codex-review
Run OpenAI Codex CLI reviews from Claude Code. Use when the user asks for a Codex review or mentions triggers like "codex-review", "/codex-review", or "codex review".
安装与使用
请根据来源仓库和你的 AI 工具说明安装此技能。先阅读技能文件和依赖要求,再添加到工作流。
查看来源 ↗使用指南技能说明
Codex Review
Overview
Run codex review and iterate on fixes until the review is clean. Claude Code applies fixes locally and re-runs the review.
Workflow
- Confirm the review target.
- If the user specifies a target, follow it.
- Otherwise, prefer
--uncommittedwhen there are local changes. - If there are no local changes, use
--base origin/main(or the repo default).
- Prepare the temp environment under
~/.claude/tmp(see "Temp directory setup"). Ensure these env vars are set before runningcodex review.- Claude Code runs each shell command in a fresh process, so
exportdoes not persist across commands. - Always run the temp setup and
codex reviewin the same shell invocation (see "Command templates").
- Claude Code runs each shell command in a fresh process, so
- If the user wants to override the review model, append
-c review_model="MODEL"to the command (default stays as-is when omitted). - If the user wants to override reasoning effort, append
-c model_reasoning_effort="EFFORT"to the command (default stays as-is when omitted). Use a value supported by the review model. - Run
codex reviewwith the chosen target and any custom prompt (plus the optional-c review_model="MODEL"from step 3 and-c model_reasoning_effort="EFFORT"from step 4). - Read the review output and extract actionable findings.
- If findings exist, fix them in the codebase and re-run the same review target.
- Repeat until no findings remain or after 10 loops, then report status.
Temp directory setup
Use a per-run temp directory to avoid /tmp failures in sandboxed environments. Set these env vars before running codex review.
Important: run the temp setup and codex review in the same shell invocation, or the exports will be lost.
mkdir -p ~/.claude/tmp
TMPDIR="$(mktemp -d ~/.claude/tmp/codex-review.XXXXXXXX)"
ZDOTDIR="$TMPDIR/zsh"
mkdir -p "$ZDOTDIR"
XCRUN_CACHE_PATH="$TMPDIR/xcrun_db"
DARWIN_USER_TEMP_DIR="$TMPDIR"
export TMPDIR ZDOTDIR XCRUN_CACHE_PATH DARWIN_USER_TEMP_DIR
Only ~/.claude/tmp is allowed for temp usage. Do not create temp directories inside the repo or under /tmp.
Timeout
Important: codex review can take a long time (several minutes). Always set a 30-minute timeout (1800000 ms) when running the command to ensure it completes.
Command templates
Always include the temp directory setup before running codex review and do not split these into multiple shell commands:
mkdir -p ~/.claude/tmp && \
TMPDIR="$(mktemp -d ~/.claude/tmp/codex-review.XXXXXXXX)" && \
ZDOTDIR="$TMPDIR/zsh" && \
mkdir -p "$ZDOTDIR" && \
XCRUN_CACHE_PATH="$TMPDIR/xcrun_db" && \
DARWIN_USER_TEMP_DIR="$TMPDIR" && \
export TMPDIR ZDOTDIR XCRUN_CACHE_PATH DARWIN_USER_TEMP_DIR && \
codex review --uncommitted [-c review_model="MODEL"] [-c model_reasoning_effort="EFFORT"]
mkdir -p ~/.claude/tmp && \
TMPDIR="$(mktemp -d ~/.claude/tmp/codex-review.XXXXXXXX)" && \
ZDOTDIR="$TMPDIR/zsh" && \
mkdir -p "$ZDOTDIR" && \
XCRUN_CACHE_PATH="$TMPDIR/xcrun_db" && \
DARWIN_USER_TEMP_DIR="$TMPDIR" && \
export TMPDIR ZDOTDIR XCRUN_CACHE_PATH DARWIN_USER_TEMP_DIR && \
codex review --base <branch> [-c review_model="MODEL"] [-c model_reasoning_effort="EFFORT"]
mkdir -p ~/.claude/tmp && \
TMPDIR="$(mktemp -d ~/.claude/tmp/codex-review.XXXXXXXX)" && \
ZDOTDIR="$TMPDIR/zsh" && \
mkdir -p "$ZDOTDIR" && \
XCRUN_CACHE_PATH="$TMPDIR/xcrun_db" && \
DARWIN_USER_TEMP_DIR="$TMPDIR" && \
export TMPDIR ZDOTDIR XCRUN_CACHE_PATH DARWIN_USER_TEMP_DIR && \
codex review --commit <sha> [-c review_model="MODEL"] [-c model_reasoning_effort="EFFORT"]
Output handling
- Summarize the review results in a short list.
- If fixes were applied, list the changed files and confirm the re-review is clean.
- If issues remain after the max loops, call out the remaining findings.
使用示例
查看来源仓库中的使用示例与配置说明。
查看来源 ↗来源历史
- 收录时间
- 2026-01-20
- 索引更新时间
- 2026-01-26
- Branch
- main
收录与同步日期属于本站记录。源码修改历史请查看来源仓库。
GitHub ↗chat Comments (0)
Sign in to join the discussion and leave a comment.
No comments yet. Be the first to share your thoughts!