nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Ask Rovo vs Rovo Agent

Vikrant Yadav

Hi Expert

I need your help fixing a Rovo Agent response issue.

I have created a prompt that includes approximately 40–50 JQL queries and filters. When I paste the prompt into Ask Rovo, it takes around five-ten minutes to generate the response, but the desired report is generated correctly.

However, when I use the same prompt in a Rovo Agent, it fails to generate the complete response and often produces only a partial report or hallucinates information.

I cannot use Ask Rovo for this requirement because I need to schedule an automated flow that runs periodically, summarizes data from 40–50 JQL queries, performs the required calculations, and creates a Confluence page in a specific space.

How can I achieve this using a Rovo Agent or another Atlassian-supported approach? I need a reliable solution that processes all the JQL queries, performs the calculations accurately, generates a complete report, and automatically publishes it to Confluence.

Accepted answers

All comments

Arkadiusz Wroblewski

Hello @Vikrant Yadav

For 40–50 JQL queries, I wouldn't rely on Rovo Agents for precise calculations. There are already reports of incorrect counts. (ROVO-396) and mathematical inaccuracies (ROVO-818).

For reliable scheduled reporting, use Jira Automation or the REST API to collect and calculate the data, then publish it to Confluence. Rovo can summarize the verified results.

Keep in mind that Automation's Lookup work items is limited to 100 results.

Best,

Arek🤠🧙‍♂️

Vikrant Yadav

Hi @Arkadiusz Wroblewski But Ask Rovo is working fine, summarizing and fetching all filter result and giving me desired result and also publishing a confluence page.
40-50 are Jira filters.

But I want to sechedule it a bi-weekly automation, rather then doing it manually.
So i tried to create an Rovo agent with same set of instruction but Rovo Agent fails to generate desired result.

Arkadiusz Wroblewski

Does it also fail when run manually, or only through Automation?

Check your Agent's reasoning tier: Quick answers allows 10 loops, while Think deeper allows up to 60. With 40–50 filters, this may explain the incomplete responses. Change it under Studio → Agents → Details → Reasoning.

Studio Automation supports Scheduled → Use agent. Ensure the agent has the required Jira and Confluence tools enabled.

Does it also fail when run manually, or only through Automation?

Atlassian documentation

Vikrant Yadav

When asked manually agent output is truncated, and most of the time, the counts in the metrics are incorrect. The agent tends to hallucinate figures and doesn't consistently follow the instructions, even though they’re clearly defined.

In Jira automation , Agent generates nothing.Blank response

While Ask Rovo generate a full report without add

Mihai_LeanZero

Hi @Vikrant Yadav,

The blank response in automation matches ROVO-1148 (https://jira.atlassian.com/browse/ROVO-1148). When the agent called a tool, like creating your page, the run ended with no text and {{agentResponse}} came back empty.

It's marked Fixed, and the fix is the "Let this agent take actions" toggle on the Use Rovo agent action (https://support.atlassian.com/cloud-automation/docs/tips-to-configure-the-action-use-rovo-agent/). The toggle is off by default, existing rules included, so switch it on.

Atlassian said on 6 Oct the rollout would take about two weeks. If you still don't see the toggle, the rollout hasn't reached your site yet.

Long responses can still get truncated in automation (https://jira.atlassian.com/browse/ROVO-850), so keep the agent's part small.

For counts you can rely on, take them out of the model:
1. Create a scheduled rule.
2. Per filter, add Send web request to your site's /rest/api/3/search/approximate-count with the body {"jql": "filter = <id>"} and a Hidden Basic Authorization header from an API token.
3. Save {{webResponse.body.count}} with Create variable.
4. Put the variables in the prompt and let the agent create the page.

That recipe won't fit 40-50 filters in one rule under the 65-step limit (https://support.atlassian.com/cloud-automation/docs/automation-service-limits/), so I'd split them across two rules that each publish their half. I haven't tested the split.

I got 417 on a test site for a filter that a paged count also put at 417, so there's no 100 cap.

The endpoint is documented as an estimate that can miss very recent changes, which is fine for a bi-weekly report. The counts follow the token user's Browse permission.

Vikrant Yadav

Hi @Mihai_LeanZero Thanks for sharing the details!

In Rovo Agent I have enabled "Think Deeper" and using AI model "Gemini Flash 3" now it's generating a report.
Yesterday i have tried to use the same agent in my automation, it works and create page.
But today the same automation fails to generate report.
Sometime it works fine, and sometime it fails to generate report.

Why result is not constant ?

Mihai_LeanZero

Hi @Vikrant Yadav ,

The changing result is a known, tracked behaviour of agents in automation. It probably isn't your config. ROVO-592 says these rules "are non-deterministic and can fail for transient, recoverable conditions (rate limiting)", and its workaround is a rerun.
https://jira.atlassian.com/browse/ROVO-592

The audit log of today's failed run is the place to look. If the same data passes on a rerun, it's transient, the ROVO-852 pattern.
https://jira.atlassian.com/browse/ROVO-852

A generic "Agent invocation failed" while other rules or agents run at the same time points to a rate limit. ROVO-968 notes the log won't say so.
https://jira.atlassian.com/browse/ROVO-968

A failure near the 15-minute mark points to the automation timeout instead.

On top of the split from my last reply, I'd stagger those rules by a few minutes so they don't hit the agent at the same moment. If you want reruns without a blank page, let the agent return the report text and add a Create page action after it, behind a condition that {{agentResponse}} is not empty. With the agent creating the page itself, a condition after it comes too late. That makes failures rarer. I haven't found anything that makes it fully deterministic.

Could you paste the failed run's status and message next to yesterday's good one?

Welcome!

It looks like you're new here. Sign in or register to get started.

Quick Links

Find more posts tagged with

confluence

jira

cloud

rovo

rovo-search