Skills are reusable instructions for handling a specific type of request. They guide the agent's approach and output, such as how to explain a variance, structure a report, or provide recommendations.
Use a Skill when you want the same reasoning pattern, checks, or response format applied consistently without changing which underlying data the agent can access.
When to create a Skill
Create a Skill when you need a consistent response method across questions. Use a Data Source to define accessible data; use a Playbook to define a repeatable multi-step procedure.
Skills control behavior, not data access. A Skill can tell an agent how to analyse or present an answer, but the relevant Specialist still needs the appropriate Data Source and permissions.
|
Need |
Skill can define |
|---|---|
|
Explain a spend variance |
Show amount, percentage, period, likely drivers, and unsupported assumptions |
|
Prepare executive spend summary |
Use headline, main changes, drivers, exposure, and actions |
|
Write a recommendation |
State finding, evidence, estimated impact, owner, and next step |
Create a Skill
-
Open Admin > AI Studio > Configure > Skills.
-
Select + Add Skill.
-
Enter Skill Name and Description. State the trigger conditions clearly so the agent can distinguish when the Skill should and should not apply.
-
In Skill Instructions, define the analysis or writing behavior, expected output, required checks, limits, and exclusions.
-
Review Expose to main Yarken agent. Enable it only when the Skill should be available for general Ask Yarken requests, rather than only through a narrower Specialist or workflow.
-
Save the Skill, then test representative requests in the supported agent experience.
Write instructions that produce consistent results
A useful Skill should specify the trigger, data checks, steps, output shape, and how to handle insufficient information.
Define:
-
Trigger: what request or condition should invoke the Skill.
-
Data checks: what values, periods, currencies, or source-backed facts must be present before analysis.
-
Method: how the agent should analyse or transform the information.
-
Output: the required structure or fields to return.
-
Limits: what the Skill must not infer or do when evidence is missing.
Example: Spend variance explanation
When asked to explain a spend variance, use the requested period and the available budget and actual figures. State variance amount and percentage. Identify the largest supported drivers. Distinguish confirmed causes from possible explanations. If source data cannot explain a driver, say what additional breakdown is needed. Finish with up to three practical next checks.
For a variance Skill, you can require fields such as scope, period, Actual, Budget, Variance, source, and missing-data notes. Do not infer unsupported root causes.
This instruction does not provide access to financial data. The relevant Specialist must have the appropriate Data Source and permissions.
Import and maintain Skills
The Skills page supports importing reusable configurations. Review imported names, instructions, and exposure settings before use. Update or deactivate instructions that no longer match current Yarken behavior or reporting conventions.
Important: Editing a shared Skill can affect Specialists, Playbooks, and Automations that depend on it. Test changes before using the updated Skill in production workflows.
Check results
-
Ask question that should invoke the Skill, such as Explain the quarter-to-date software variance for this cost centre.
-
Confirm the expected structure is present, the answer uses authorized data, and no unsupported figures or causes are invented.
-
Ask an unrelated question. Check the Skill does not impose its format when not relevant.
-
Refine the description or instructions if the agent applies Skill too broadly or omits required sections.
Related content