Operations

How LeoYeAI Built teammate-skill to Turn a Teammate into an AI Skill

Built by LeoYeAI · · Source: GitHub

Quick answer

LeoYeAI published teammate-skill, an open source skill for Claude Code and OpenClaw. It takes a teammate's Slack, GitHub, email and docs and generates a Work Skill plus a five-layer Persona, so an agent can review PRs, answer questions and write the way that person does.

Result: 260 GitHub stars and 11 forks when we read the repo. The README states no accuracy or usage results.

What they built

LeoYeAI published teammate-skill, an open source (MIT) skill for Claude Code, OpenClaw and other AgentSkills-compatible agents. The README frames the problem as the knowledge a team loses when someone leaves. According to the README, the skill takes a person's Slack messages, GitHub PRs, emails, docs and your own description of them, and produces an AI skill that writes code in their style, reviews PRs to their standards and answers in their voice. The repo says it is powered by MyClaw.ai, which it lists as managed OpenClaw hosting.

How it works

You install it as a skill called create-teammate and run /create-teammate in a new session. The README describes the flow like this:

  • Intake. You answer three questions describing the teammate.
  • Source material. You provide data or skip this step. Slack and GitHub can be collected automatically with a Slack bot token and a GitHub token. Gmail, Teams, Notion, Confluence, JIRA, Linear, PDFs, images and plain text come in as exports or uploads.
  • Two-track analysis. One track builds a Work Skill covering the systems they own, their tech standards, what they focus on in code review, their workflows and their tribal knowledge. The other builds a five-layer Persona that runs from hard rules through identity, expression, decision patterns and interpersonal style.
  • Output. The skill writes SKILL.md, work.md, persona.md and meta.json into a folder for that teammate. You then call the teammate with a slash command named after them, with work-only and persona-only versions available.

The README also lists a few checks. A quality gate tests seven rules before you preview the result, and a smoke test runs three prompts after creation: a domain question, a pushback scenario and an out-of-scope question. A privacy scan looks for emails, phone numbers, API tokens and other personal data before export, and raw source files are kept out of git and exports by default.

Skills can be updated later. You can add new material, correct mistakes in conversation, and roll back, since every update archives the previous version. There are also commands to compare two teammates and export one as a package.

The README's own advice is that source quality decides skill quality. It says design docs and code review comments are the strongest inputs, GitHub PRs help the Work Skill most, and Slack threads help the Persona most. It recommends starting from a description and adding real data later.

The result

The README gives no accuracy numbers or user results. The repo showed 260 stars and 11 forks when we read it.

How to build something similar

  • Get the person's consent first, and check your company's rules on using chat and email history. This kind of data is full of personal information.
  • Start with written work that shows judgment, such as code review comments and design docs, before adding chat logs.
  • Keep what the person knows separate from how they talk. Two files are easier to fix than one mixed prompt.
  • Test the skill with a few real questions the person would have answered, including one outside their area, and correct it where it gets them wrong.

The original post

LeoYeAI · View on GitHub

Want to build something like this?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week.

Join Shipping Skool — Free

More use cases

← All AI agent use cases