Agent Behavior

This section helps you configure not the agent interface, but its behavior: how to respond, when to look for documentation, when to show an element on the page, and when to submit a question to support.

These materials are useful for project owners, instruction editors, and developers who connect tools. The section describes not one cabinet screen, but response rules: role boundaries, safety, information search, and handoff of complex questions to support.

What to configure

• Agent role - boundaries of responsibility and response style.
• Safety Rules for Answers - what cannot be promised, done or disclosed.
• When to send a user to support - signs that it is better for the agent not to continue on his own.
• Search dictionary - words and synonyms that help connect questions with the right articles.

How to write rules

First describe the agent role in simple language: who it helps, which tasks it solves, and where its boundaries are. Then add safety rules, knowledge sources, and escalation signals. If interface highlighting is needed, use exact data-ai-context-id values from documentation instead of free-form descriptions.

What to check after a change

After changing behavior, test several real questions: a simple documentation question, an access-limited question, a question about a dangerous action, and a question where an element needs to be shown on the page. This helps catch cases where answers start promising more than the system has actually done.