REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
A portable project-planning skill for Codex, Claude Code, pi, Hermes, and Agent Skills-compatible harnesses. Evidence before build advice.
作者提供的原文
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/aaravkashyap12-advise-project-approach)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 advise-project-approach | rea | andrej-karpathy-skills | graphify | MoneyPrinterTurbo | |
|---|---|---|---|---|---|
| 作者 | AaravKashyap12 | morluto | multica-ai | Graphify-Labs | harry0703 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 330 | 468,733 | 234,332 | 150,287 | 144,566 |
| 最近代码更新 | 2026年8月31日 | 2026年11月19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 |
AaravKashyap12 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。
Claude CodeCodex
依据开发者 README 和 repo 中的文件夹
GitHub 账号 AaravKashyap12(个人),已创建 3 年
★ 330 星,目前用户较少 · 代码最近更新 2026年8月31日
没有,都是纯文本文件,由 AI 阅读后照做
没有脚本,所以不会自己联网
没有发现 hook 或写入电脑设置的操作
根据仓库 commit abdde26 中的文件自动检查(只读取文件,没有运行任何内容)。这不是安全保证,请自己阅读后再决定
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
无法上传到 claude.ai,因为描述超过 200 个字符(634 个字符)。请改用"让 AI 帮你安装"
复制这句话,粘贴到 Claude Code 的对话中
如果只用于一个项目,请粘贴到在该项目文件夹中打开的对话里 · AI 在安装前一定会先告诉你路径
AI 会先说明要做什么,每一步都会征求你的同意
重新打开 Claude Code,输入 /advise-project-approach,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
AI 在安装前一定会先告诉你路径
AI 会先说明要做什么,每一步都会征求你的同意
把 advise-project-approach 文件夹放到 ~/.agents/skills/advise-project-approach
~ 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
如果只用于一个项目,请粘贴到在该项目文件夹中打开的对话里 · AI 在安装前一定会先告诉你路径
AI 会先说明要做什么,每一步都会征求你的同意
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
如果只用于一个项目,请粘贴到在该项目文件夹中打开的对话里 · AI 在安装前一定会先告诉你路径
AI 会先说明要做什么,每一步都会征求你的同意
在 Cursor 的 Agent 聊天中输入 / 并选择 advise-project-approach
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 · 复制这句话,粘贴到应用的对话中
如果只用于一个项目,请粘贴到在该项目文件夹中打开的对话里 · AI 在安装前一定会先告诉你路径
AI 会先说明要做什么,每一步都会征求你的同意
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
显示前 8,000 个字符 · 在 GitHub 打开完整文件
---
name: advise-project-approach
description: Research and advise on the best way to approach a software project, including architecture, tech stack, implementation strategy, pricing/operating-cost tradeoffs, benchmark research, and comparisons with similar real-world projects. Use before building, mid-build, or after completion when the user asks for project strategy, optimal approach, research comparables, similar projects, stack selection, vendor/service choice, repo analysis, architecture critique, implementation feedback, or a prioritized improvement plan. Avoid for narrow single-bug debugging or isolated file edits unless the user asks for broader project direction.
---
# Advise Project Approach
Help the user decide, validate, or improve how a project should be built. This skill automates the research loop a strong engineer would normally do manually: understand the project goal, inspect any existing work, study credible comparables, evaluate tech-stack and architecture choices, then recommend the highest-leverage path.
## Non-Negotiable Protocol
Apply these gates before all other instructions:
1. **Stop for vague intake.** For an initial broad pre-build request, if two or more facts material to the requested decision are unknown, ask the concise intake batch and end the response. Do not invent a product direction or constraints. Skip repetitive intake for bounded questions or an already answered interview; handle accepted unknowns as described in Project Intake. If the user asks to skip questions, proceed with visible assumptions.
2. **Keep repository review read-only.** A request to inspect or review a repository does not authorize dependency installation or execution of its tests, builds, linters, audits, benchmarks, scripts, or application code. Ask before running them.
3. **Do not outsource judgment to popularity.** Never select or copy a stack because a repository has the most stars or adoption. If the user requests that shortcut, explain why it is not a fit test and continue only with visible assumptions or known constraints.
4. **Require receipts before recommendation.** For a substantive recommendation, inspect relevant local evidence and normally two comparables plus primary documentation or pricing sources when available. State what was inspected, what each source supports, its limits, and the observed date for time-sensitive claims.
