When a Teammate Leaves, Their Knowledge Doesn't Have To

We've all been there. A key teammate moves on to a new role, and suddenly there's a gap. Who handled that specific deployment quirk? Who knew the exact incantation to debug the legacy billing service? Tribal knowledge walks out the door, and the team scrambles to fill the void.

What if you could capture a slice of that expertise before they go? Not just through frantic documentation, but as an interactive, queryable resource? That's the intriguing premise behind Colleague-Skill, a project that turns departed teammates into reusable AI skills.

What It Does

Colleague-Skill is a tool that leverages the Humanizer framework. In simple terms, it allows you to create a "skill" modeled after a colleague by feeding it examples of their past work—think Slack conversations, code reviews, pull request comments, and documentation they've written. The system processes this data to build a profile of their communication style, technical preferences, and problem-solving approach.

Once created, you can interact with this AI skill through chat. You can ask it questions like "How would you approach debugging this API timeout?" or "What's your preferred pattern for this React component?" and get responses that mimic how your former colleague might have answered, based on their historical data.

Why It's Cool

The clever part isn't just the AI—it's the specific, pragmatic use case. Instead of a generic assistant, you get a contextual one trained on your team's actual history and jargon. It's a form of knowledge preservation that's more dynamic than a static wiki.

The implementation is also developer-focused. It uses the Humanizer framework, which is designed to "humanize" AI interactions by grounding them in specific personas and data sources. This means the skill isn't just a language model spitting facts; it's structured to emulate a particular person's tone and depth, making the interactions feel more relevant and useful.

Potential use cases are pretty clear:

• Onboarding new team members by letting them "query" the previous expert.
• Preserving niche, institutional knowledge about specific systems.
• Providing a consistent reference for agreed-upon code styles and architectural decisions long after the original advocate has left.

How to Try It

The project is open source and available on GitHub. You'll need to have Python and uv installed to get started.

1. Clone the repository:

   git clone https://github.com/blader/humanizer.git
   cd humanizer
   

2. Install the package in editable mode:

   uv sync
   

3. The project provides a framework. You'll need to configure a data source (like exported Slack data or code review history) to train a specific "colleague" skill. Check the examples/ directory and the project's README for detailed guidance on setting up your own data pipeline and defining a skill.

Final Thoughts

Is this a perfect, seamless replacement for a great teammate? Of course not. AI can't replicate human intuition, creativity, or the nuances of real-time collaboration. It also raises valid questions about privacy and data usage that any team would need to address thoughtfully.

But as a practical tool for mitigating one of the most painful parts of team change—the loss of hard-won, specific knowledge—it's a fascinating experiment. It shifts the focus from documenting everything to capturing how a specific person reasoned. For developers dealing with complex legacy systems or frequent team turnover, that's a concept worth exploring. It's less about hype and more about a clever, slightly eerie, but potentially very useful form of backup.

Follow for more interesting projects: @githubprojects