A Claude skills library: 40 shared, 80+ behind the scenes

What a skill is

A Claude skill is a set of written instructions Claude follows for one kind of task: the steps, the standards, and the files, knowledge bases and tools it should draw on. It’s plain text, so people can read and edit it, and Claude follows it the same way every time.

Two layers

Shared. About 40 skills and automations I’ve shared across departments, so a process one team figured out becomes something every team can run.

Behind the scenes. About 80 more that stay internal and are part of my own process and agentic workflows: project specs and instruction sets that point Claude to the right knowledge bases, so every agent follows the same standards and teams get consistent results.

Tracking and metrics

I track the library the same way I track every project in the AI portfolio:

• What exists. Every skill, spec and instruction set is counted: about 40 shared across departments and 80+ internal. Before anyone builds something new, I can check whether it already exists.
• No duplicates. I keep an eye on everything being built, so when two skills overlap I merge them.
• Ownership. Skills that come out of a project are tracked in Linear alongside that project, with its portfolio owner and executive stakeholder, so every skill has someone accountable for it.
• Usage and cost. The AI usage and cost dashboard I’m building shows each person their own token usage and spend, and gives admins a view by person, so we can measure adoption and cost.
• Change history. Skills and their code live in GitHub, so every change is tracked, reviewed and reversible, and the reasoning behind them is documented in Obsidian.
• Adoption. Most skill use starts in my training sessions.

Getting people to use them

The library and the training go together. More than 50 people have been through my sessions, classes and demos. I also sit down with people to build their own skills and connect them to the third-party services they already use.