "TASK_ID",
  "data": { "reason": "Explain what went wrong and why" }
}

Publishing to the Library

When you produce research, reports, documentation, analysis, or any reference material, publish it to the Library so it's organized and searchable.

Publishing a New Document

{
  "event": "library:publish",
  "agentId": "YOUR_AGENT_ID",
  "data": {
    "title": "Document Title",
    "content": "Full document content in markdown, JSON, or plain text",
    "doc_type": "research",
    "format": "markdown",
    "collection_id": "col_research",
    "projectId": "PROJECT_ID_IF_RELEVANT",
    "source_path": "/workspace/docs/my-report.md"
  }
}

Document types: research, report, documentation, reference, template, note, analysis, brief, guide, other

Formats: markdown, html, code, json, text, csv

Default collections (use these IDs):

• col_research — Research findings, competitive analysis, market data
• col_reports — Status reports, weekly summaries, metrics
• col_docs — Technical documentation, API docs, guides
• col_reference — Reference material, specifications, standards
• col_templates — Reusable templates and boilerplate
• col_notes — Meeting notes, scratch work, ideas

Updating an Existing Document

{
  "event": "library:update",
  "agentId": "YOUR_AGENT_ID",
  "data": {
    "doc_id": "DOCUMENT_ID",
    "content": "Updated content",
    "change_note": "What changed and why"
  }
}

Requesting Approvals

When you need human approval before proceeding (deploying to production, making purchases, deleting data, etc.), create a workflow gate:

{
  "event": "approval:needed",
  "agentId": "YOUR_AGENT_ID",
  "data": {
    "type": "workflow_gate",
    "title": "Deploy to production?",
    "description": "All tests pass. 3 files changed. Ready to deploy.",
    "urgency": "urgent",
    "entityType": "task",
    "entityId": "TASK_ID",
    "resumeToken": "UNIQUE_TOKEN_TO_RESUME"
  }
}

The human will see this in their Inbox and can approve or reject. If you provided a resumeToken, the backend stores it so you can check whether approval was granted.

Urgency levels: critical, urgent, normal, low

Submitting Project Requests

When you identify new work that should be done (a bug you found, an improvement idea, research that's needed), submit a project request:

{
  "event": "request:submit",
  "agentId": "YOUR_AGENT_ID",
  "data": {
    "title": "Request title",
    "description": "Detailed description of what's needed and why",
    "category": "bug",
    "urgency": "urgent"
  }
}

Categories: feature, bug, research, content, ops, automation, general

Reporting Session Lifecycle

When a Session Starts

Report session creation so Mission Control tracks active sessions:

{
  "event": "session:created",
  "agentId": "YOUR_AGENT_ID",
  "data": { "sessionId": "sess_abc123" }
}

When a Session Ends (Reporting Costs)

When your session ends, report token usage so the Costs dashboard stays accurate:

{
  "event": "session:ended",
  "agentId": "YOUR_AGENT_ID",
  "data": {
    "tokens": {
      "model": "claude-sonnet-4-5",
      "provider": "anthropic",
      "input": 45000,
      "output": 12000,
      "cacheRead": 30000,
      "totalCost": 0.82
    }
  }
}

Reporting Agent Status

When Idle (no more tasks)

{
  "event": "agent:idle",
  "agentId": "YOUR_AGENT_ID",
  "data": { "message": "All assigned tasks complete" }
}

When an Error Occurs

{
  "event": "agent:error",
  "agentId": "YOUR_AGENT_ID",
  "data": { "error": "Description of the error", "context": "What you were trying to do" }
}

Using the /mc Shorthand

If your session supports slash commands, you can use these shortcuts instead of crafting full HTTP requests:

/mc task:started <taskId>
/mc task:completed <taskId>
/mc task:failed <taskId> <reason>
/mc task:review <taskId>
/mc task:progress <taskId> <percentage> <message>

The lifecycle hook translates these into the proper API calls automatically.

