Codex Reset Radar — Usage Reset Window Monitor

Monitors the Codex Reset Radar current.json endpoint to detect Codex usage quota reset windows ("speed windows"), pushing alerts via OpenClaw cron + Feishu.

How It Works

Cron triggers
     │
     ▼
① Data collection: python3 scripts/codex-radar-check.py
   → fetch current.json + compare against local cache
     │
     ▼
② JSON diff: outputs has_changes + events[] or has_changes: false
   → no changes → agent replies NO_REPLY (silent)
     │
     ▼
③ LLM formatting: reads JSON diff only
   → formats into Feishu message
     │
     ▼
④ Feishu push: announce → user session

Detection Script

scripts/codex-radar-check.py — stdlib only, zero dependencies:

• Fetches https://codex-reset-radar.pages.dev/current.json
• Compares against cache/codex-radar-last.json from previous run
• Creates baseline on first run

Change types detected:

Output JSON Examples

No changes:

{"has_changes": false}

Changes detected:

{
  "has_changes": true,
  "events": [
    {"type": "window_opened", "detail": "Codex usage reset window opened", "opened_at": "2026-05-24T08:21:33+08:00", "scope": "Codex users"}
  ],
  "current_status": {
    "window_open": true,
    "status": "open",
    "last_window_id": "codex-speed-window-2026-05-24-codex",
    "prediction_level": "low",
    "probability_24h": 0.06
  }
}

Installation

clawhub install codex-reset-radar

Cron Setup

Create an OpenClaw cron job (recommended: hourly 8 AM–11 PM, silent overnight):

{
  "name": "Codex Reset Radar Monitor",
  "schedule": {"kind": "cron", "expr": "0 8-23 * * *", "tz": "Asia/Shanghai"},
  "sessionTarget": "isolated",
  "payload": {
    "kind": "agentTurn",
    "timeoutSeconds": 60,
    "lightContext": true,
    "message": "Codex Reset Radar monitor.\
1. cd ~/.openclaw/workspace && python3 skills/codex-reset-radar/scripts/codex-radar-check.py\
2. If has_changes: false, reply NO_REPLY\
3. If changes detected, format and push to Feishu"
  },
  "delivery": {"mode": "announce", "channel": "feishu", "to": "user:your_open_id"}
}

Recommended schedules:

• 0 8-23 * * * — hourly during waking hours, silent 0-7
• */10 * * * * — every 10 minutes (urgent monitoring)

Design Principles

• LLM only sees JSON diffs — no raw RSS/HTML semantic analysis, minimal token usage
• Zero-cost silence — has_changes: false → agent replies NO_REPLY → nothing pushed to Feishu
• Stdlib only — uses json, urllib, datetime, os; zero external dependencies
• Fault-tolerant — network errors output {"has_changes": false, "error": "..."}, never false-trigger or crash