Replica:11 个 Skill,把任何 App 克隆一遍
Replica: 11 Skills to Clone Any App from Scratch
如果你看了一个 App,觉得「这个我也能做,而且能做得更好」——replica-skill 就是把这个想法走完的 11 步流程。
MIT,1.3K stars,2026-10-03 发布,一周内涨到这个数。
GitHub: https://github.com/Jakeschincariol/replica-skill
11 个 Skill 的分工
| Skill | 做什么 |
|---|---|
/replica-recon | 逆向分析目标 App 的页面流程、组件、数据模型(只读公开页面和自己的账号) |
/replica-architect | 规划技术栈、数据库 Schema、API 设计 |
/replica-design | 重建设计系统(颜色 token、字体、间距、组件) |
/replica-build | 逐屏重实现 App |
/replica-backend | 接入 Auth、数据库、支付、第三方集成 |
/replica-test | 跑完所有流程,按严重度记录 Bug |
/replica-diff | 对比克隆版与原版,给出功能完整度评分 + 缺漏清单 |
/replica-entrepreneur | 读真实用户差评,提取痛点,给出差异化定位方向 |
/replica-brand | 重命名品牌,扫描并清除所有原产品痕迹(名称/域名/颜色) |
/replica-launch | 生成落地页文案、定价页、应用商店描述 |
/replica-deploy | 预检通过后部署到你的域名 |
随附的 Python 工具脚本
Skills 调用这 6 个 Python 工具脚本(Python 3.8+,只用标准库,无需 pip install):
imgdiff.py— 截图布局对比(忽略颜色差异,专注结构差异)parity.py— 功能完整度评分reviews.py— 用户差评排序(只保留带链接的有据可查的评论)contrast.py— WCAG 对比度检查sweep.py— 代码库扫描原产品名称/域名/颜色残留listing.py— 应用商店文案字数和抄袭检查
/replica-deploy 在 sweep.py 扫描通过之前会拒绝部署。
/replica-entrepreneur:从差评里找机会
这是整套流程里最有意思的一个 Skill。
原版 App 的用户差评里记录着真实的痛点——功能缺失、交互反直觉、性能问题、价格不合理。/replica-entrepreneur 读这些差评,提取共性痛点,转化成「你的版本在哪里可以做得不一样」的定位方向。
它不是帮你抄一个一模一样的克隆,而是帮你造一个从用户抱怨出发的改进版。
Clean-Room 机制
项目的 Clean-Room 定义:
- 只看:功能做了什么、用户怎么走流程
- 不碰:源代码、素材、Logo、商标、私有 API
- 从零写:所有实现代码由 Claude Code 生成,不复制原版
/replica-recon 限定在公开页面和你自己的账号内操作,不绕过登录墙或 ToS。
项目明确建议上线前做商标检索和法律审查——clean-room 不等于免责,它只是降低代码层面的侵权风险。
技术栈
Skills 本身是 Claude Code 的 skill 文件夹(每个含 SKILL.md),文档举例的参考架构是 Next.js + Postgres + Stripe + Resend,但 /replica-architect 会根据目标 App 实际情况规划,不强制这个栈。
已知边界
- 复杂度上限:订票类 App 可行,电子表格引擎级别的产品不现实
- 不提供法律建议,上线前务必自行做商标检索和法律审查
- 只支持 Claude Code(Skill 格式),不是通用 Agent 脚本
- 1.3K stars,2026-10-03 才发布,尚未经过大量实战验证
一句话说清楚
replica-skill 是一套 Claude Code Skill,把「克隆一个 App」分成 11 步系统化:逆向分析功能→重建设计和代码→差评驱动差异化→清除原产品痕迹→部署上线。MIT,1.3K stars,一周内增速很快。
MIT。Jake Schincariol(opusjake.ai),2026-10-03 发布,1.3K stars。开源仅供学习参考,上线前请做法律审查。
Replica: 11 Skills to Clone Any App from Scratch
If you’ve looked at an app and thought “I could build this, and build it better” — replica-skill is the 11-step system for turning that thought into a shipped product.
MIT, 1.3K stars, published 2026-10-03, reached 1,000+ in one week.
