case-studies
Installation
- Make sure Claude is on your device and in your terminal.
Skills load from
~/.claude/skills/when Claude Code starts up — so you need it on your machine first. If you don't have it yet, install it once with the command below, then runclaudein any terminal to verify.One-time setupnpm i -g @anthropic-ai/claude-codeAlready have it? Skip ahead.
- Paste into Claude Code or into your terminal.
This copies the whole skill folder into
~/.claude/skills/case-studies-boyang-hu/— the SKILL.md plus any scripts, reference docs, or templates the skill ships with. Safe default: works for every skill.Faster alternative (instruction-only skills)
Skips the clone and grabs only the SKILL.md file. Don't use this if the skill ships Python scripts, reference markdowns, or asset templates — they won't be downloaded and the skill will fail when it tries to load them.
Quick install (SKILL.md only)Sign up to copy - Restart Claude Code.
Quit and reopen Claude Code (or any other agent that loads from
~/.claude/skills/). New skills are picked up on startup. - Just ask Claude.
Skills auto-activate when your request matches the skill's description — no slash command needed. Trigger phrases live in the skill's own frontmatter; you can read them in the “What this skill does” section above.
Prefer to read the source first? Open on GitHub.
When Claude uses it
Claude activates this skill based on the context of your request.
What this skill does
case-studies/skill.md — SKILL.md 的实证记录
何时加载本文件:不在必经集合里。只在想知道 SKILL.md 某条规则为什么存在、或要核对它的实证强度时读;小节名与 SKILL.md 一一对应。
使用前提与授权
法务考量不得削减镜像完整性或门的覆盖面:不得以"反正不公开""不该多存一份"这类法务理由留洞。实证:某项目以"产出永不公开"为由对一类资产"登记、不补抓",缺了约 60% 的资产而五道门始终全绿,藏了四个里程碑【objectarchive】。
适用范围
C1(重构式逆向)——路线:flight-decode 建坐标系 → 重构一个可构建的 Next 工程(客户端一方组件按 C2 逐字译,服务端组件从 flight 树反推为显式登记的推断物)→ verify-flight 语义门收口(模块 id 全局双射)。实测 rauchg.com(Next 16/Turbopack):18/18 路由语义一致,盲逆向对答案结构 ≈95%/行为 ≈98%。
C2(可做,按 A 类跑)——实测一个 R3F 站:useFrame 回调里就是 MathUtils.damp(x, y, 7, t) 这样的普通命令式代码,逐字切片 18 个模块换进页面后 CLEAN、8 个 canvas 齐全、跨侧 99.5%。切片器不关心范式——它切的是字节。 渲染器当平台层从镜像伺服。
X 类(可抢救)——⭐ 抢救产出是标准镜像——X 类可走完 L3 全程(实测 first-launch:策略 A 外壳 + 数值门 9,856 样本全等 + 像素带宽内 + 自包含门,无一道门需改语义)。
X 类(可抢救)——⛔ "CDX 无覆盖才是真不可做"按资产层读,不按站读:IA 爬虫不执行 JS,清单/拼接驱动的站可以代码层覆盖 100% 而画面层为零(实测 157/160 资产任何年代零捕获,断网即白屏)——Step 0 先做分层覆盖侦察(推导 + CDX 前缀查询)预判抢救深度,见 archival-rescue.md §1.9。
Workflow / Flow — M1
⛔ 第一个动作是判 bundle 形态(扁平拼接 / 模块化打包 / 多 chunk),再选工具——分层表扫顶层声明,而 webpack 打包产物的顶层声明数是 0,边界与依赖边由打包器给定(实测 24,378 行 → 569 个现成模块,用 scripts/module-map.mjs)。
Workflow / Flow — M(n+1)
到 M(n) 为止产物已证明正确但人读不了(实测:14,271 行挤在一个文件里,e 出现 2962 次,注释占 0.2%)。
遇到巨型模块时先测「延迟绑定少数末尾单例」的收益曲线再决定(实测 6 个绑定即从 11,246 行降到 1,013 行,而换模块系统要赔上整条工具链才换来同样粒度)。
⭐ "这件事做不到"这个判断极不可靠——实测两次判为结构性不可能,两次真凶都是自己工具里的一行 bug;先怀疑测量它的工具,再怀疑对象(readable-source.md §3.1–3.1.3)。
Script Directory
⛔ 不要改成手写词法器——本 skill 里试过,一个含引号的正则字面量把它带偏了 16,177 行(F27)。token 流上的括号匹配是精确的,文本上的括号匹配是对字符串/正则/注释的猜测。
⚠ 这条线是被违反之后才被发现的:module-map.mjs 依赖 @babel/*,却在 scripts/ 里住了整整八个版本,而同一份纪律的原话就写在它上面三行。一条只写在文档里、没有任何东西去查的规矩,会安静地失效。
Related skills
Skill Builder & Optimizer
anthropics
Create, edit, and optimize Claude skills with performance testing and benchmarking.
Org Change Management
alirezarezvani
Guide teams through organizational changes using the ADKAR model and communication strategies.
Audio/Video Transcription
daymade
Transcribe audio and video files to text with fast local or remote processing.
Claude Export Conversation Fixer
daymade
Repair broken line wrapping in Claude Code exported conversation files.