Wangziru Analysis Skill V1 — Development skill for Claude Code

王自如式表达拆解（第一版）· wangziru-analysis · V1

**简体中文** · [English](README.en.md)

学结构，不学腔调。

   

**这是第一版（V1，对应原仓库 v1.0–v1.3）——最初版本，原始配方**：方法不贴样本、逻辑更通用，不少小伙伴反馈这版效果更好，因此单独成库。工作流是蓝图式：先交《拆解蓝图》、过 13 道质量门，再成稿。**第二版是数据加强版**：基于更多视频数据迭代，更像他本人，但对原样本可能过拟合，见 [think-like-wangziru](https://github.com/chrislitality-svg/think-like-wangziru)（原 wangziru-analysis-skill-v2，已改名「像王自如一样思考」）。两版建议都试试。

**This is Edition 1 (V1, i.e. v1.0–v1.3 of the original repo) — the original edition, the original recipe**: a general-purpose method that doesn't hug the sample; many users report better results with this one, which is why it now lives in its own repo. Workflow is blueprint-style: produce a "deconstruction blueprint", pass 13 quality gates, then write the script. **Edition 2 is the data-enhanced edition** — iterated on more video data, closer to the man himself, but possibly overfitted to the original sample; see [think-like-wangziru](https://github.com/chrislitality-svg/think-like-wangziru) (formerly wangziru-analysis-skill-v2, renamed "Think Like Wang Ziru"). Try both.

一个给 AI 编程助手（Claude Code / Codex / Cursor 等）用的 **Agent Skill**：用王自如的表达系统，把一个"大家都在聊、没人讲清楚"的选题，拆成**观众听完能复述的思考框架**，再写成能直接念的口播稿；也能给旧稿做体检，或者把 AI 写的稿子改成人话。

推荐模型 · Recommended Models

优先级	模型	说明
🥇 首选 / Primary	Gemini 3.8 Flash	本 skill 推荐的第一模型：长上下文能一次读完全部 references，逻辑链表和长句骨架排得稳，性价比最高 / First choice: holds all reference files in one context, stable logic-chain tables and long-sentence skeletons, best value
🥈 次选 / Secondary	DeepSeek V4 Flash	首选不可用时的替代：中文语感接近口播基准，跑 tools/ 检测脚本表现稳