Skills
Skills are executable playbooks that tell Nousu how to handle specific request types, with guards and approval rules enforced by the platform.
Last updated
Skills are executable playbooks that tell Nousu how to handle specific request types, with guards and approval rules enforced by the platform.
Skills (under Agent → Skills) replace the old Learning & Evaluation dashboard and Policy Center. They are optional playbooks for repeatable workflows — returns, goodwill gestures, escalations, and other store-specific procedures.
Nousu works out of the box without Skills. Add them when you want your team's way of working encoded for specific scenarios.
A Skill is a structured document that includes:
Procedure — steps Nousu should follow in prose
Guards — hard rules checked before an action runs (order value, time window, product type)
Actions — connector operations such as creating a return or issuing a discount
Authority — what Nousu may do alone vs. what needs approval
The model interprets the procedure. The platform verifies every consequential action — approval never bypasses a failed guard.
Draft
Work in progress, not live
Testing
Try in the test environment
Active
Used in production conversations
Retired
Removed from the index, history kept for audit
Create a draft, iterate, test, then Activate. Major edits should go through Edit as new version so you keep an audit trail.
Multi-step workflows
Static facts and policies
Conditional logic ("if order < €100…")
FAQ answers
Actions that need approval
Opening hours, shipping times
Escalation playbooks
Product care instructions
Escalation rules now live in Skills instead of Integrations → Escalation rules. Define when Nousu should hand off to a human, what to tell the customer, and what brief the team receives.
Pair Skills with Slack notifications so the team sees escalations immediately.
Each action has an authority level:
Autonomous — Nousu executes when guards pass
Approval required — teammate must approve in the Inbox
Denied — action unavailable
Example: a Skill may allow refunds under €50 automatically but require approval above that — even if a teammate clicks approve on a €120 order, guards still block it.
Open Agent → Skills
Click New skill
Name it and write the procedure
Add guards and actions as needed
Test in Agent → Test environment
Activate when ready
Tip: Start with one high-volume workflow (standard returns, damaged-in-transit, late delivery) before covering every edge case.
Agent Trace — see decisions, evidence, and tasks per conversation
Inbox — where approvals land
Agent & Tools — enable/disable base tools
Last updated