Eternal Autonomous Mode

Eternal Autonomous Mode

Eternal mode lets the Leader self-patrol, continuously monitoring and managing engineering state without human intervention. Through state machine drive, budget circuit breaking, and 3-layer health checks, the autonomous mode is safe and controllable.

State Machine

IDLE→CHECK→PATROL→THINK→WAIT↺loop

State	Description
IDLE	Idle waiting, base 30-second interval
CHECK	Check system status, task queue, and environment health
PATROL	Patrol engineering directory, detect file changes and anomalies
THINK	Analyze discovered issues, decide whether action is needed
WAIT	Wait for next cycle, exponential backoff

State Flow

IDLE (30s) → CHECK → Issues? → PATROL → THINK → Action? → Execute

↘No issues → WAIT → IDLE (backoff) ↘Not needed → WAIT

Budget Circuit Breaker

To prevent Eternal mode from running away, a budget circuit breaker is in place:

• Consecutive failure counter: Tracks consecutive failure count
• Circuit breaker threshold: Pauses Eternal mode after 8 consecutive failures
• Auto recovery: After circuit breaking, requires manual restart or Supervisor auto-recovery
• Exponential backoff: Cycle interval grows exponentially with failure count

Base interval

30s

Backoff strategy

30s × 2^n (n = consecutive failure count)

Circuit breaker

8 consecutive failures → pause

EternalSupervisor

EternalSupervisor provides 3-layer health checks and auto-restart:

3-Layer Health Check

Layer	Check Method	Description
PID Check	Process PID liveness detection	Check if the Eternal process is running
Watchdog	Heartbeat watchdog	Check if Eternal is heartbeating normally
HTTP Probe	HTTP health endpoint	Check if Eternal can respond normally

Auto Restart

When any layer check fails, the Supervisor will:

1. Record the failure reason and time
2. Clean up residual processes and locks
3. Restore last state from SQLite
4. Restart the Eternal state machine
5. If consecutive restart failures exceed the threshold, enter a full stop state and notify the user

Configuration

Configure Eternal mode in ~/.lingxiao/settings.json:

{
  "eternal": {
    "enabled": true,
    "baseInterval": 30000,
    "maxBackoff": 600000,
    "budgetCircuitBreaker": 8,
    "supervisor": {
      "enabled": true,
      "checkInterval": 60000,
      "httpTimeout": 5000
    }
  }
}

Use Cases

• Continuous monitoring: Long-running engineering projects needing automatic inspection
• Auto repair: Automatically trigger repair flows when issues are detected
• Night patrol: Monitor build and test status during off-hours
• CI/CD assistance: Continuously observe deployment status and respond to anomalies