Codex

Codex Skill Guide

Running a Task

1. Default to gpt-5.2 model. Ask the user (via AskUserQuestion) which reasoning effort to use (xhigh,high, medium, or low). User can override model if needed (see Model Options below).
2. Select the sandbox mode required for the task; default to --sandbox read-only unless edits or network access are necessary.
3. Assemble the command with the appropriate options:
   • -m, --model <MODEL>
   • --config model_reasoning_effort="<high|medium|low>"
   • --sandbox <read-only|workspace-write|danger-full-access>
   • --full-auto
   • -C, --cd <DIR>
   • --skip-git-repo-check
4. Always use --skip-git-repo-check.
5. When continuing a previous session, use codex exec --skip-git-repo-check resume --last via stdin. When resuming don't use any configuration flags unless explicitly requested by the user e.g. if he species the model or the reasoning effort when requesting to resume a session. Resume syntax: echo "your prompt here" | codex exec --skip-git-repo-check resume --last 2>/dev/null. All flags have to be inserted between exec and resume.
6. IMPORTANT: By default, append 2>/dev/null to all codex exec commands to suppress thinking tokens (stderr). Only show stderr if the user explicitly requests to see thinking tokens or if debugging is needed.
7. Run the command, capture stdout/stderr (filtered as appropriate), and summarize the outcome for the user.
8. 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."

Safety Boundaries

• Do not run codex exec until the user has explicitly asked to use Codex for the task.
• Do not select --full-auto or a broader sandbox mode than the task requires.
• Do not overwrite user files, create commits, or perform destructive operations without explicit user approval.
• Do not use networked or danger-full-access runs unless the user requested work that actually needs them.

Quick Reference

Model Options

GPT-5.2 Advantages: 76.3% SWE-bench (vs 72.8% GPT-5), 30% faster on average tasks, better tool handling, reduced hallucinations, improved code quality. Knowledge cutoff: September 30, 2024.

Reasoning Effort Levels:

• xhigh - Ultra-complex tasks (deep problem analysis, complex reasoning, deep understanding of the problem)
• high - Complex tasks (refactoring, architecture, security analysis, performance optimization)
• medium - Standard tasks (refactoring, code organization, feature additions, bug fixes)
• low - Simple tasks (quick fixes, simple changes, code formatting, documentation)

Cached Input Discount: Cached context is substantially cheaper than first-pass context. Check current vendor pricing before quoting costs.

Following Up

• After every codex command, immediately use AskUserQuestion to confirm next steps, collect clarifications, or decide whether to resume with codex exec resume --last.
• When resuming, pipe the new prompt via stdin: echo "new prompt" | codex exec resume --last 2>/dev/null. The resumed session automatically uses the same model, reasoning effort, and sandbox mode from the original session.
• Restate the chosen model, reasoning effort, and sandbox mode when proposing follow-up actions.

Error Handling

• Stop and report failures whenever codex --version or a codex exec command exits non-zero; request direction before retrying.
• Before you use high-impact flags (--full-auto, --sandbox danger-full-access, --skip-git-repo-check) ask the user for permission using AskUserQuestion unless it was already given.
• When output includes warnings or partial results, summarize them and ask how to adjust using AskUserQuestion.

CLI Version

Requires Codex CLI v0.57.0 or later for GPT-5.2 model support. The CLI defaults to gpt-5.2 on macOS/Linux and gpt-5.2 on Windows. Check version: codex --version

Use /model slash command within a Codex session to switch models, or configure default in ~/.codex/config.toml.