cto-platform（CTO 平台）

CTO Platform Skill

Comprehensive knowledge of the CTO (Cognitive Task Orchestrator) platform for AI-driven development workflows.

When to Use

• Running intake to generate tasks from PRDs
• Starting and monitoring Play workflows
• Understanding agent sequencing and responsibilities
• Debugging workflow issues

Intake Workflow

MCP Tool

mcp_cto_intake(project_name="my-project")

Flow

PRD Document → mcp_cto_intake() → Linear Project Created
                                → PRD Issue with Morgan as delegate
                                → PM Webhook → Argo Workflow
                                → Morgan Pod generates tasks.json
                                → Task issues created in Linear

Key Points

• local: false in cto-config.json = Argo workflow (production)
• local: true = local binary execution (debugging only)
• Morgan auto-assigned triggers intake via PM webhook

Play Workflow

MCP Tools

mcp_cto_play()           # Start workflow, auto-detects next task
mcp_cto_play_status()    # Query progress
mcp_cto_jobs()           # List running workflows
mcp_cto_stop_job()       # Cancel workflow

Agent Sequence (Per Task)

┌─────────────────────────────────────────────────────────────────────────────┐
│                           PLAY WORKFLOW (Per Task)                           │
│                                                                              │
│   1. IMPLEMENTATION AGENT (based on task.agentHint)                         │
│      ├─ bolt (infrastructure - Task 1 ALWAYS, plus any infra tasks)        │
│      ├─ rex (Rust: axum, tokio, sqlx)                                       │
│      ├─ grizz (Go: chi, grpc, pgx)                                          │
│      ├─ nova (Node.js/Bun: Elysia, Effect, Drizzle)                         │
│      ├─ blaze (React/Web: Next.js, shadcn/ui, TailwindCSS)                  │
│      ├─ tap (Mobile: Expo, React Native)                                    │
│      └─ spark (Desktop: Electron)                                           │
│                                                                              │
│   2. SUPPORT AGENTS (sequential, run for EVERY task)                        │
│      ├─ Cleo (quality) ──── max 5 retries                                   │
│      ├─ Cipher (security) ─ max 2 retries                                   │
│      ├─ Tess (testing) ──── max 5 retries                                   │
│      └─ Atlas (integration) final merge gate                                │
└─────────────────────────────────────────────────────────────────────────────┘

cto-config.json Structure

{
  "version": "1.0",
  "orgName": "5DLabs",
  "defaults": {
    "intake": {
      "local": false,           // MUST be false for production
      "githubApp": "5DLabs-Morgan"
    },
    "linear": {
      "teamId": "CTOPA",
      "pmServerUrl": "http://localhost:8081"
    },
    "play": {
      "repository": "5dlabs/cto",
      "healerEndpoint": "http://localhost:8083",
      "autoMerge": true,        // Atlas auto-merges PRs
      "implementationMaxRetries": 10,
      "qualityMaxRetries": 5,
      "securityMaxRetries": 2,
      "testingMaxRetries": 5
    }
  },
  "agents": {
    "morgan": { ... },
    "bolt": { ... },
    "rex": { ... },
    // ...
  }
}

Key Fields

Execution Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                        IN-CLUSTER EXECUTION FLOW                             │
│                                                                              │
│   MCP Tool Call (Droid/Cursor)                                              │
│       └─→ Submits Argo Workflow                                             │
│           └─→ Controller creates CodeRuns (pods)                            │
│               └─→ Each agent runs in isolated container                     │
│                   └─→ Linear sidecar streams activities                     │
│                       └─→ Healer monitors via Loki                          │
└─────────────────────────────────────────────────────────────────────────────┘

Local Development

Services to Run

just dev-controller       # Controller
just dev-pm              # PM Server (:8081)
just dev-healer-play-api # Healer (:8083)

# Or all at once:
mprocs

Port Forwards

kubectl port-forward svc/argocd-server -n argocd 8080:80
kubectl port-forward svc/argo-workflows-server -n automation 2746:2746

Autonomy Permissions

When running E2E tests with full autonomy:

1. MAY merge pull requests via Atlas
2. MAY update Linear issues directly
3. MAY execute all MCP tools without confirmation
4. MAY handle all agent transitions
5. MAY create and close GitHub issues

Reference Documentation

• docs/linear-integration-workflow.md - Linear webhook flow
• docs/heal-play.md - Healer monitoring
• cto-config.template.json - Config field documentation
• AGENTS.md - Agent overview