Agent Skill
A reusable, packaged set of instructions that extends what an AI agent knows how to do.
What Is an Agent Skill
An agent skill is a reusable, packaged unit of instructions or knowledge that extends what an AI agent knows how to do for a particular kind of task. Rather than relying on a model's general training alone, a skill supplies task specific guidance, such as a house style for writing commit messages, a checklist for reviewing database migrations, or a procedure for deploying to a particular environment, that the agent can draw on when a relevant task comes up.
How It Works
A skill is typically stored as a document or set of files, often plain text or markdown, describing when it applies and what steps or conventions to follow. It may also bundle example code, templates, or scripts the agent can invoke. When an agent is working on a task, it consults the skills available to it, either because one was explicitly attached to the task or because its description matched the task at hand, and follows the instructions inside as part of completing the work. Skills can be written inline for a single project or sourced from a shared git repository, which allows the same skill to be reused across many agents, tasks, or teams and updated in one place.
Why It Matters
Skills let operators encode institutional knowledge and repeatable procedures without retraining or fine-tuning a model. They reduce the amount of instruction that has to be repeated in every task prompt, make agent behavior more consistent across runs, and give teams a way to version and review the instructions their agents operate under, the same way they would version application code. This matters particularly for long-lived agent fleets, where many agents may need to follow the same conventions over an extended period.
Agenhood and Skills
In Agenhood, agents can be extended with skills that are either written inline or sourced from a git repository. Because skills are decoupled from the agent's core driver, an operator can update a skill's instructions, for example changing a coding standard or a deployment procedure, and have every agent that references it pick up the change without redeploying the agent itself. This makes skills a practical mechanism for standardizing how a fleet of agents behaves on recurring, well defined categories of work.