SKILL.md file, that agents follow when the task in front of them matches it. Write your triage steps, review checklist, or naming convention once, and agents apply it without you repeating it in every prompt.
When to use a skill
One test decides it: if agents should follow it step by step, make it a skill. If they should look things up in it, put it in a knowledge base. A “check CPU, then credits, then traffic” triage runbook is a skill. A 40-page vendor manual or a service inventory goes in a knowledge base.
A skill can point agents at a knowledge base document, so the two work well together. Skills are one of several surfaces that hold the context you give CloudThinker; Teach CloudThinker compares all of them.
How a skill gets used
A skill is used only when all three are true:- It is enabled. The switch on the skill is on. New skills start enabled.
- It is available on the surface. The surface where the work happens is ticked under Available in (see the table below). New skills start with Chat only. This applies to
/too: the/menu lists every enabled skill, but chat can’t use one without Chat ticked, and the agent carries on without it. Tick Chat on any skill you plan to run by name. - It fits the task. On every surface, including Review, the agent sees only the skills available there and loads the ones whose name and description match the work. It can load several skills for one task. So write the description in the words of the request. Bad: “CPU stuff”. Good: “Investigate high CPU or a slow EC2 instance”.
You can also run a skill by name: type
/ in the chat prompt box and pick it from the menu. When an agent uses a skill, its activity shows a Load Skill step.
CloudThinker also publishes ready-made skills in the Skills Hub. An installed hub skill becomes a workspace skill you manage like your own, and it doesn’t change when the hub version does. To install one, see custom skills.
A new workspace’s skills list starts empty. CloudThinker’s built-in skills, such as pdf, docx, pptx, xlsx, and deep-research, work without setup. They don’t appear in the list, so you can’t edit or turn them off. Their names are reserved: your own skill can’t use one.
Who can do what
In More → Workspace settings → Skills, only workspace Admins can create, edit, attach, or delete skills, including turning the switch on or off and changing Available in. Developers can see the skills list and run skills with/. Viewers can see skills but can’t chat, so they can’t run them. See roles.
Three paths add a skill without an Admin, and each one goes live in Chat for every agent:
- Skills Hub install. The install doesn’t check the workspace role. Any member who opens a hub skill’s install link from the public catalog can install it into their current workspace.
/create-skillin chat. Any member who can chat can run it, and the agent writes the skill itself.- Learning Mode. A custom agent on Auto or Offer saves skills from chat. Audit Logs don’t record these. Set it to Off if every skill should come from an Admin.
Ways to create a skill
Custom skills has the steps. For the file itself, see Skill format.
Next steps
Custom skills
Write effective skills and attach them to Review, Incident, and Assessment
Skill format
Learn the SKILL.md structure with a complete worked example
Custom agent quickstart
Pair a skill with a custom agent and test it end to end
Knowledge base
Give agents documents and websites to look things up in