What is a skill in an AI agent?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A skill is a reusable set of instructions for carrying out a particular kind of task. It explains the purpose, the information required, the steps to follow and what a useful result should contain. Think of it as a procedure an agent can consult, often written largely in ordinary language.
An agent is software that can choose actions while pursuing a goal. A skill might tell it how to prepare an invoice summary: locate the supplier and date, identify missing information, calculate totals using an appropriate tool and present exceptions for review. The skill does not itself become a separate employee or an independent agent.
Some implementations package skills in folders containing an instruction file and optional scripts, templates or reference material. A script is a small program that carries out specified operations. A template provides a starting structure for a result. The Agent Skills format is one documented approach, but not every product uses the word “skill” in exactly the same way.
A tool is different. It performs a callable operation, such as opening a file or calculating a sum. A skill can tell an agent when and how to use that tool. A plugin is a package that may contain several skills and other capabilities. A subagent is a separate agent instance assigned work. These describe a procedure, a function, a package and an executing worker respectively.
A skill also does not automatically run on a schedule. A scheduler is software that starts a task at a specified time or when a specified event occurs. Writing “prepare this report every Friday” inside an instruction file is not enough unless the surrounding system provides and configures that scheduling capability.
The distinction between instruction and enforcement matters. A skill can say that an email must not be sent without approval. The application should enforce that boundary through permissions rather than relying only on the text being followed.
A clinic could write a skill for preparing its daily referral review: retrieve current cases, identify missing prerequisites, show the source of each status and produce a list of unresolved actions with owners. The skill should say what to do when records disagree and which decisions require staff judgment. A scheduler would start the daily task; the skill would describe how to perform it. Testing the procedure on difficult cases matters because reusable instructions reproduce weaknesses as efficiently as strengths.
Source: Agent Skills — Overview.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy