Skills
Skills are what an agent knows. They make its answers specific to your work.
A skill is a piece of knowledge written in plain markdown: a runbook, a decision guide, the steps your team follows, or the quirks of an API. Skills are what make an agent's answers specific to your work instead of generic.
What makes a good skill
A good skill adds something the model doesn't already know. Kavela scores each skill against the model, so the bar isn't "is this well written," it's "does this make the agent more capable."
Good skills are usually:
- Specific. The exact steps, the real edge cases, the names of your systems.
- Focused. One skill, one topic. Small skills are easier to find than one large document.
- From real work. Paste in docs, runbooks, notes, or an API spec you already have.
How skills are used
Skills are not added to every prompt. The agent retrieves only what a task needs:
Find
The agent searches its skills and picks the few most relevant.
Load
It reads the full content of those skills.
Apply
It uses that knowledge to answer or act.
This is the context loop. It's why an agent gets better as you add skills, instead of slower.
Organizing skills
- Groups keep related skills together and control what an agent can reach.
- Versions record every change, so you can see what changed and roll back if you need to.
- Live editing lets your team work on skills together, with cursors and presence.