一键导入
codex-review
Run a code review using Codex CLI. Use when the user wants a code review of uncommitted changes, a specific commit, or changes against a base branch.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Run a code review using Codex CLI. Use when the user wants a code review of uncommitted changes, a specific commit, or changes against a base branch.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Guides execute-inspect-adjust development and t-wada Red-Green-Refactor TDD. Use for exploratory implementation or when stable behaviour needs an executable test-first contract.
Prevents and handles GitHub API rate limits with gh-nix. Use when Nix, flakes, nixpkgs commands, or comma may fetch GitHub-backed inputs.
Manages worktree lifecycle operations through git-wt without replacing an existing linked worktree. Use when creating, switching, listing, renaming, or deleting worktrees.
Runs the full PR workflow — creates a feature branch, commits, pushes, and opens the pull request. Use when the user asks to create or open a PR ("create a PR", "push this up and open a PR").
Resolves missing CLI tools. Use when a command is unavailable, a shell reports command not found, or a tool must be run without installing it globally.
Guides agent-skill creation and updates following Anthropic's SKILL.md best practices. Use when adding or editing skills under `agents/skills/`, writing SKILL.md frontmatter, references, or skill routing.
| name | codex-review |
| description | Run a code review using Codex CLI. Use when the user wants a code review of uncommitted changes, a specific commit, or changes against a base branch. |
You are a code review coordinator. When invoked, run a code review using Codex.
This skill is shared between agents (Claude Code, Codex, etc.). Pick the path that matches the session you are running in:
codex exec review gives you an independent second opinion from a different agent.codex exec review from inside Codex starts a whole new Codex process (cold start, fresh model load, duplicated context) just to call the same engine you are already running. The overhead is large and pointless — delegate to a subagent instead.Use codex exec review with the appropriate flags.
codex exec review --uncommitted
codex exec review --base main --model <model-slug>
codex exec review --commit <SHA> --model <model-slug>
[PROMPT] is a positional argument that cannot be combined with --uncommitted, --base, or --commit. Use it alone for a free-form review prompt:
codex exec review "Focus on error handling and edge cases"
Do not run codex exec review here. Instead, launch a subagent (Codex multi-agent is enabled) and have it perform the review in-process:
git diff HEAD plus untracked files.git diff main...HEAD.git show <SHA>.--uncommitted (or the equivalent git diff HEAD scope for the subagent path) when the user says "review my changes" without further detail--base main (or git diff main...HEAD) when reviewing a feature branch's full diff!codex exec review --help
!jq -r '.models[] | "- \(.slug): \(.description)"' "$CODEX_HOME/models_cache.json"
Pick a --model slug from the list above. Prefer Spark models for speed. Use the latest non-Spark model for deeper analysis.