| name | codex-doctor |
| description | Diagnose current Codex quota windows, account token activity, context pressure, session age, model and reasoning usage, burn pace, and evidence-based next actions. Use when a user asks how much Codex usage remains, whether usage is high or may run out, when a limit resets, whether to start a new session, or whether current model/reasoning choices fit the remaining quota. Do not use for generic Codex installation, configuration, or authentication troubleshooting handled by the built-in `codex doctor` command. |
Codex Doctor
Run the deterministic reporter immediately. Never call update_plan or create a plan for this skill. The reporter owns collection, calculations, diagnosis, recommendations, partial-data handling, and terminal formatting.
- Run
node <skill-dir>/scripts/codex-doctor.mjs from the user's current working directory, where <skill-dir> contains this file.
- Make complete stdout the first user-visible response, verbatim, in one fenced
text block. Preserve every character, line, meter, and space.
Do not send a preamble, acknowledgement, status update, progress message, or execution announcement. Do not inspect, research, or explain anything before running. For a bare $codex-doctor invocation, stop after stdout. If the prompt includes a question, answer briefly after the report using its evidence; investigate further only when needed for that question.
Do not contradict Diagnosis or Suggested action, invent unavailable windows, equate token activity with quota consumption, or infer that descriptive model/reasoning fields caused usage.
For an explicit machine-readable request, add --json and return stdout verbatim in one fenced json block. On failure, return the exact error without inventing a fallback.
Use only this reporter; never scrape the TUI, credentials, logs, or transcripts.
