Knowledge and workspace tools
Skills for repeatable work
A skill is a saved, checked way of doing a kind of job, so your AI agent doesn't have to guess the method fresh each time. For example, if you always want quotes costed the same way, or handover packs built in the same order, a skill can hold that method.
Find a skill before inventing a process
If your agent is new to the workspace, ask it to start with a quick orientation. Then ask it to search for a skill that matches the job you need done. Read the most relevant skill before asking for work that needs a set method, such as a step-by-step process or a specific kind of document.
Skills can belong to your workspace, or be shared ones anyone can use. Ask your agent to check both. Only follow a skill that is marked active: a draft can be read while someone is still improving it, but shouldn't be used to do real work yet.
Keep the boundary clear
A skill gives the method, but it doesn't make the decision for you and it can't grant your agent access it doesn't already have. Ask what the skill needs from you before you start, give it the facts it asks for, then check the result before anything is sent, saved for good, or otherwise hard to undo.
A prompt to try
Search the active skills for a repeatable way to prepare a site visit. Show me the best matches and why they fit. Don't run one yet.
Read the skill you found and list what you'll need from me before we begin.
Improve repeatability carefully
If a process keeps coming up, turning it into a skill can make it easier to follow next time. Only change a skill when you can describe the improved method clearly and try it out against real work.
See Workspace knowledge and Automations.