dsh-codex-usage

If you live in DeepSeek Harness all day while still watching your OpenAI Codex quota, the worst part is switching to a terminal, typing a codex query, or opening another tab just to check what is left. dsh-codex-usage puts Codex usage exactly where you already look: a 28 px ring at the far right of the Settings row, matched in geometry and design tokens to the conversation context meter, with the OpenAI mark inside. Hover over it or give it keyboard focus and a full quota panel unfolds, showing rolling-window percentages, reset times, plan type, remaining credits, spend controls, and reset credits at a glance. The data flows from a locally managed codex app-server process; OAuth and credential refresh stay inside that official process, while the browser receives only a strict, normalized summary over a loopback-only GET route, with no configurable proxy, no direct auth.json reading, and no token handling on the client side.

It is built for engineers who use DeepSeek Harness as their daily workbench while holding a ChatGPT Plus, Pro, or enterprise Codex allocation. A five-minute background poll keeps the numbers fresh by default, a 250 ms hover debounce triggers an on-demand refresh, and a last-good snapshot means a brief network hiccup or a temporary provider outage never erases the data you were reading. Cross-platform command launching covers Windows (including PowerShell shims), macOS, and Linux.