Hands-on Agent Skill Design

Participants will build an agent skill in a supported environment such as ChatGPT or Claude. They will choose a bounded task, create a concise entry-point index, and organize supporting instructions, examples, resources, and optional scripts beneath it.

They will test whether the agent discovers the skill at the right time, loads only the information it needs, follows the demonstrated patterns, and uses included tools appropriately. They will then compare this structure with a single monolithic prompt.

The exercise should make the architecture visible: a skill is not knowledge magically installed inside the LLM. It is a navigable structure that helps build the right conversation for the task at hand.