Cyber Immortality: Packing a Coworker Into an AI

目录

I recently came across a really fun open-source project, colleague-skill. It can “digitize” your coworker — not by simply backing up chat logs, but by actually generating an AI Skill that can do their job in their place.

Why do we need a “Coworker Skill”?

The image at the top of the project README is a great joke — someone says the people building large models are “code traitors” who killed off their frontend brothers, and who will go on to kill off backend, QA, ops, network security, and IC brothers too, and eventually all of humanity.

But joking aside, the problem the project actually wants to solve is very real:

• A coworker jumps ship, leaving behind a pile of docs nobody maintains
• An intern quits, leaving only a half-finished project and an empty desk
• A mentor graduates, taking all their experience and context with them
• A partner transfers teams, and years of rapport reset to zero overnight
• A predecessor hands off, trying to sum up three years of accumulated knowledge in three pages of docs

Plainly put: it’s knowledge drain.

What is “cyber immortality”?

The core idea of this project is: distill your coworker’s way of working, their technical standards, their tone of voice, even their blame-shifting skills, all into a single AI Skill.

The generated Skill isn’t a simple Q&A bot — it’s split into a two-layer architecture:

Part	Role
Work Skill	System architecture, technical standards, workflows, experience knowledge base
Persona	Five-layer personality structure: hard rules → identity → expression style → decision-making mode → interpersonal behavior

The runtime logic is clear:

Receive task → Persona decides attitude → Work Skill executes → Output in their tone

What data can it collect?

The project supports a variety of data sources:

Source	Message logs	Docs/Wiki	Multi-dimensional tables
Feishu (auto-collect)	✅ API	✅	✅
DingTalk (auto-collect)	⚠️ Browser	✅	✅
Slack (auto-collect)	✅ API	—	—
WeChat chat logs	✅ SQLite	—	—
PDF/images/screenshots	✅	—	—
Email .eml	✅	—	—
Markdown/text	✅	✅	—

Feishu is the most complete — just type in a name and it collects everything automatically. WeChat is less stable due to encryption issues; the project recommends first exporting with tools like WeChatMsg or 留痕 and then importing.

How realistic is the result?

The README gives a few examples, and I laughed out loud reading them:

Scenario 1: Code Review

User        ❯ Take a look at this API design for me

colleague.skill ❯ Hold on — what's the impact of this API? The background isn't spelled out.
                  (after reading) N+1 query, fix it. Use the unified return structure
                  {code, message, data}, that's the standard, no need to ask why.

Scenario 2: Blame-shifting

User        ❯ This bug was introduced by you, right?

colleague.skill ❯ Do the release times line up? That requirement touched several places, and there were other changes too.

That tone — it nails it.

The tag system is a hoot

The project supports all kinds of “personality tags” and “corporate cultures”:

Personality tags: conscientious and responsible, master blamer, perfectionist, good-enough-is-fine, procrastinator, PUA expert, workplace politics player, managing-up expert, passive-aggressive, flip-flopper, man of few words, reads-but-never-replies…

Corporate cultures: ByteDance style, Alibaba flavor, Tencent flavor, Huawei flavor, Baidu flavor, Meituan flavor, first-principles thinking, OKR zealot, big-tech assembly line, startup camp…

Level support: ByteDance 2-13-3+, Alibaba P5P11, Tencent T1T4, Baidu T5T9, Meituan P4P8, Huawei levels 1321…

It basically covers the entire “jargon system” of China’s internet giants.

How to use it

The project is a Skill designed for Claude Code, and installation is simple:

# Install into the current project
mkdir -p .claude/skills
git clone https://github.com/titanwings/colleague-skill .claude/skills/create-colleague

# Or install globally
git clone https://github.com/titanwings/colleague-skill ~/.claude/skills/create-colleague

Then in Claude Code type /create-colleague, and follow the prompts to enter the coworker’s name, level, and personality tags, and select the data source.

Once generated, you can invoke it with these commands:

Command	Description
/list-colleagues	List all coworker Skills
/{slug}	Invoke the full Skill (Persona + Work)
/{slug}-work	Work ability only
/{slug}-persona	Personality only
/colleague-rollback {slug} {version}	Roll back to a historical version

Evolution mechanism

This project isn’t just “copy-paste” — it also has an evolution mechanism:

• Append files → automatically analyzes the delta → merges into the corresponding part, without overwriting existing conclusions
• Correct via conversation → say “he wouldn’t do it that way, he should be xxx” → written into the Correction layer, takes effect immediately
• Version management → auto-archives on every update, supports rolling back to any historical version

This means your “cyber coworker” gets more and more like the real person the more you use it.

Technical report

The project team also wrote a technical report, Colleague.Skill: Automated AI Skill Generation via Expert Knowledge Distillation, which details the system design, the two-layer architecture, multi-source data collection, and the Skill generation and evolution mechanisms.

You can tell this isn’t just playing for laughs — there’s serious technical thinking behind it.

Some reflections

This project got me thinking about a few questions:

1. A new way to pass on knowledge: Traditional documentation handoffs often become a mere formality, whereas this kind of Skill can preserve even tacit knowledge (like “when to shift the blame”).

2. The boundary of AI anthropomorphism: When AI can imitate a specific person’s way of working and tone of voice, the concept of a “digital legacy” may need to be redefined.

3. A reflection on workplace culture: Those tags (master blamer, PUA expert, managing-up expert) are funny, but they also reflect certain not-so-healthy workplace cultures.

Of course, the project README also stresses: this is still a demo version, and if you find bugs please file an issue.

Final words

Honestly, my first reaction after reading through this project was to laugh; my second was admiration.

I laughed at those all-too-real scenarios and tags; I admired how the team took the concept of “knowledge distillation” and implemented it in such a down-to-earth way.

If you too have that kind of coworker — the one who, once they quit, can never be replaced — maybe give this project a try. Pack them into an AI and achieve true “cyber immortality.”

---

Project address: https://github.com/titanwings/colleague-skill

Created by @titanwings | Powered by Shanghai AI Lab · AI Safety Center

文章作者：阿文

文章链接： https://www.awen.me/post/616e548f.html

版权声明：本博客所有文章除特别声明外，均采用 CC BY-NC-SA 4.0 许可协议。转载请注明来自 阿文的博客！