Rovo Agent – Output Format Not Respected After Scenario to Subagent Migration

Question

After the forced migration from scenario-based to subagent model, my Rovo agent is no longer respecting the output format defined in its instructions. Specifically, it is not returning results in the required 7-column table format.

EXPECTED BEHAVIOUR (working correctly before migration)
The agent returned a table with exactly 7 columns in this order:
1. JIRA Issue key
2. Summary of the issue
3. Relevance Percentage
4. reason why you think it is related issue
5. fix on the issue done
6. Root Cause
7. Status

ACTUAL BEHAVIOUR (after migration)
The agent is consistently returning a simplified table with only 4-5 columns, with renamed headers such as "JIRA Issue", "Summary", "Relevance", "Resolution", "Connection to [issue]". The columns Fix Applied, Root Cause, and Why Related are either dropped or merged into other columns.

Anyone having the same and recommendation how to way around it - Already consumed all possible instruction format recommendation to try fix the instruction however nothing work.

Our agent is being used across teams and now it no longer what we wanted to work.

Dr Valeri Colon (Connect Centric) · Answer

@May Rosales your testing is valuable because it demonstrates the model can produce the correct 7-column output but is not consistently applying the instruction after migration. The fact that the formatting works more reliably when placed in the Trigger field suggests instruction prioritization may have changed between the scenario-based and sub-agent models. This appears less like a prompt-writing issue and more like a behavioral change in how the new agent architecture interprets instructions. I'd recommend raising this with Atlassian Support as a potential regression.

Rebekka Heilmann (viadee) · Answer

Hi @May Rosales - welcome to the Community.

Very interesting (and frustrating).

The switch to Sub Agents should not have done anything to the logic behind it. It was more of a name change than anything else. The only other thing that was changed is that the Agent itself doesn't have a behaviour anymore, but instructions and it's sort of the default now.

I've flagged your question for the Atlassian Team.

May Rosales_trhvhw · Answer

I used the manual setup. My agent was originally built using the
scenario-based model and it currently deploy in production, and our teams are already using it and was working perfectly with the exact 7 column output format I specified in the instructions.

Now I need to do some updates to our agent, so I duplicate the production agent and notice that scenario base is become sub-agent, and the output is now incorrect, I have tried rebuilding the instructions from scratch manually multiple times with no success. Interestingly the format rules work better when placed in the TRIGGER field rather than the INSTRUCTIONS field, which suggests the new model processes these two fields differently, or gives the instructions field less weight when determining output format.

Did the agent builder guide handle output format instructions differently? Would love to know if there is a recommended way to enforce strict table formatting in the new subagent model.