Clade — Control

Start, stop, or manage autonomous coding loops on a remote machine.

When to use

User wants to:

• Start a new coding loop with a goal
• Stop a running loop
• Clear blockers so a loop can resume

API

Endpoint: POST {CLADE_BASE_URL}/control

Headers:

Authorization: Bearer {CLADE_API_KEY}
Content-Type: application/json

Action: start

Start a new autonomous coding loop with a goal.

Request:

{
  "project": "/home/user/myproject",
  "action": "start",
  "goal": "Fix all failing tests and ensure 100% pass rate",
  "max_iter": 5,
  "max_workers": 4,
  "model": "sonnet"
}

Translating user intent to parameters:

• "run 3 times" / "跑3次" → max_iter: 3
• "use opus" / "用opus" → model: "opus"
• "2 workers" / "2个worker" → max_workers: 2
• "quick run" → max_iter: 3, model: "haiku"
• "thorough run" → max_iter: 10, model: "opus"
• If user doesn't specify: default to max_iter: 10, max_workers: 4, model: "sonnet"

The goal can be:

• A natural language description: "Fix the login page CSS on mobile"
• A file path on the server: "/home/user/myproject/goals/feature-x.md"

Response:

{
  "ok": true,
  "action": "start",
  "pid": 12345,
  "goal": "/home/user/myproject/.claude/monitor-goal.md",
  "max_iter": 5
}

Action: stop

Stop the currently running loop gracefully.

Request:

{
  "project": "/home/user/myproject",
  "action": "stop"
}

Response:

{"ok": true, "action": "stop", "sentinel": "/home/user/myproject/.claude/stop-start"}

Action: clear-blockers

Clear blockers so the loop can be restarted.

Request:

{
  "project": "/home/user/myproject",
  "action": "clear-blockers"
}

How to format the response

On start:

🚀 Loop started!
📋 Goal: {goal summary, first 80 chars}
⚙️ Config: {max_iter} iterations, {max_workers} workers, {model}
🔢 PID: {pid}

Use "status" to check progress.

On stop:

🛑 Stop signal sent. The loop will finish its current task and exit gracefully.

On clear-blockers:

🧹 Blockers cleared. You can now restart the loop.

Error handling

• Connection refused: "Monitor is not running."
• 401 Unauthorized: "Check your CLADE_API_KEY."
• 400 "goal is required": Ask the user what they want the AI to work on.
• 500 "start.sh not found": "Clade is not installed on the server. Run ./install.sh first."