Codex Skill Guide
Workflow
- Confirm task mode:
- New run: use codex exec. - Continue prior run: use codex exec... resume --last with stdin prompt.
- Set defaults unless user overrides:
- Model: gpt-5.4. - Reasoning effort: ask user to choose xhigh, high, medium, or low. - Sandbox: read-only unless edits/network are required.
- Build command with required flags:
- Always include --skip-git-repo-check. - Add 2>/dev/null by default to suppress thinking tokens on stderr. - Show stderr only if user asks or debugging is needed.
- Run command, summarize outcome, and ask what to do next.
- After completion, remind user they can continue with
codex resume.
Quick Reference
| Use case | Sandbox mode | Key flags | |
|---|---|---|---|
| Read-only review or analysis | read-only | --sandbox read-only 2>/dev/null | |
| Apply local edits | workspace-write | --sandbox workspace-write --full-auto 2>/dev/null | |
| Permit network or broad access | danger-full-access | --sandbox danger-full-access --full-auto 2>/dev/null | |
| Resume recent session | Inherited from original | `echo "prompt" \ | codex exec --skip-git-repo-check resume --last 2>/dev/null` |
| Run from another directory | Match task needs | -C <DIR> plus other flags 2>/dev/null |
Command Patterns
New run
codex exec --skip-git-repo-check \
--model gpt-5.4 \
--config model_reasoning_effort="high" \
--sandbox read-only \
"your prompt here" 2>/dev/nullResume latest session
Use stdin and keep flags between exec and resume.
echo "your prompt here" | codex exec --skip-git-repo-check resume --last 2>/dev/nullWhen resuming, do not add configuration flags unless the user explicitly asks for changes (for example, different model or reasoning effort).
Model Options
| Model | Best for | Context window | Key features |
|---|---|---|---|
gpt-5.4 ⭐ | Default for most coding tasks in Codex | N/A in this skill | OpenAI's recommended default for general-purpose coding |
gpt-5.4-pro | Harder problems that benefit from more compute | N/A in this skill | More compute for deeper reasoning on difficult tasks |
gpt-5-mini | Faster/cost-effective option for lighter tasks | N/A in this skill | Smaller GPT-5 model for lower-cost coding and chat tasks |
gpt-5.3-codex | Legacy specialized alternative | N/A in this skill | Prior Codex-tuned model; generally superseded by GPT-5.4 |
gpt-5.4 is the default for software engineering tasks.
Reasoning Effort
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)
Following Up
- After every run, ask for next steps or clarifications.
- When proposing another run, restate model, reasoning effort, and sandbox mode.
- For continuation, use stdin with
resume --last.
Error Handling
- If
codex --versionorcodex execexits non-zero, report failure and ask before retrying. - Ask permission before high-impact flags unless already granted:
--full-auto,--sandbox danger-full-access,--skip-git-repo-check. - If output includes warnings or partial results, summarize and ask how to proceed.
CLI Version
Use a current Codex CLI version that supports gpt-5.4. Check with:
codex --versionUse /model inside Codex to switch models, or set defaults in ~/.codex/config.toml.
Update Check
On first use in a session, silently check for a newer version:
- Fetch
https://raw.githubusercontent.com/abpai/skills/main/versions.json. - Compare the version for
codexagainst this file'smetadata.version. - If the remote version is newer, pause before the main task and ask: codex update available (local {X.Y} → remote {A.B}). Would you like me to update it for you first? I can run
npx skills update codexfor you. - If the user says yes, run the update before continuing.
- If the user says no, continue with the current local version.
- If the fetch fails or web access is unavailable, skip silently.