Codex Skill Guide

Running a Task

1. If the user did not specify a model or reasoning effort, use the installed Codex default or ask once with the available user-question mechanism. Do not hardcode a model list; model names change over time.
2. Select the sandbox mode required for the task; default to --sandbox read-only unless edits or network access are necessary.
3. Run codex --version first. Stop and report the failure if Codex is unavailable.
4. Assemble the command with the appropriate options:
   • -m, --model <MODEL>
   • --config model_reasoning_effort="<xhigh|high|medium|low>"
   • --sandbox <read-only|workspace-write|danger-full-access>; only use other modes when codex exec --help lists them
   • -C, --cd <DIR>
   • --add-dir <DIR>
   • --skip-git-repo-check
   • --dangerously-bypass-approvals-and-sandbox
5. Do not use --skip-git-repo-check by default. Use it only when the user explicitly asks to run outside a Git repository or has approved that boundary bypass for this command.
6. When continuing a previous session, use codex exec resume --last via stdin. Do not add model, reasoning, or sandbox flags on resume unless the user explicitly requests an override.
7. IMPORTANT: By default, append 2>/dev/null to codex exec commands to suppress thinking tokens (stderr). Only show stderr if the user explicitly requests it or if debugging is needed.
8. Run the command, capture stdout/stderr (filtered as appropriate), and summarize the outcome for the user.
9. After Codex completes, inform the user: "You can resume this Codex session at any time by saying 'codex resume' or asking me to continue with additional analysis or changes."