Studio Agent / Automation ignores Confluence instructions

[Damiano Di Battista]

Hi community,
I created a Rovo AI agent to categorize incoming alerts in my Jira instance.
The goal is to route each alert to the most appropriate technical team based on its content.

I configured the agent with specific instructions and added an automation rule that triggers the analysis when a ticket is created.

Now I need to apply some overrides to this logic.
For certain servers, I want every alert — regardless of its content — to be assigned to a specific team.

To achieve this, I created a Confluence page where I listed the servers and their forced routing:

Example:
server01 – DBA_Oracle
server02 – Network
server03 – Infrastruttura

Then I added the following instruction to the agent:

Mandatory Rule (Override)
You have access to a Confluence page named "Forced_Route".
Before classifying an alert, extract the hostname from the ticket summary or description.
If the hostname is present in the Forced_Route page, you MUST return only the forced_category and ignore all other rules.

I tested this by opening a ticket related to server02, with an alert clearly belonging to the Oracle category.
However, the agent still returned DBA_Oracle instead of Network (as defined in the forced routing).

Questions / What else can I check?

• Am I missing any configuration step in Studio or Jira Automation that could prevent the agent from reading the Confluence page correctly?

• Could there be formatting issues (spaces, non-breaking characters, uppercase vs lowercase) that prevent matching?

• Are there limits on how the agent reads Confluence pages (text length, table formatting, or special characters)?

• Is the forced override rule applied correctly if it comes after other classification logic in the prompt?

• Any best practices for ensuring absolute overrides for specific hosts?

[Rebekka Heilmann _viadee_]

Hi @Damiano Di Battista 

How is your list formatted in Confluence? I'd recommend a table as you then can clearly state if column 1 matches, take info from column 2.

Rovo CAN read tables and understand their structure. Just created Epics from a table and it did exactly the right thing with the right columns. In this case it also needs to search though and i think that could be an issue.

I am not sure how the "search" feature on a Confluence page is implemented at the moment but it might be an issue that it's too fuzzy. A classic RAG search would find trees if you searched for forest. However I think for the "look for information on the page" implementation, Atlassian is using a different algorithm.

I'd also structure your prompt slightly different. There are several frameworks that work, but explaining the page structure (or table) and then giving an example "For example, if the host is xyz you should find the value abc in the table and give out xxx as a result".