System instructions
System instructions are the core prompt for the agent. They define the agent’s role, boundaries, tone, escalation rules, and how it should use knowledge or tools.
What you can do here
- Write or paste instructions in the Profile tab editor.
- Use Markdown-style structure for roles, rules, and examples.
- See validation errors when required instructions are missing.
- Use Enhance with AI to build or improve the prompt from requirements.
How to write instructions
- Open the agent in the builder.
- Go to Profile.
- Find Instructions.
- Describe the agent’s role and responsibilities.
- Add boundaries, escalation rules, and output expectations.
- Save the agent.
Field behavior
| Field | Required | Accepted values | What it controls |
|---|---|---|---|
| Instructions editor | Yes for most publishable agents | Markdown-style plain text | The system prompt used for every response the agent gives. |
| Enhance with AI | No | Button action | Opens the AI instruction generation dialog. |
Good instruction structure
- Start with the agent’s job.
- State what the agent should do.
- State what the agent must not do.
- Define when to ask clarifying questions.
- Define when to escalate to a person.
- Specify the expected answer format when consistency matters.