Scenario Troubles

[Mike Stone]

I have an agent that has two scenarios that sometimes execute instructions from both scenarios in one session when it should only be executing the instructions for one scenario.

Debugging a bit more, it appears it switches out of the correct scenario into the other scenario and merges.

[Ryan Boyd]

What are the triggers and scenarios? if you can share. 

[Mike Stone]

The agent assists with investment requests. 

• Scenario A is for internally funded requests. This is the default scenario.
• Scenario B is for customer funded requests.

The user is presented with two conversation starters:

I have an internal R&D and/or product investment request.

I have a customer SOW/PWS deliverable I need development assistance with.

Scenario B trigger instructions:

Trigger this scenario when a user has a request for their customer and has a PWS or SOW.

## Positive examples:
This is a customer request or need.
There is a PWS or SOW.

## Negative examples:
This is an internal investment request.
This is a product or R&D request.

[Mike Stone]

What appears to happen is the user selects the conversation starter below and the agent executions instructions correctly for Scenario B.

I have a customer SOW/PWS deliverable I need development assistance with.

When the agent gets near the end of those instructions it then starts executing Scenario A instructions when it should finish and submit the request instead which are the last instructions for Scenario B.

[Ryan Boyd]

Ah. What do the debug logs say for that one? 

[Mike Stone]

Where can I see logs?

[Ryan Boyd]

https://community.atlassian.com/forums/Rovo-articles/Peek-Behind-the-Curtain-New-Debug-View-for-Rovo-Agents-in-Chat-%EF%B8%8F/ba-p/3132492

^^ has the pointers on where to see it. That should also show you directly which scenario is running. 

[Mike Stone]

Fantastic. I will report back. Thank you!

[Mike Stone]

After reviewing the logs, it said it decided to switch to the other scenario when a user answered TBD to a question. I need the agent to stay in the current scenario and not switch out.

[Mike Stone]

After some research this sums up the problem.

"It sounds like you’re hitting the "hallucination loop" or "intent drift"—a common frustration in Rovo Studio (2026) when an agent’s internal reasoning (the LLM) decides another scenario is a "better fit" mid-stream."