Recently, GitHub has witnessed what can only be described as an absurd 'open-source frenzy.'

The 'Zhang Xuefeng.skill' project went live, refining Zhang Xuefeng's cognitive framework and conversational style into a functional AI avatar using his 5 books, 15 media interviews, and over 30 quotes. It was dubbed 'cyber immortality' by netizens.

At the same time, 'Colleague.skill' emerged, creating a workplace digital twin capable of 'continuing' a departed employee's role by scraping their Feishu records and DingTalk documents. It garnered 6.6k stars in 5 days, spawning over 850,000 variants like 'Boss.skill,' 'Mentor.skill,' and 'Ex.skill.'

Netizens spread memes while sharing: 'My skill has been uploaded.'

Behind this seemingly absurd open-source movement lies the rapid transition of digital immortality from a fringe sci-fi concept into everyday life. Controversy has followed: Can humans be commercially replicated? Where do legal, ethical, and technological boundaries lie? But today, we'd rather discuss why 'Zhang Xuefeng.skill' and its ilk went viral—and whether they represent the next standard form of AI applications.

1. Why Did 'Zhang Xuefeng.skill' Go Viral?

Over the past year, nearly everyone who has deeply used general-purpose large models has experienced a similar journey: from awe to indifference to boredom.

When you ask, 'Is journalism a good career choice?' a standard general-purpose large model typically responds:

'Journalism is a profession with profound social value, cultivating students' reporting skills and critical thinking. However, with traditional media's transformation, the job market faces challenges. While new media and corporate PR still demand content talent, traditional print media roles are shrinking. We recommend considering your personal interests, family financial situation, and long-term career planning...'

Is this wrong? Not at all—it's flawlessly correct. But is it useful? No, and it's purely verbose.

The responses from general-purpose large models are essentially 'correctly mediocre' statistical probabilities. Trained for broad applicability, they become neutral parrots, stripped of all emotion, stance, and edge. They'll list SWOT analyses but never dare to make decisions for you.

Yet real-world decisions involve trade-offs, information asymmetry, and resource competition.

This is the fatal flaw of general-purpose large models—and the fertile ground for 'Zhang Xuefeng.skill.'

Technologically, the barrier to entry is low. It relies on the Claude Skills architecture released by Anthropic in October 2025, which encapsulates domain knowledge into dynamically loadable 'capability modules.'

After OpenClaw gained popularity in early 2026, demand for Skills surged, prompting Microsoft, OpenAI, and others to follow suit. Implementation is simple: scrape chat logs, documents, emails, and use structured prompts to simulate someone's work habits and speech patterns.

Take 'Zhang Xuefeng.skill' as an example. Based on 5 books, 15+ in-depth interviews, and 30+ public quotes, it distills 5 core mental models and 8 decision heuristics. When activated, the AI discusses educational choices 'in a Northeastern Chinese bro's tone, with rapid-fire, joke-laden delivery.'

When faced with the same question, 'Zhang Xuefeng.skill' replies:

'Do you have family wealth? Capital to afford trial and error?'

'If not, blindly choose engineering, computer science, or a profession that teaches tangible skills.'

'Stop talking about journalistic ideals. When you can't eat, ideals are worthless. Choose journalism, and I'll knock you out.'

While controversial—critics call him an 'anxiety peddler'—it's undeniable that Zhang Xuefeng became a phenomenon not just by selling anxiety. He offered a 'survivalist decision-making model': choosing majors based on employment prospects and students' circumstances. For parents and students desperate for clear answers, this holds value.

This explains the frenzy around 'Zhang Xuefeng.skill' after his passing.

Currently, 'Zhang Xuefeng.skill' is limited to dialogue and advice. An ideal 'college admission agent' would not only scold you but also, after learning your scores and preferences, automatically call a national university admissions API to generate a strategy-based application list—even simulating one-click submissions.

This is why one Zhang Xuefeng trumps one ChatGPT: in complex real-world problems, users need not a general AI but a knowledgeable insider—or an expert who gives definitive answers.

2. Do We Need 100 'Zhang Xuefengs'?

Zooming out from college admissions to entire industries, 'Zhang Xuefeng.skill' is just a precursor. It hints at the ultimate direction of AI-era product ecosystems:

General-purpose AI will become as ubiquitous as utilities like water and electricity, while thousands of vertical 'expert agents' will thrive.

Why 100 'Zhang Xuefengs'? Primarily because human societies have extremely high expertise barriers and low tolerance for error.

The fatal flaw of general-purpose large models is hallucination. While harmless in casual chat, hallucinations are deadly in serious fields like medicine, law, and accounting.

You consult doctors for illness, lawyers for contract disputes, and designers for home renovations. You'd never ask a philosopher versed in astronomy and geography, 'How should my appendix be removed?'

General-purpose models like ChatGPT are those philosophers. They answer everything but lack medical licenses or courtroom experience.

Thus, we might need: 'Luo Xiang Agent' for law—trained on China's latest statutes and millions of public judgments, it explains labor arbitration odds in plain language and generates demand letters. Or 'Big Four Agent' for finance—linked to corporate ERPs, it uses top-tier auditor logic to flag tax loopholes and cash flow risks in real time.

Vertical agents' core strength: their search scope is strictly confined to specific knowledge bases, drastically reducing hallucination rates. They don't need to know everything—just be absolutely reliable in their domain.

After creating 100 vertical agents, the next breakthrough will be multi-agent collaboration. In the future, starting a company won't require hiring a full team. You'll simply 'hire' agents from an AI app store: a 'Musk Agent' for strategy, a 'Zhang Xiaolong Agent' for product details, a 'Li Jiaqi Agent' for livestream copywriting...

Your system becomes a group chat. You propose a complex business need, and the 100 vertical experts discuss, correct each other, even argue—ultimately deriving an execution plan.

This should be AI's ideal form for reshaping productivity: modularizing, reusing, and collaborating human expertise.

Before embracing '100 Zhang Xuefengs,' however, we must confront real challenges: privacy breaches, commercial exploitation, workplace ethical crises—topics dominating current discourse.

Technology is neutral; its impact depends on usage boundaries and rules. Only with regulatory frameworks, an end to wild growth, and standardized development will '100 Zhang Xuefengs' transition from open-source frenzy to reality.