Shared Skills Repository — AI System Brief | Maneesh Maddala Skip to main content
From Test to Trust TEST TRUST
Connect
← Back to AI Systems

Shared Skills Repository

From every team building skills in isolation to one GitHub repository the whole company publishes to and reuses.

Claude Skills · GitHub · Claude Desktop · MCP Connectors · skill-creator · Claude Enterprise · Knowledge Sharing

Across our company, teams were writing their own Claude skills and keeping them to themselves. When two teams worked on similar problems, they often built the same skill twice, because neither could see the other’s. A skill that broke had no name attached, so no one knew who to ask.

Our team built a shared skills repository on GitHub where any team can publish its skills and check for an existing one before building a new copy. Every skill carries its author’s username, so a broken or unclear one traces straight back.

The repository lives in GitHub, where everyone already has access. Skills are split into two groups, generic ones any team can use and team-specific ones, so an author checks the right set first. Developers browse and search it directly.

For staff who do not work in GitHub, a connector links Claude Desktop to the repo. They ask Claude to search for an existing skill and get the same catalog from the chat window they already use.

To add a skill, an author runs /skill-creator, describes what it should do, and gets a working scaffold back. The connector publishes it to the repo with the author’s username attached.

Architecture diagram
Shared skills repository flow Developers browse and commit to a GitHub repository directly. Non-technical staff reach the same repository through a Claude Desktop connector, searching existing skills from chat. The repository holds generic skills any team can use and team-specific skills. To add a skill, an author runs the /skill-creator skill to scaffold it, then publishes it; the publish step stamps the author's username. A returning path shows that finding an existing skill lets an author skip building a new one. GITHUB · SHARED SKILLS REPO browse + commit search new skill scaffold + username found an existing skill · skip the rebuild Developer Non-tech staff Claude Desktop connector /skill-creator Publish stamps username Generic skills Team-specific skills search / publish flow reuse an existing skill GitHub · shared repo

Two people can still start the same skill on the same day. The workflow cannot prevent that, but it makes checking cheap: anyone can search the repo first, in GitHub or through the connector, before they build anything.

When a skill from another team malfunctions, you need to know who to ask. Every skill is published with its author’s username attached, so a question or a bug goes straight to the person who wrote it.

A GitHub-based catalog could leave out everyone who does not work in GitHub. The Claude Desktop connector gives them the same search and the same way to publish, from chat, so the repository is not developer-only.

Teams stopped rebuilding skills that already existed and started pulling the published ones.

Developers and non-developers work from one catalog, one through GitHub and the other through the connector.

Every skill has a named owner, so a broken or unclear one traces back to the person who wrote it.