GitHub: https://github.com/Jakeschincariol/replica-skill
11-Skill Pipeline
| Skill | What it does |
|---|---|
/replica-recon | Reverse-analyzes flows, components, data models (public pages + your own account only) |
/replica-architect | Plans tech stack, DB schema, API design |
/replica-design | Rebuilds the design system (color tokens, type, spacing, components) |
/replica-build | Implements the app screen by screen |
/replica-backend | Integrates auth, database, payments, third-party services |
/replica-test | Tests all flows, logs bugs by severity |
/replica-diff | Compares clone vs. original, produces completeness score + gap list |
/replica-entrepreneur | Mines real negative reviews, extracts pain points, generates differentiation angles |
/replica-brand | Renames brand, scans and removes all original product traces |
/replica-launch | Generates landing page copy, pricing page, app store description |
/replica-deploy | Deploys to your domain after pre-checks pass |
Bundled Python Tools
Six utility scripts (Python 3.8+, stdlib only, no pip install):
imgdiff.py— screenshot layout diff (color-agnostic, structural diff only)parity.py— feature completeness scoringreviews.py— negative review ranking (keeps only citable, linked reviews)contrast.py— WCAG contrast checksweep.py— codebase scan for original product name/domain/color traceslisting.py— app store copy length and plagiarism check
/replica-deploy refuses to deploy until sweep.py passes clean.
/replica-entrepreneur: Mining Competitor Pain Points
This is the most interesting Skill in the set.
Real negative reviews document genuine pain points — missing features, counterintuitive interactions, performance issues, pricing frustration. /replica-entrepreneur reads these reviews, extracts common themes, and turns them into positioning angles: where your version can be meaningfully different.
It’s not helping you clone the app identically — it’s helping you build the version that users actually wanted.
Clean-Room Mechanism
The project’s clean-room definition:
- Analyzes: what features do, how users flow through them
- Does not touch: source code, assets, logos, trademarks, private APIs
- Builds from scratch: all implementation is written by Claude Code, not copied
The project explicitly recommends doing trademark searches and legal review before launch — clean-room reduces code-level infringement risk, not all risk.
Known Limits
- Complexity ceiling: booking apps are feasible; spreadsheet-engine-level products are not
- No legal advice included; do your own review
- Claude Code only (Skill format)
- Very new (2026-10-03), limited real-world validation
TL;DR
replica-skill is a Claude Code Skill set that systematizes “clone an app” into 11 steps: reverse-analyze features → rebuild design and code → differentiate from negative reviews → sweep for original traces → deploy. MIT, 1.3K stars, fast-growing week one.
MIT. Jake Schincariol (opusjake.ai), published 2026-10-03, 1.3K stars. For reference only — do legal review before launching.
关于本站 · 免责声明
🍄 Mushroom Research Blog 是非营利、免费公开的个人科技观察博客与公众号 XStack18,不接受商业合作、不代表任何企业或机构立场,也不谋求商业利益。我们以个人视角客观中立地记录和分析 AI、Web3 等领域的最新模型发布与技术动态——不止转述新闻标题或二手信息,而是给出有独立思考的深入分析,希望帮更多人获得有价值的一手科技认知。
⚠️ 文中介绍的开源代码与模型,仅供学习交流与技术借鉴。它们大多仍处于早期阶段,有待进一步研究和验证,请勿直接用于工作或生产环境;如需采用,请先自行充分测试,并核实其许可证与安全性。
Open-source code and models featured here are shared for learning and reference only. Most are early-stage and still need further study and verification — please don't use them directly in your work or in production. Test them thoroughly and check their licenses and security first.
- 本站文章均为作者基于公开信息的个人研究与观点整理,不代表文中提及的任何公司、产品、模型的官方立场,未与其构成商业关联或合作关系。
- 科技行业信息更新极快,我们尽力保证内容准确、及时,但不对完整性、实时性做绝对保证,具体请以相关企业/项目官方公告为准。
- 文中引用的第三方商标、产品名称、图片、数据等版权归原权利人所有,我们会尽量注明来源;如你认为存在版权疑问或侵权,请通过下方邮箱联系我们,收到通知后会尽快核实处理(更正、加注来源或删除)。
- 文章内容仅为技术科普与个人观点,不构成投资、法律或其他专业建议,据此进行任何决策的后果需自行判断和承担。
📮 侵权 / 勘误 / 合作咨询:[email protected]
💬 评论与讨论
使用 GitHub 账号登录后发表评论