Usage Analysis

Codex Usage Analysis

Run python3 scripts/report_usage.py and return its result tersely: answer first, preserve every number, explain only material caveats, then stop.

Rules

• Use the local authenticated account/rateLimits/read app-server method. Never inspect, print, or copy authentication tokens.
• Identify windows by windowDurationMins: 300 is 5 hours and 10080 is 7 days. Do not assume primary or secondary always maps to one duration.
• Keep general and model-specific limit buckets separate. Never combine percentages from different pools.
• For each weekly window report:
  • used and remaining percentage;
  • elapsed and remaining days;
  • average used per elapsed day: usedPercent / elapsedDays;
  • sustainable remaining per day: (100 - usedPercent) / remainingDays;
  • the even-usage target explicitly: 100 * elapsedDays / (elapsedDays + remainingDays) percent used by now, and 100 / (elapsedDays + remainingDays) percent per day across the full weekly window;
  • actual points ahead of or behind that target: usedPercent - evenUsageTarget. Say the target in prose, rather than only reporting "N points over/under even pace."
• For each 5-hour window report used, remaining, and reset time. Say not reported when a pool has no 5-hour window.
• Treat backend resetsAt as authoritative. Do not describe a window as fixed or rolling unless current official documentation establishes that behavior.
• Percentages are backend-rounded integers; derived pace values are estimates. Use the machine's local timezone unless the user requests another.
• Mention available reset-credit count, but never redeem one without explicit authorization.
• If the sandbox blocks the local state runtime, request scoped permission to run this script. Do not work around it by extracting credentials.

Use python3 scripts/report_usage.py --json when another tool needs structured output.