Rovo Agent remembers too much — how do you control context between Automation runs?

[Aditya Srivastav]

Hi everyone,

I’m experimenting with Rovo Agents in Jira Automation and ran into something I’m not completely sure how to handle.

I have an automation that calls the same Rovo Agent whenever an issue moves to a particular status. The prompt is fairly simple and includes the issue details, but I noticed that the agent’s response sometimes seems to be influenced by what happened in previous executions.

For example, if the agent processed Issue A first and then Issue B, the response for Issue B can occasionally feel as though some context from the earlier execution is being carried over.

I was expecting each automation execution to be completely independent unless I explicitly provide previous context in the prompt.

So I’m trying to understand:

• Does a Rovo Agent invoked through Jira Automation maintain any conversation/session context between executions?
• If it does, how is that context scoped — per rule, agent, user, or issue?
• Is there a recommended way to force every automation execution to start with a completely fresh context?
• Are there any differences in context handling between Use Rovo and Use Rovo Agent?

I couldn’t find a clear explanation of this behavior in the documentation, so I’d be interested to hear how others are handling it.

[Robin Surland]

You have to add strict constraints to your agent’s  prompt to explicitly forbid referencing prior chats like this:

"Treat every new session as a completely independent interaction."
"Do not assume, infer, or pull context from any past conversations, user history, or prior sessions."
"If information is missing from the current message payload, ask the user for it instead of guessing based on previous interactions."