If .claude/skills/codex-cli/SKILL.md.disabled exists, stop before reading a fallback.
Project configuration: If the current project contains .claude/skills/codex-cli/SKILL.md, read and apply it instead of this bundled default. The project copy is the capability source of truth.
Repository context: Read ../setup/references/task-context.md when resolving a task, project commands, named agent roles, or legacy .claude/ resource paths. Use repository evidence and applicable project instructions; a missing local workflow copy does not require setup. Resolve bundled resources from the installed skill, never from the target cwd.
Codex CLI
Announcement: Begin with: "I'm using the codex-cli skill for cross-AI validation with Codex."
Use this wrapper for one-shot Codex review, approach validation, or a delegated
second opinion. Read the shared lifecycle in
references/cross-ai-run.md before running it.
Provider adapter
- Check
command -v codex, then codex --version. Read codex --help and
codex exec --help when the installed version or requested scope is unclear.
- Use the locally supported non-interactive command and output option. Choose
the caller's model, or omit a model override for the configured default.
Never pass a hardcoded model from this skill.
- Prefer the provider's least-privilege mode for review. Do not use an
autonomous write or permission-bypass mode unless explicitly requested.
- Capture stdout/stderr according to the installed CLI's help, read the result,
and return the shared receipt. Historical
codex review versus codex exec review flag differences belong in reference.md; verify them locally before
use.
If codex is absent or needs an update, report the exact missing prerequisite
and stop. Installing or updating it is a separate user action; this skill never
runs npm i -g, curl | bash, or changes Codex settings.
Prompt requirements
The prompt is a cold-start handoff. Include the repository root, task path when
task-attached, changed files or diff scope, requirements, and the specific
questions to answer. Ask for findings or an approach only; the result does not
authorize repository edits.
For task-attached calls, resolve ../setup/references/task-context.md and save
the result receipt in that task's evidence area. Standalone calls return the
result without creating a task.
When not to use
- The user asked for the primary review or implementation workflow only
- The request needs an interactive conversation
- The question is trivial or can be answered from local evidence
References
references/cross-ai-run.md — shared preflight, context, capture, and evidence lifecycle
reference.md — exact flags only after checking the installed CLI; historical compatibility notes
templates.md — prompt shapes, with placeholders that must be resolved before execution
1---2name: codex-cli-43description: Run OpenAI Codex CLI for one-shot cross-AI code review or approach validation. Invoke ONLY when the user explicitly4 asks ('second opinion', 'codex review', 'ask codex', 'run codex', 'cross-AI check'), or when another skill passes an explicit5 instruction to delegate to codex. Do not invoke proactively on general review requests. Not for interactive conversations.6allowed-tools: Bash Read7compatibility: Portable clients may not enforce Claude Code invocation guards or project setup behavior.8metadata:9 source: https://skillmd.com/skills/alexandrbasis/codex-cli-410---1112<!-- claudops-build: project-config-pointer -->13> If `.claude/skills/codex-cli/SKILL.md.disabled` exists, stop before reading a fallback.14> **Project configuration:** If the current project contains `.claude/skills/codex-cli/SKILL.md`, read and apply it instead of this bundled default. The project copy is the capability source of truth.15> **Repository context:** Read `../setup/references/task-context.md` when resolving a task, project commands, named agent roles, or legacy `.claude/` resource paths. Use repository evidence and applicable project instructions; a missing local workflow copy does not require setup. Resolve bundled resources from the installed skill, never from the target cwd.1617# Codex CLI1819> **Announcement**: Begin with: "I'm using the **codex-cli** skill for cross-AI validation with Codex."2021Use this wrapper for one-shot Codex review, approach validation, or a delegated22second opinion. Read the shared lifecycle in23`references/cross-ai-run.md` before running it.2425## Provider adapter26271. Check `command -v codex`, then `codex --version`. Read `codex --help` and28 `codex exec --help` when the installed version or requested scope is unclear.292. Use the locally supported non-interactive command and output option. Choose30 the caller's model, or omit a model override for the configured default.31 Never pass a hardcoded model from this skill.323. Prefer the provider's least-privilege mode for review. Do not use an33 autonomous write or permission-bypass mode unless explicitly requested.344. Capture stdout/stderr according to the installed CLI's help, read the result,35 and return the shared receipt. Historical `codex review` versus `codex exec36 review` flag differences belong in `reference.md`; verify them locally before37 use.3839If `codex` is absent or needs an update, report the exact missing prerequisite40and stop. Installing or updating it is a separate user action; this skill never41runs `npm i -g`, `curl | bash`, or changes Codex settings.4243## Prompt requirements4445The prompt is a cold-start handoff. Include the repository root, task path when46task-attached, changed files or diff scope, requirements, and the specific47questions to answer. Ask for findings or an approach only; the result does not48authorize repository edits.4950For task-attached calls, resolve `../setup/references/task-context.md` and save51the result receipt in that task's evidence area. Standalone calls return the52result without creating a task.5354## When not to use5556- The user asked for the primary review or implementation workflow only57- The request needs an interactive conversation58- The question is trivial or can be answered from local evidence5960## References6162- `references/cross-ai-run.md` — shared preflight, context, capture, and evidence lifecycle63- `reference.md` — exact flags only after checking the installed CLI; historical compatibility notes64- `templates.md` — prompt shapes, with placeholders that must be resolved before execution