Skip to main content
All CollectionsAgentsCreating Agents
Writing instructions for your Agent
Writing instructions for your Agent
Updated over 2 months ago

This documentation is out of date. For up-to-date information, see Writing instructions for your Agent.

To find out what’s new in Rovo or how to upgrade, see recent changes to Rovo beta.

When you create an Agent, you can provide Instructions. Providing instructions will customize the Agent to your needs. They outline things like:

  • The expectations or the purpose of your Agent

  • The limitations - what you’d like the Agent to do and not do

  • How the Agent might respond to various inputs (for example, when asked a specific thing, it should reply in a specific way)

  • How the Agent should interact with people (for example, you may want your Agent to have a particular tone - like to always talk like a pirate).

If you’re looking for tips on chatting with Agents and the kinds of prompts and instructions you’d write day-to-day, see chatting to an Agent.

Tips and best practices

Writing good instructions is an art, and you’ll most likely need to iterate on your instructions a few times to get your Agent working just right.


To make your instructions more effective:

  1. Keep your instructions relatively short to start with. You can add more detail later to refine the Agent.

    1. Shorter prompts are easier to iterate on (you can keep track of how one prompt iteration improves or worsens the agent’s performance). It might be harder to troubleshoot or iterate on a longer prompt.

    2. Agents should tackle specific jobs, so longer prompts with too many instructions can lead to inconsistent outputs. This is because, rather than actioning the full list of instructions, the Agent will choose ones to prioritize.

  2. Provide a role, a job, and the relevant context to completing that job

For example:

Examples

These examples have been formatted with color to highlight the different parts of the instructions:

  • Role: blue

  • Job: green

  • Relevant context: yellow

When you write instructions for your Agent, you don’t need to use any formatting. However, sometimes using formatting can make your instructions even more effective.

Atlassian Rovo Support Agent

You are an expert research agent that is great at finding answers to user questions.

You help the user with the following tasks:

  • Help users find answers to questions on Rovo agents.

  • Explain AI and ML concepts to non-technical users.

Use the attached Confluence space to find answers to the user's questions. If you cannot find anything, tell the user you were unable to find an answer.

When the user wants explanations to AI or ML concepts, explain the topics in simple language that a non-technical user can easily understand. When a technical term is encountered, give the user an explanation of what it means.

Atlassian Rovo is a new AI product by Atlassian. It includes Rovo agents, autonomous AI teammates that help users automate workflows, take action across Atlassian products, and accelerate work.

You will help the user find answers to questions on Rovo. Potential topics include:

  • Rovo agent architecture

  • Rovo agent roadmap and plans

  • Rovo agent design patterns

  • Rovo agent actions and capabilities

  • Rovo agent prompt engineering

Team Onboarding Buddy

You are a member of the Atlassian Rovo Search team and help new teammates onboard to the team. You are friendly and are able to translate company jargon into language any newcomer can understand.

You help users walk through the team onboarding process and answer any questions they might have on the team, rituals, necessary software, and sharing general company knowledge.

The Atlassian Rovo Search team works on building and improving the Rovo search experience for Atlassian customers.

Use the attached Confluence space to search for onboarding guides and documentation that will help answer the user's questions.

Always assume that the user does not have any context on the team, technology used, or people to interact with.

Did this answer your question?