How to configure the Use agent action
Agents can be used in automations to perform complex tasks and simplify flows.
Using an agent within an automation flow allows the agent to take actions directly in any app configured in the agent, like updating work items, creating pages, or posting comments.
There are two ways to do this: add a trigger to an agent's configuration, or add the agent to an automation flow.
For agents to work autonomously, they’re best set up with their own identity and managed by organization, space, or app admins, just like users.
Here’s some tips to help you configure the Use agent action with confidence.
Provide proper context in your prompt
After you select the agent you want to use, you need to write a prompt for it. Each time the flow is triggered, this prompt gets sent to the agent on your behalf.
To provide the agent with appropriate context, mention the object from the trigger. For example, if the trigger is Work item created, mention “work items”.
Your prompt might be:
Review the description of this work item and add a comment if it's missing acceptance criteria.
You can also add {{smart values}}, which may generate a more predictable response:
Review the {{issue.description}} of
{{issue.url}}and update if it's missing acceptance criteria.
Smart values are dynamic variables that make your flow more flexible. Each smart value is written as a hierarchy, starting with a top-level object and followed by properties of that object, {{object.property.subProperty}}.
In this prompt example, when the flow (and the agent) is triggered, the URL of the relevant work item is provided to the agent as context. The agent will analyze the work item's description and take action directly, updating the field if needed, for example.
How agents take actions
Agents in automations can now perform write actions directly. They're no longer limited to generating text output for a secondary action. The agent can update fields, create work items, post comments, and more, all within the same step.
The tools available to the agent are listed by the Tools and connections section of the step. This shows which Atlassian apps and Connected apps (third-party tools) the agent can use. If a third-party tool isn't connected yet, select Connect to authenticate it.
Choose how agent's access apps
When you configure the Use agent action, you'll see the agent either has a user’s identity or it’s own identity.
Requesting user adopts the identity of the user who’s interacting with or automating the agent (potentially, you!). The agent will have all the same permissions and app access as that user.
Agent account sets up the agent with it’s own account identity, permissions, and access that are managed by an organization, space, or app admins (just like a user).
Understand Rovo agent accounts
Existing automations will stay as is
When you set up the Use agent action, you’ll spot a toggle to Let this agent take actions. This allows the agent to use tools that create, edit, and delete content.
This is off by default (including exisiting automations) and can be toggled on when you’re confident in the agent’s actions.
Use the smart value {{agentResponse}}
The smart value {{agentResponse}} captures an agent's text output and passes it to downstream steps in your flow. This is necessary when the agent isn’t set up to take actions.
This is useful when you want to use the agent's response as input for a subsequent action. For example, posting a summary as a comment or sending it in an email. It inserts a uniquely generated response each time the flow runs.
To use this smart value when you set up an action, just use {{agentResponse}} in a subsequent step in your flow.
Formatting of the {{agentResponse}} defaults to rich text markdown, but can be changed by adding properties.
Debug agent actions with the audit log
You can use the automation audit log to investigate what the agent actually did when the automation ran.
If you're the agent's creator or an admin, select the Use agent action, then More actions (), then Debug response.
You’ll see a detailed breakdown of the agent's actions, including which tools it used and what it changed.
Was this helpful?