On this page
Quick answer
Write your best on-call runbook once as a skill, and Rubix follows it in every chat and every RCA, the same way for everyone on the team. Skills use the open Agent Skills format, SKILL.md, so a runbook you already wrote for another agent works here too.
Every team has a runbook that only works when the right person runs it. The steps live in a wiki, the judgement lives in one engineer's head, and the 3am version skips half of it. A skill keeps the steps and the judgement together, and Rubix runs it every time a matching incident or question comes in.
How it works
- Pick a runbook your team already trusts, such as checking a service after a deploy.
- Create the skill. In Customize > Skills, choose New skill, or Fill from a SKILL.md file. Give it a name, a line on when to use it, and the steps.
- It is on as soon as you create it. Rubix uses it in chat and in RCAs when an incident or a question matches its description.
- Try it. Start a new chat, pin the skill with
@> Skills and ask a real question. Then ask again without pinning, to check Rubix picks it on its own.
Example: a post-deploy check
---
name: post-deploy-verification
description: Use after a production rollout, or when an incident starts within an hour of one.
---
1. Find the rollout: service, version and the time it finished.
2. Take a 30-minute baseline of error rate and latency before it.
3. Compare the 10 minutes after the rollout with the baseline.
4. Flag anything more than 20% worse, with the metric and the time.
5. If something is worse, check the diff for config changes first.
What a good skill has
- One clear job, described in one line, so Rubix knows when to reach for it.
- Steps written for a new teammate: what to check, in what order, and what counts as bad.
- Your team's own knowledge: names of services, the setting that usually causes trouble, the dashboard that matters.
Built-in skills you can replace
Rubix ships with skills for Kubernetes, AWS, Google Cloud, Azure, machines, incidents, RCA review and RubixKube itself. A skill you create with the same name replaces the built-in one, and Make a copy gives you a starting point.
Skills guide Rubix, they do not grant access
A skill changes how Rubix works, not what it may touch. Modes, approvals and hooks still apply, and the tools a skill lists are guidance, not a permission. To keep a skill from writing, keep its users in Observe or Assist, or block the tool with a hook. Only admins create skills.
For a prompt you run by hand, such as /postmortem, create a saved command in Customize > Commands instead.
Start free, or read Add custom agent skills and Skills.
Frequently asked questions
What format do skills use?
The open Agent Skills format: a SKILL.md file with a name, a description of when to use it, and instructions. You can paste one in with Fill from a SKILL.md file.
Does a skill run in RCAs or only in chat?
Both. Rubix uses a skill whenever an incident or question matches its description.
Can a skill make changes on its own?
No. A skill grants no permissions. Every change still goes through your mode, approvals and hooks.
How long can a skill be?
The description can be up to 1,024 characters and the instructions up to 16 KB.