This might be a trivial question, but my attempt to have an automation using a rovo agent to summarise steps on a work item into a KB article guide isn't outputing any of the content into the confluence page.
The steps are:
The page gets created, except there is nothing on the page.
I've reviewed the automation logs, there is output from the agent, but its not coming over to the confluence page.
Seems like something basic im missing?
I've looked at many Atlassian articles on how to use rovo agents and in automations, none of them help.
e.g.
How Do You Use {{agentResponse}} in Atlassian Auto... - Atlassian Community
Automating Rovo agents | Rovo | Atlassian Support
To resolve, the agent instructions (not the prompt in my automation) required adjustment.
From:
Create a well-formatted Confluence page with the structured guide containing
To:
Create a well-formatted Knowledge Base Article with the structured guide containing
Small detail, but its what resolved the issue.
Hello
Can you provide the details of the Use Agent step, please?
Are you triggering this rule while viewing a Jira issue?
Before the step to create the Confluence can you add a Log action to print the agent response smart value to the rule execution log? Then try executing the rule again and share the audit log output with us.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Agent action:
Are you triggering this rule while viewing a Jira issue?
Yes, when viewing a JSM work item
Log action only shows "Log": when attempting to log the {{agentResponse}} response.
I've also logged a support ticket and will track and place an update on this.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Thank you for the additional information @Phillip C
I think that the problem is that your agent doesn't know what issue you want it to reference.
Your prompt says "the ticket". I think if you try replacing that with {{issue.key}} it will then know which ticket to reference and your agent response will not be empty. That is what worked in my simple test.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
thanks Trudy for the help. But no luck with using the {{issue.key}} smart field. I do like that approach though for improving my prompt.
The log action still only shows 'log', nothing else. the debug agent response shows a good amount of data indicating it's done exactly what i needed, but it's not outputting.
I've got my support ticket still open, so waiting on their response!
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Can you share with us the new version of the prompt? And if you changed the rule in any other way provide a new image showing the rule. And again show the log output.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Also please show the comments of the issue for which you trigger the rule so that we can see that there are Internal comments on the issue.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hey @Trudy P Claspill ,
just to close the loop on this, the fix was to update the agent prompt.
I changed a line in it from:
Create a well-formatted Confluence page with the structured guide containing
To:
Create a well-formatted Knowledge Base Article with the structured guide containing
Silly mistake, but good to know and be on the lookout for moving forward! Appreciate your help :)
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
I'm glad you found the solution!
If my responses helped you, please consider clicking the Accept Answer button above my responses.
If you found that the solution was not related to any of my suggestions, I encourage you to add your own Suggest an answer response with the actual solution and mark that one as Answer Accepted.
Clicking the Answer Accepted button marks your Question as Solved and is very helpful to people searching the community when they are searching for solutions.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi @Phillip C - I'm having the same problem. Your solution seems quite inexplicable to me. Do you know if the output of the agent was plain text or used rich text formatting for each version - i.e. the initial version (Confluence page) and the version that worked (Knowledge Base Article)?
I have an agent that produces output that includes tables and other formatting. I have found that smart variables with anything more than plain text get their content stripped by the Publish New Page action. This is poorly documented by Atlassian and quite problematic for achieving various use cases.
I'm hoping that you can provide more detail about what the output of your agent looks like in the {{agentResponse}} smart value variable. I would love to be able to get my simple automation to create a page successfully from the output of my agent.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
update: got it working, but no idea how. Maybe filling empty cells in a table with "--" helped? (I saw some issue with empty cells written up elsewhere)
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Since that time I made some other changes and then was getting failures again on creating the page with the Publish New Page action using the {{agentResponse}} smart value variable from my agent.
Running the agent manually worked perfectly. Running the agent in the Automation did not work. Adding a Log action showed no response from the agent, but a final step in the Agent log in the Automation Audit log showed Create Page producing what looked like the right content composed into a JSON structure in the "content" field. It turns out this does not feed into the {{agentResponse}} variable for some reason.
The solution I found was to remove the Create Page skill that I had added to the Agent. Now the agent doesn't create a page when run manually, however it creates the content for the page. Then when used in the Automation, this content is successfully passed via the {{agentResponse}} variable to the Publish New Page action.
I thought that at some point I had been able to run both via Automation and Manually, having the Agent successfully create a page in both use cases. But I don't know exactly what everything looked like at that time anymore, as I've only been working on this once per week. Each week I've made some adjustments and successfully made my page, but looking back I don't think I've always checked both Automation and manual methods of running the Agent. I think I've been assuming if it worked manually then it should work in the Automation. I was wrong. I don't know if there's a way it could be written to work both ways, but for now I just need the Automation to work, so I'm happy to sacrifice having the manual operation successfully create the page (even though it can create the content in both operation modes).
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
neater summary:
When the Agent is run in Automation with the Create Page skill attached, it creates output in JSON format. When the Agent is run manually, this allows a Confluence page to be created. But in the Automation, the JSON from the Create Page skill appears to be stripped from populating {{agentResponse}}, which is then returned empty, meaning nothing arrives in the content of the Publish New Page action (or any other actions using the {{agentResponse}} variable.
It would be nice if Atlassian could allow the output of Create Page from the Agent to populate the {{agentResponse}} variable, and allow a Publish New Page action (or similar) that will accepts this JSON format that the Agent execution environment clearly accepts. This way we could have an Agent that can be used manually or via Automation to successfully create and publish a new Confluence page. If there is already a way to do this, I haven't yet found it.
Essentially, the agent environment and the automation environments are not harmonised. This seems like a bit of a flaw when getting real business value out of Agentic capabilities require a cooperation between pure agentic and deterministic workloads. Sure, the agent has deterministic skills that seem to work well, but the converse situation - use of agentic elements within the deterministic Automation environment - seems more stunted.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.