Skip to main content
Compliance rule for prohibited AI practices under the EU AI Act with Critical severity and Enabled status
Compliance rules turn your internal policies into automated checks. Instead of reviewing every agent and skill by hand, you define a rule once and choose whether it checks agents, skills, or both. Agent results appear in each agent’s compliance checks. Skill results stay on the rule, so you can see which resources meet your standards and which need attention.
Only workspace admins can create, edit, delete, or turn compliance rules on or off. Anyone with Governance access still sees rule results on the agents and skills they can open.

How rules work

Choose Agents, Skills, or both under Applies to. A new rule starts with Agents selected. A rule that already exists stays on agents until you change Applies to. At least one of the two has to stay selected. When the rule applies to agents, an agent either passes or fails. That check runs when the agent is created and whenever a new version is published. You can also rerun the rule on the agents and skills in its current scope, for example after you change the failure condition. When the rule applies to skills, a skill is checked after it is created or imported, and after changes to its instructions, description, or files. Saves within 10 seconds are grouped into one check. The check covers the name, description, instructions, and readable text and script files. PDF, Office, image, and other binary files are not analyzed. When the analyzed content has no violation and some files were skipped, the result is Partial analysis. A violation is still a failure. The skill result stays on the rule. It does not change agents that use the skill.
Skills tab of a compliance rule with a skill marked Partial analysis because some files were not checked
When an agent fails a rule, the failure appears in the agent’s compliance checks in the agent detail view, together with the rule’s severity. From there you can decide whether to flag or disable the agent.

Create a rule

You can create a rule from scratch or start from a template.
1

Open Compliance rules

Go to Compliance rules in Governance and click New rule. Pick one of the compliance templates as a starting point, or choose Start from scratch.
Template picker for new compliance rules with Start from scratch and pre-written templates like EU AI Act and GDPR
2

Describe the rule

Give the rule a name and a description. A clear description helps other admins understand what the rule enforces and why.
New compliance rule dialog with name and description fields
3

Set the severity

Choose how serious a failure of this rule is: Low, Medium, High, or Critical. Severity helps you prioritize which failures to review first.
Severity selector with the options Low, Medium, High, and Critical
4

Choose what the rule checks

Under Applies to, select Agents, Skills, or both.
Applies to control with Agents selected and Skills available
5

Define the failure condition

Define the condition that makes the check fail. Governance evaluates each agent or skill in the rule’s scope against it.
Failure condition field with an example rule that fails agents exposing personal data
6

Run it on existing resources

Optionally, select Check existing resources now. The rule then runs on the agents and skills in the selected scope as soon as you create it. Otherwise, agents in that scope are checked when they are created or when their next version is published. Skills in that scope are checked when their content is saved.
Check existing resources now checkbox with the count of agents and skills in the selected scope
Match the severity to the action you would take. If a failure would make you disable an agent immediately, the rule is Critical. If it is something you would mention to the editor at the next review, Low or Medium is appropriate.

Rule analytics

Each rule has its own analytics view that shows how the rule behaves over time. When the rule applies to both agents and skills, switch between Agents and Skills. A rule that only checks skills opens the skill view.
  • Rule runs over time: how often the rule ran, and how many agents or skills were checked.
  • Flagged agents over time or Flagged skills over time: how many failed the rule, so you can see whether compliance in your workspace is improving.
  • Rule cost over time: what running the rule costs.
In the skill view, the chart labels are Skills checked and Flagged skills. When the selected range includes checks that skipped files, the view also shows how many of those checks were partial.
Skill analytics for a compliance rule with Skills checked, Flagged skills, and a partial check count
API costs for rule executions are shown for transparency. They are currently not billed to your workspace and do not count toward user usage limits.