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."