codex-second-opinion
Get Codex to audit Claude's mistakes.
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
Get Codex to audit Claude's mistakes.
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。
基于 SOC 职业分类
This skill should be used when the user asks about libraries, frameworks, API references, or needs code examples. Activates for setup questions, code generation involving libraries, or mentions of specific frameworks like React, Vue, Next.js, Prisma, Supabase, etc.
Respond to code review comments point by point, quoting each reviewer point verbatim and answering directly beneath it.
Use when moving local sessions between Claude Code and Codex, converting Claude Code JSONL into Codex rollout format, converting Codex rollout JSONL into Claude scrollback, or debugging cross-agent resume state.
Use when moving local sessions between Codex and Claude Code, converting Codex rollout JSONL into Claude scrollback, converting Claude Code JSONL into Codex rollout format, or debugging cross-agent resume state.
Operate remote hosts over SSH through the ssh-mcp server, driving a box like a person at a terminal instead of firing blind one-off commands.
Use when a task involves SSH or a remote host, including remote commands, interactive shells, sudo over SSH, tailing logs, deployments, scp or rsync file transfer, and port forwards or tunnels.
| name | codex-second-opinion |
| description | Get Codex to audit Claude's mistakes. |
| when_to_use | Use when the user asks for a second opinion, a Codex audit, an independent review, or wants to check Claude's work. |
| argument-hint | [focus area or file path] |
| allowed-tools | ["Bash"] |
| effort | high |
Invokes Codex in read-only sandbox mode to hunt for bugs, dead code, tech debt, and architectural errors that Claude left behind.
Run: ${CLAUDE_SKILL_DIR}/codex-audit $ARGUMENTS
The script detects the project language, builds an audit prompt, and runs codex exec in read-only mode. If the argument is a file/directory path it scopes the audit there; if it is a topic (e.g. "error handling") it focuses on that category; if empty it does a broad sweep.
Print the full output as a direct text response so the user can read it without expanding tool results.
Do not add your own commentary or soften the results. Let the audit stand on its own.