REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
PavedPath Code: reusable Skill for finding proven implementation paths from GitHub and open-source evidence
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/jia-ethan-pavedpath-code)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 pavedpath-code | rea | graphify | archify | security-audit-skill | |
|---|---|---|---|---|---|
| 作者 | Jia-Ethan | morluto | Graphify-Labs | tt-a1i | cloudflare |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 439 | 463,254 | 148,775 | 97,049 | 73,584 |
| 最近代码更新 | 2026年6月30日 | 2026年11月16日 | 2026年11月16日 | 2026年11月15日 | 2026年9月15日 |
| 许可证 | MIT | MIT | Apache-2.0 | Apache-2.0 | MIT |
Jia-Ethan 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 CodeCodexCursor
依据开发者 README 和 repo 中的文件夹
GitHub account Jia-Ethan (person) is only 7 months old: take extra care
★ 439 stars, few users so far · code last changed 2026年6月30日
No, plain text files the AI reads and follows
No scripts, so it does not go online by itself
No hooks or writes to machine settings found
Checked automatically from the repo files at commit 97a319f (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 (265). 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 /pavedpath-code,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。
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 pavedpath-code 文件夹放到 ~/.agents/skills/pavedpath-code
~ 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 pavedpath-code
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
显示前 8,000 个字符 · 在 GitHub 打开完整文件
---
name: pavedpath-code
description: PavedPath Code is the code-focused edition of PavedPath. It helps agents solve software engineering problems by finding proven implementation paths from GitHub repositories, issues, pull requests, discussions, code examples, release notes, and open-source evidence.
---
# PavedPath Code
PavedPath Code(代码版)helps agents avoid reinventing solutions for software engineering problems. It searches GitHub and open-source evidence for proven implementation paths, then adapts the strongest pattern to the local codebase with minimal, verifiable changes.
Use GitHub as a primary evidence source, not as a link dump. The goal is to define the local code problem, find paths already walked in public repositories, issues, pull requests, discussions, examples, release notes, and source code, evaluate the strength of that evidence, then translate the strongest pattern into a small local fix, implementation path, or verification plan.
This is the code-focused edition of PavedPath. Do not broaden it into the future general-purpose PavedPath. It is for software engineering work: bugs, runtime errors, build/test/deploy failures, dependency issues, framework or API usage, integration blockers, implementation patterns, engineering tool selection, and open-source solution adaptation.
## When to Use
- Runtime, build, test, deploy, package, SDK, API, dependency, framework, or integration errors.
- A feature implementation is blocked by an unclear edge case, missing usage pattern, version behavior, or uncertain API contract.
- A local issue resembles something that maintainers or other open-source users may have resolved in issues, PRs, examples, code, release notes, or discussions.
- The user asks whether GitHub or open-source projects have already solved a concrete engineering problem.
- Mature implementation examples or reusable projects would reduce uncertainty for one specific capability.
- The answer should compare suitable GitHub repositories and explain how to adapt one proven path locally.
Do not use PavedPath Code for copy edits, purely local refactors where the codebase already dictates the answer, non-code life decisions, self-media workflows, study methods, consumer purchasing decisions, or requests that explicitly forbid web/GitHub research. Do not inspect private repositories unless the user explicitly scopes and authorizes that access.
## Default Workflow
1. **Frame the local code problem first.** Capture the goal, symptom, error signature, reproduction path, versions, runtime, dependency/framework names, recent changes, constraints, and attempted fixes. If a discoverable fact is missing, inspect local files/logs before asking.
2. **Choose the evidence mode.** For errors/regressions, search issues, PRs, releases, and code first. For capability or tool needs, search repository candidates first. For feature implementation, use both repository candidates and issue/PR/code evidence.
3. **Evaluate subagent usefulness.** Before substantial GitHub research, decide whether conditional subagent work would improve breadth, evidence quality, or review coverage. If not using subagents, state the reason briefly when reporting the search path.
