Skills are like automated standard operating procedures for AI. You teach Claude or ChatGPT the steps of something you do repeatedly (writing grant reports, drafting board updates, applying brand standards) and it invokes those steps automatically whenever the task comes up, instead of you re-explaining yourself every single time.
Below are a video and slide deck of our meetup presentation on August 20th, 2026, where Ken Krugler presented a 50-minute workshop covering how to use, build, and safely manage AI skills for nonprofits, including live demos of nonprofit branding and organizing draft content. He also showed how to convert a project into a skill.
One topic that surfaced multiple times is how to be safe when using skills. A simple sequencing (specifically for Claude) is the following…
- Start by only installing skills that are published by Anthropic.
- And have your Claude Team admin disable the sharing (either between people, or to the entire org) of skills by an individual.
- Then provide basic training on “safe skills” to all Claude users. This should include a process for them to request access to additional Anthropic/3rd party skills.
- After that, enable individual skill creation, but still disable sharing of skills by individuals.
- Have someone in the organization learn the risks associated with using non-Anthropic skills, including execution of code packaged inside a skill (via CoWork/Claude), and “prompt injection”.
- Once you have this person, it can be safe to add internally-created and 3rd party skills that have been vetted and approved by them.