oem-orchestrator - SKILL.md Agent Skill

name: oem-orchestrator description: Traffic Controller — central orchestrator that monitors, retries, and escalates across all 19 OEMs autonomously

OEM Orchestrator (Traffic Controller)

Central intelligence skill that provides autonomous oversight of all 19 OEMs. Runs every 2 hours via OpenClaw cron, with on-demand access via admin API.

Capabilities

1. Health Monitoring

Queries import_runs from the last 24 hours per OEM
Classifies each OEM: healthy (>70% success), degraded (30-70%), failing (<30%), stale (0 runs/36h)
Tracks per-OEM metrics: success rate, run count, last success time, last error

2. Auto-Retry with Exponential Backoff

Automatically re-triggers crawls for failing and stale OEMs
Exponential backoff: 30min → 60min → 2h → 4h → 8h → 12h max
Maximum 5 consecutive retries before escalation
Retry state persisted in R2 (memory/controller/retry-state.json)

3. Slack Escalation

Sends alerts when OEMs are degraded, failing, or stale
Per-OEM status with emoji indicators (🔴 failing, 🟡 degraded, ⚪ stale)
Escalation alert when max retries exceeded (human intervention needed)
Retry count + last error included in alerts

4. System Status API

GET /api/v1/oem-agent/admin/system-status — on-demand health check
Returns full per-OEM breakdown: status, success rate, runs, products, offers, retry count
Used by dashboard and support staff

Configuration

{
  "id": "oem-orchestrator",
  "schedule": "0 */2 * * *",
  "skill": "oem-orchestrator",
  "config": {
    "max_consecutive_failures": 5,
    "base_backoff_minutes": 30,
    "max_backoff_minutes": 720,
    "success_rate_threshold": 0.7,
    "stale_hours_threshold": 36,
    "notify_on_degradation": true
  }
}

Implementation

Controller: src/sync/orchestrator-controller.ts
Cron handler: src/routes/cron.ts → case 'oem-orchestrator'
Admin API: src/routes/oem-agent.ts → GET /admin/system-status
R2 state: memory/controller/retry-state.json

Companion Skills

Skill	Schedule	Relationship
`oem-agent-hooks` (health_check)	Every 4h	Detailed per-OEM import_run analysis, Slack alerts
`oem-agent-hooks` (memory_sync)	Every 30min	R2 backup of OEM configs + discovered APIs
`oem-agent-hooks` (generate_report)	Monday 9am	Weekly metrics digest to Slack
`oem-data-sync`	Daily 3am	Color + pricing sync (Kia, Hyundai, Mazda, Mitsubishi, VW + generic)
`oem-extract`	Daily 6am	Page crawling across all 19 OEMs
`oem-brand-ambassador`	Tuesday 4am	AI page generation (respects manual edits)

Status Classification

Status	Criteria	Action
✅ Healthy	>70% success rate, runs in last 36h	None
🟡 Degraded	30-70% success rate	Monitor, Slack alert
🔴 Failing	<30% success rate	Auto-retry with backoff, Slack alert
⚪ Stale	0 runs in 36h	Auto-retry, investigate if persists
🚨 Escalated	5+ consecutive failures	Slack escalation, human intervention needed