4. **Create targeted searches.** Prefer exact error text, package/API names, version numbers, framework + symptom, file names, config keys, stack trace fragments, failing command names, or capability + framework/runtime/API names.
5. **Find suitable GitHub projects when relevant.** Prefer high-fit, high-Star, active, non-archived repositories with clear licenses and real examples. Lower the Star threshold when the high-Star set is too broad or misses the exact problem.
6. **Search open-source evidence surfaces.** Use issues, PRs, discussions, code, examples, release notes, and official project docs within relevant open-source repos. Repository search is required when a project itself may solve the problem.
7. **Rank by problem fit first, with Stars as a maturity signal.** A high-Star repository is a strong candidate for inspection, but maintainer-confirmed issues, merged PRs, released fixes, official examples, and exact matching code beat popular adjacent projects. Use [research-rubric.md](references/research-rubric.md) when ranking matters.
8. **Deep-read the strongest projects and evidence.** Use [extraction-playbook.md](references/extraction-playbook.md) to extract project basics, reusable surfaces, root cause or implementation pattern, version constraints, risks, adaptation boundaries, and verification steps.
9. **Translate to local work with minimal adaptation.** Preserve the proven workflow, API, configuration shape, or architecture where it fits. Adapt only the parts required by the user's local interfaces, configuration, data/auth model, deployment target, or language/runtime.
10. **Verify before claiming success.** Use tests, builds, reproduction commands, real requests, logs, browser checks, or manual inspection appropriate to the local problem.
11. **If evidence is weak, say so.** Do not stretch weak matches into a confident recommendation. Mark the recommendation as first-principles or local-only when open-source evidence is insufficient.
## Subagent / Parallel Research Guidance
Subagents are conditional research aids, not a default requirement. The controller remains responsible for problem framing, scope control, evidence ranking, local adaptation, and final verification.
Use subagents when at least one of these applies:
- The problem spans 2+ independent ecosystems, frameworks, languages, tools, deployment surfaces, or GitHub communities.
- Repository discovery needs broad candidate coverage across multiple query families.
- Issue, PR, discussion, code, release, and example evidence can be split cleanly by project, version, or search surface.
- A final recommendation benefits from independent evidence review, risk review, or candidate rejection review.
Do not use subagents when any of these applies:
- The task is a narrow error with one obvious package, repository, API, or maintainer surface.
- Local repository context, logs, config, or reproduction details must be understood before external research can be scoped safely.
- GitHub rate limits, authorization boundaries, private repositories, secrets, production data, or sensitive logs would make delegation risky.
- Subagents would mostly duplicate the same searches or edit the same local files.
When using subagents, the controller must:
- Define each subagent's query family, repository scope, evidence surface, constraints, allowed write scope, and expected output.
- Require direct links, verified metadata, problem-fit rationale, risk notes, and explicit rejection reasons.
- Keep subagents read-only unless a separate implementation phase has a narrow allowed write scope.
- Merge and deduplicate results before ranking; do not count repeated reports of the same issue, PR, code path, or repository as independent evidence.
- Directly verify the strongest claims with `gh`, source reads, tests, logs, real requests, or official docs before finalizing.
## GitHub CLI First
Use the GitHub CLI (`gh`) as the default search and inspection surface. Prefer `gh search repos`, `gh search issues`, `gh search prs`, `gh search code`, `gh repo view`, `gh issue view`, `gh pr view`, and `gh api` before browser scraping or custom scripts. Do not add or rely on bundled search scripts for this skill.
Use these short command templates as starting points, then adjust the query, repo, fields, and limits to the local problem:
```bash
gh search repos "<query>" --archived=false --sort stars --order desc --limit 10 --json fullName,url,description,stargazersCount,forksCount,language,license,pushedAt,isArchived,openIssuesCount
gh search issues "<query>" --repo owner/repo --sort updated --order desc --limit 10 --json title,url,state,updatedAt,commentsCount,repository,body
gh search prs "<query>" --repo owner/repo --merged --sort updated --order desc -提问、分享技巧或报告无法使用。按“有帮助”数量排序。