Token Meter is a small Codex skill that reports token usage using Codex’s own goal telemetry instead of guessing from transcript length or local files. It was built mainly for developers using Codex through a ChatGPT Plus subscription who want a clearer sense of usage than the generic percentage indicators shown by OpenAI. When invoked, it starts a bounded telemetry goal for the current task, lets Codex do the work, then reports the platform-provided token count when the task is complete.
The skill is intentionally conservative: it only reports values returned by Codex runtime tools such as get_goal, create_goal, and update_goal. If telemetry is unavailable, it says so rather than estimating. It also avoids polling and avoids reading local logs, cache files, credentials, or transcript stores to infer usage.
The main limitation is that Codex currently exposes reliable token counts through the Goal workflow. Persistent tracking across multiple messages requires leaving a Codex goal active, and Codex Desktop may show repeated timestamped goal/status entries while that goal is running. For that reason, Token Meter now defaults to clean bounded per-task reporting, with persistent per-message tracking treated as an opt-in mode with a visible UI tradeoff.
Interested in any feedback.
If interested in using it, you can find it in my github Angry-TacoZ/token-meter