Polling for Pending Events

Mission Control pushes events to you via the gateway, but delivery isn't guaranteed (gateway might be down, session might have restarted). At the start of each session, check if any events are waiting for you:

GET $MISSION_CONTROL_URL/api/dispatch/pending/YOUR_AGENT_ID

Response:

{
  "events": [
    {
      "id": 1,
      "event": "mc:project_kickoff",
      "payload": { "type": "mc:event", "event": "mc:project_kickoff", "projectId": "abc", "projectName": "New Feature", "..." },
      "created_at": "2026-03-21T20:00:00Z"
    }
  ],
  "count": 1
}

Calling this endpoint marks the events as delivered. Process each one according to the event type (see the outbound events section below).

To check without consuming: GET /api/dispatch/pending/YOUR_AGENT_ID?peek=true

To explicitly acknowledge after processing: POST /api/dispatch/ack with { "ids": [1, 2, 3] }

Best practice: Check for pending events at session start and after each heartbeat.

Best Practices

1. Always report task:started when you begin work — this moves the task to "In Progress" on the dashboard
2. Use task:review when work is done — don't use task:completed directly. Let the human review first.
3. Include deliverables in your review submission — list every file you created or modified
4. Publish to the Library when you create lasting content — reports, research, documentation
5. Pick the right collection — don't dump everything in "Notes". Research goes in Research, docs go in Documentation.
6. Request approval for risky actions — deployments, deletions, purchases, external API calls with side effects
7. Report failures honestly — include what went wrong and what you tried. The human needs context to help.
8. Include a work_summary in reviews — explain what you did, what choices you made, and what the human should focus on when reviewing.

Quick Reference

Events You Will RECEIVE from Mission Control

Mission Control sends events back to you when a human makes a decision. These arrive as messages in your session with type: "mc:event". When you receive one, act on it.

mc:review_approved

Your work was approved. The task is now "done". Look for the next task.

Fields: taskId, taskTitle, projectId, projectName, round, qualityScore, notes, nextAction

What to do: Check if the project has more tasks in "todo" stage. If so, pick the next one and start it.

mc:changes_requested

Your work needs revisions. The task has been moved back to "doing". Read the feedback and fix the issues.

Fields: taskId, taskTitle, round, feedback, currentStage

What to do: Read the feedback field carefully. It contains the human's specific concerns. Revise your work, then resubmit with task:review.

mc:review_rejected

Your work was rejected. The task is back in "todo" and may be reassigned.

Fields: taskId, taskTitle, round, reason, currentStage

What to do: Read the reason. If you can address it, pick the task back up. If not, move on to other work.

mc:approval_granted

A workflow gate you requested was approved. Resume your paused work.

Fields: approvalId, approvalType, title, resumeToken, entityType, entityId, notes, nextAction

What to do: If resumeToken is present, use it to resume the paused workflow. If nextAction is "resume_workflow", continue from where you stopped.

mc:approval_denied

A workflow gate you requested was denied.

Fields: approvalId, approvalType, title, entityType, entityId, reason, nextAction

What to do: Read the reason. You may need to take a different approach or abandon the action.

mc:project_kickoff

A new project was created (from an approved request or directly). You are the owner agent. Start planning.

Fields: projectId, projectName, description, priority, fromRequest, requestTitle, hint

What to do: Break the project down into tasks. Create them via POST /api/tasks with project_id set. Put tasks in "backlog" or "todo" stage. Then start working on the highest-priority one.

mc:task_assigned

A task was assigned to you.

Fields: taskId, taskTitle, projectId, priority, currentStage, previousAgent, nextAction

What to do: If currentStage is "todo", start when ready. If "doing", continue work immediately.

mc:task_resume

A task you own was moved back to "doing" (e.g., after changes requested).

Fields: taskId, taskTitle, fromStage, toStage, movedBy, nextAction

What to do: Resume work on this task. Check if there's review feedback in the task metadata.