finds.dev← search

// the find

alchaincyf/zhangxuefeng-skill

★ 10,405 · MIT · updated Aug 2026

张雪峰.skill — 张雪峰的认知操作系统。高考志愿/考研/职业规划的实战思维框架。由女娲.skill生成。

A skill that loads a distilled model of Zhang Xuefeng's reasoning about gaokao choices, postgraduate study, and career planning into an agent as a SKILL.md file. It is aimed at people who want advice framed through his heuristics, such as the social sieve argument and working backward from employment data, rather than a quote collection. There is no code here. The product is one markdown prompt and the research behind it.

- The research is sourced in the open. Six research files cover writings, interviews, expression style, outside criticism, major decisions, and a timeline. The README names its primary and secondary sources and says it excludes Zhihu, WeChat posts, and Baidu Baike on purpose.

- Each of the five mental models is tied to where it came from. The social sieve idea, for example, is marked as used more than 20 times in lectures and livestreams, so the framing can be checked against the source.

- The tensions section keeps the contradictions, such as preaching restraint while talking faster than he thinks. That stops the persona from collapsing into a one-note caricature of a fast-talking exam coach.

- The format follows the open Agent Skills convention (markdown with YAML frontmatter), so it works across many runtimes and can also be pasted into a chat as plain text.

- No evaluation results are published. The Nuwa process is described as including three known-answer tests, an edge case, and a style test, but the repo shows no outputs or pass rates. The only evidence is the example dialogues the author chose to show.

- The demo answers state figures as fact with no source attached, such as 80% of art graduates not working in their major and the median finance graduate ending up at a neighborhood bank branch. A parent who reads those as sourced would be misled, and the persona voice makes the claims sound more certain than they are.

- Nearly all of the advice assumes the gaokao score system and the 985/211/双非 tier structure, and the README has no English summary. The heuristics would not carry over to other countries' admissions systems without a rewrite.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →