Colleague-Skill turns departing coworkers into AI clones

Colleague-Skill is an open-source project designed to clone departed teammates into usable AI agents. By ingesting chat logs and work documents, it captures a former coworker's coding standards, specific conversational quirks, and even their workplace defense reflexes. The repository went viral immediately, racking up over 6.7k stars on GitHub within three days.

Instead of training an expensive base model from scratch, the tool takes a lightweight persona-packaging approach. The architecture is split into two primary engines: a Persona layer that dictates behavioral attitude, and a Work Skill layer that handles actual task execution.

What it can do

Colleague-Skill targets the messy handover gap when a colleague leaves your team.

It works by processing chat logs, email snippets, or documentation exported from platforms like WeChat, Lark, DingTalk, Slack, or iMessage. The project integrates standard chat export tools such as WeChatMsg and PyWxDump to automate data ingestion.

In practice, it mimics professional habits and conversational cadence rather than just regurgitating dry knowledge. If set up as a senior backend engineer, it will enforce their specific API guidelines during code reviews. When asked about a bug, it reproduces their typical pushback and deflection logic.

If the generated response feels inaccurate, you can tune it through an active Correction layer. Simply telling the system that the person is usually more blunt or starts answers with a question mark will update the layer for future conversations.

The concept has already inspired multiple spin-offs, including clones for thesis supervisors, bosses, and ex-partners.

How it works

The framework organizes behavioral traits into a five-tier persona structure. This covers non-negotiable engineering rules, self-identity, surface communication style, decision-making logic under uncertainty, and interpersonal workplace dynamics.

When a prompt arrives, it first passes through the Persona layer, which decides how the character would react emotionally and whether they would push back. The Work Skill layer then produces the actual output following their specific technical patterns.

The code and persona framework are fully open-source and hosted on GitHub.

Why it matters

We think Colleague-Skill is a clever demonstration of behavioral prompt engineering over brute-force fine-tuning. By cleanly separating demeanor from technical competence, it offers an accessible way to preserve operational habits and project context that usually vanish during employee churn.

However, feeding internal communications and chat logs into third-party AI pipelines poses undeniable privacy and compliance hazards. If you plan to test it, treat it as a lightweight playground for organizational knowledge and strip out personal chats and proprietary data first.