5. **Complete the decision.** A substantive choice needs constraint fit, a credible alternative, tradeoffs, reversal conditions, and next actions. Mark unavailable evidence explicitly. For narrow implementation guidance, do not reopen the user's settled choices just to fill these fields.
6. **Stop when evidence is sufficient or exhausted.** Track whether each lookup adds new support. Stop after two consecutive lookups add no decision-relevant evidence, even if the question remains unresolved. Use the bounded research and repository-inspection rules below.
7. **Make advice disprovable.** Pair the first action with observable acceptance and a check that actually exercises the failure mode. Proposed checks are neither passing results nor architecture diagnoses. A failing invariant triggers debugging of the write path, test, or requirement first, not an automatic new layer.
8. **Honor the requested scope and format.** Respect explicit length, paragraph, and step limits in every mode; combine essential evidence and advice rather than filling template headings. When the direction is fixed and the question is bounded, use the narrow-advice route. Gather fresh evidence only if it could materially change the answer; that does not expand the requested output.
9. **Sequence observable outcomes.** Prefer end-to-end capabilities over speculative abstractions. A required safety, recovery, or correctness check is a valid earlier step when the user or evidence makes it a prerequisite. Avoid layer extraction solely for anticipated reuse.
## Operating Modes
First identify which mode applies:
- **Pre-build strategy** - no repo exists yet, or the user is deciding how to build. Focus on requirements, constraints, comparable projects, stack choices, architecture options, risks, and a recommended implementation path.
- **Mid-build course correction** - a repo or partial implementation exists. Inspect the code, compare it with the intended goal and external references, then recommend what to keep, change, or defer.
- **Post-build review** - the project is mostly complete. Review architecture, quality, maintainability, deployment readiness, security posture, and gaps against similar mature projects.
Mode selection rule:
- Use the user's explicit stage for the subject project first. A comparable, dependency, or upstream-template URL is reference material, not evidence that the user's own project exists or is finished.
- If the subject has no implementation yet, use **pre-build strategy**. If work is underway, use **mid-build course correction**. If the subject is finished, deployed, or being assessed for launch, use **post-build review**.
- Use repository and language clues only when the subject's stage is unclear. Distinguish current state from an intended future launch or scale.
If a mid-build or post-build request provides only a description and no repo/code, proceed as an **advisory review from description**. Say that file-level findings require a repo or code sample; do not pretend local evidence was inspected.
### Narrow-advice route
Use this route when the user has fixed the main direction and asks a bounded question such as "give me the first three steps," "how should I validate this choice," or "compare these two options." It overrides the full workflow and output contracts for that response.
- Preserve the requested count and format. For `N` steps, return exactly `N` primary steps.
- Put the action, acceptance behavior, and focused check inside each step.
- Keep checks concrete: for a race, name independently competing operations and how overlap is synchronized. Label unsynchronized parallel requests as a stress probe, not deterministic proof.
- State assumptions or evidence limits in one compact sentence only when material.
- End with at most one compact failure or escalation condition when it changes the advice.
- Express that escalation as a testable threshold or event, such as the same invariant duplicated across two entry points, a focused check failing, or a measured performance/cost limit being crossed. Do not use "awkward," "complex," or "hard to maintain" without an observable proxy.
- Do not browse, research comparables, restate the chosen stack, or emit full-report headings unless fresh evidence is necessary to answer the bounded question safely.
- Prefer the framework's default organization. Do not introduce a service layer, repository pattern, queue, cache, microservice, or other architectural boundary without evidence that the current requirement needs it.
- Give each numbered step an observable capability or risk-reducing result, including prerequisite safety checks when necessary. Keep early slices in the framework's ordinary structure; extract abstractions only for evidenced duplication or conflicting entry points, not merely because a test failed.
## Project Intake
Use a lightweight intake interview before research when a pre-build request is vague enough that different answers would materially change the recommendation.
Decision-critical facts include the primary user, core workflow, project stage, must-haves, builder/team capability, budget or deadline, deployment target, and dominant priority.
Do not interrogate users who already supplied clear constraints. If the project, users, must-have workflow, stage, and major constraints are sufficiently specified, begin research immediately and ask only the missing decision-critical question.
For a vague idea, ask these questions in one concise batch and accept "not sure" answers:
1. What are you trying to build, and who is it for?
2. 提问、分享技巧或报告无法使用。按“有帮助”数量排序。