REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
协调探索、执行和独立评审,让复杂任务按明确职责并行推进,由主 Agent 统一验收。
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/oil-oil-codex-team-mode)在同一工作中把 AI 连到服务的 MCP
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
Codebase intelligence for TypeScript and JavaScript, exposed as structured MCP analysis tools.
MCP server for terminal commands, file operations, and process management
同类工作,排序与 Skill 列表相同
| 本项 codex-team-mode | rea | andrej-karpathy-skills | graphify | MoneyPrinterTurbo | |
|---|---|---|---|---|---|
| 作者 | oil-oil | morluto | multica-ai | Graphify-Labs | harry0703 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 237 | 467,235 | 234,125 | 149,873 | 144,332 |
| 最近代码更新 | 2026年10月1日 | 2026年11月18日 | 2026年4月20日 | 2026年11月18日 | 2026年11月18日 |
| 许可证 | MIT | MIT | – | Apache-2.0 | MIT |
oil-oil 提供的 1 个 Skill(指令集),AI 读取后照做 · 不是 MCP server
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
Codex
依据开发者 README 和 repo 中的文件夹
GitHub account oil-oil (person), 7 years old
★ 237 stars, few users so far · code last changed 2026年10月1日
2 scripts that can run on a computer: scripts/current_model.py, scripts/usage_by_model.py
No internet calls found in the scripts
Scripts that touch settings folders (~/.claude or settings.json): scripts/current_model.py, scripts/usage_by_model.py
Checked automatically from the repo files at commit 8eab25b (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
Uploading to claude.ai would fail: the description is over 200 characters (411). Use "Let the AI install it" instead
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /team-mode,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -vMac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号: node -v
安装一次,所有需要 Node.js 的工具都能用
正在检查命令 · 通过后才会显示复制按钮
摘自开发者 README,会下载并运行开发者的代码
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
ChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
把 team-mode 文件夹放到 ~/.agents/skills/team-mode
~ 是你的主文件夹($HOME)
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 team-mode
Checked against official docs 2026年11月12日 · Source: cursor.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
5,250 个字符 · 在 GitHub 打开完整文件
---
name: team-mode
description: "Use when a task may benefit from a bounded subagent for implementation, large-codebase discovery at task start, independent review, requested or pre-commit code simplification, or expert work on complex decisions, modeling, automation, or repeated failures. The main agent chooses whether to delegate and accepts the result. Do not use for simple questions or short work with no meaningful review or delegation."
metadata:
compatibility: "Codex with custom subagents; local model and usage diagnostics require Python 3.10+ and retained session logs."
---
# Team Mode
The main agent decomposes the user's task, decides what to delegate, and accepts the result. A task's size alone does not require subagents. On activation, send one brief commentary update in the user's language prefixed with `👾` (in Chinese: `👾 已开启小队模式。`).
## When to dispatch
Delegate a defined part of the task when a child can make useful progress through implementation, codebase discovery, review, or expert work. The main agent owns how the parts fit together, unresolved product decisions, and final acceptance. There is no target number of agents or required sequence.
Choose the smallest count that covers the independent work. One child is the default for one bounded question or review lens. Add children only when each can work and report independently, with a distinct question, scope, or lens. A large diff alone is not a reason to fan out.
Consider coordination and usage overhead. Respect the host's active-agent limit; if parallel dispatch is unavailable, continue sequentially or inline without silently dropping coverage.
- `Explorer` — use for substantial codebase discovery at the start of a task. Handle small lookups and general research directly. Read [Explore](references/explore.md) for this route.
- `Executor` — use when the intended result and file ownership are clear enough for independent implementation. Give each target one owner.
- `Reviewer` — use when the user asks for review or a completed result has a meaningful risk of unnoticed defects. Review the assigned result without steering toward a suspected answer; save a Markdown report when a lasting record helps.
- `ExpertAdvisor` — use for a complex architecture or high-impact decision, a problem unresolved after repeated attempts, or modeling or complex computer automation the main agent cannot handle well. Ask for an independent plan or assign the expert a concrete outcome to produce.
When the user asks to simplify code, or before committing a code change, follow [Simplify](references/simplify.md) for scoped cleanup and final review. The main agent decides which findings to apply and accepts the final result.
For each dispatch, name the intended `agent_type`, count, independent scope, expected return, and write ownership. Keep write scopes separate when agents share a workspace.
Other configured roles and models are allowed when appropriate. Model and effort defaults, plus the optional `default.toml` sentinel, are described in [profile setup](references/custom-agents.md); normal dispatch does not require reading it.
## Context and handoff
Start a new child with `fork_turns="none"` by default. Always use it for `Reviewer` and `ExpertAdvisor`, whose value depends on an independent view. Use it for `Explorer` too: a question and codebase path are usually enough.
An `Executor` may inherit a small number of recent parent turns when the assignment depends on decisions in that conversation that a short brief would likely omit. Use the smallest useful positive `fork_turns` value; inherit the full conversation only when the whole history is genuinely needed. Reuse an existing child for a direct continuation of its own assignment.
Give the child its assigned question or deliverable, relevant context, and scope. Include the broader user goal when it helps explain the assignment. Pass files, symptoms, and prior attempts as leads, not a fixed diagnosis or checklist. The child chooses its method within the assigned part; the main agent handles changes to the task breakdown and checks the returned work.
## Expert help
When stronger expertise would help, use `python3 scripts/current_model.py` if the parent model is unclear. Choose an available stronger model and supported effort for the model-free `ExpertAdvisor` profile.
Give it the decision or outcome, relevant context, evidence, constraints, and scope without presenting a proposed cause as fact. The expert may produce a plan or carry out assigned modeling or complex automation. Check its result before accepting it. If no stronger model is available, continue without claiming a stronger-model consultation.
A model fixed in a custom profile takes precedence over a spawn-time choice. See [profile setup](references/custom-agents.md) for model changes; do not assume an explicit model overrides `Explorer`, `Executor`, or `Reviewer`.
## Diagnostics
For a requested usage report, run `python3 scripts/usage_by_model.py`; read [evaluation guidance](references/evaluation.md) when evaluating Team Mode itself. For an interactive-state task that code inspection cannot establish, read [interactive testing guidance](references/interactive-testing.md).
提问、分享技巧或报告无法使用。按“有帮助”数量排序。