What Rovo Agents have you created?

[Jason Krewson]

Today I wanted to start exploring what Rovo Agents others have built and found valuable within their organizations.

I haven’t built a lot yet myself. Right now, I’m focused on creating simple easy to follow Rovo documentation that points users to relevant Atlassian resources if they want to go deeper, along with some recommended training materials.

I’ve also created a couple of Confluence pages that walk through the different ways users can leverage Rovo Chat to create Jira work items and build or update Confluence content. On top of that, I’ve been promoting Rovo through our monthly newsletter, May’s edition was heavily focused on Rovo, and June’s will be as well, I have been calling it the Rovo Takeover to help drive adoption. I did something similar around this time last year.

As for Rovo Agents, the only one I’ve built so far is a “Jira & Confluence Guide” agent. It helps answer user questions by pointing them to the most relevant internal Confluence documentation, while also including links to official Atlassian docs. It also lets users know how to reach out to me directly if they need additional help.

If anyone has ideas for useful Rovo Agents please share as I’d love to hear about them and I'm sure others would benefit as well.

[Jason Krewson]

I just found this shortly after creating this post and wanted to share, 3 great ideas for Rovo agents. 

3 Rovo Agents That Make Your Life Easier! (It's a Reddit Post) 

[Jason Krewson]

I made the "Change My Mind" Rovo agent that is in the reference link in my comment above but updated it a bit as it would randomly look for something to review if you didn't provide something, and that really shouldn't be how it works, instead it should ask you for the content you want reviewed. 

I'll also say its fantastic, give it a Confluence page, it will surprise you with its feedback.

Below is the breakdown for it if anyone wants to create it for their organization also. 

Name

Change My Mind

Description

This agent is designed to stress test your content by actively challenging it. No matter what you provide, a Confluence page, idea, or proposal it will always find something to question, critique, or improve.

Instructions

Agent Behavior:

You are a debate expert whose primary job is to thoughtfully evaluate and challenge the provided content.

You consistently take a critical stance, offering meaningful and well-reasoned arguments that push for improvement. Your feedback is direct, insightful, and always aimed at strengthening the content.

Core Instructions:

• Focus on evaluating and questioning the provided content

• Always find something to challenge — the current state is never acceptable as-is

• Provide clear, valid, and logical arguments

• Be specific and detail-oriented, actively nitpicking weak points or gaps

• Do not ask open-ended questions — instead, give direct recommendations for improvement

• Never accept the content as complete — there is always a better version

• Maintain a confident stance and do not back down from your critique

• Communicate in a provocative but friendly tone

• Operate with the mindset: "agree to disagree", staying respectful while being challenging

• Do not proactively search for or analyze any external or internal documentation until the user has explicitly provided the content or a specific URL they wish to have reviewed. If the user has not provided content in their initial prompt, your only response should be to confidently and directly ask them to provide the material they want challenged.

Expected Output Style:

• Opinionated and assertive

• Constructively critical

• Actionable recommendations (not vague feedback)

• Focused on improvement, not validation

Skills

Get page

Knowledge

All organizational knowledge

Conversation starters

• What I wrote here is the best thing ever...change my mind
• I think this is ready to share… prove me wrong.
• Find the weakest part of this and tell me how to fix it.

Web search 

This is disabled.

[Sang Park]

Would you mind sharing some details about your “Jira & Confluence Guide” agent?
TIA

[Jason Krewson]

Absolutely, see below for “Jira & Confluence Guide” agent. Note that I remove company specific information and put in like "<<Insert company name here>>" instead. 

---------------------------------------------------------------------

Name

Jira & Confluence Guide

Description

Answers Jira and Confluence questions by guiding users to the most helpful documentation in the <<Insert company name here>> Atlassian Confluence space and Atlassian Documentation.

Instructions

You are a helpful virtual assistant that supports <<Insert company name here>> users with Jira and Confluence questions.

Your primary goal is to provide clear, accurate, and easy-to-understand answers while prioritizing <<Insert company name here>> internal documentation.

Always start by searching the <<Insert company name here>> Atlassian Confluence space: <<Insert Confluence Space URL here>>

When responding:

• Begin with information from <<Insert company name here>> Confluence whenever possible

• Provide a clear explanation based on the internal documentation

• Include direct links to the most relevant page(s)

• If multiple pages apply, include a short summary of each with links

If internal documentation is incomplete or does not fully answer the question:

• Supplement the response using Atlassian Documentation and Atlassian Community content

• Clearly label these as external sources (e.g., "Atlassian Documentation" or "Atlassian Community")

• Use external sources only to enhance or clarify, not replace internal guidance when it exists

You may use web search when needed to improve accuracy or fill gaps, but always prioritize <<Insert company name here>> Confluence content first.

If no relevant internal documentation is found:

• Provide the best available answer using Atlassian sources

• Recommend contacting support

Always include this message at the end of every response: "For additional help or questions not covered here, feel free to reach out to the Atlassian Team at <<Insert your teams email distro here>>. They’re happy to help!"

Tone and style:

• Friendly, patient, and easy to understand

• Avoid technical jargon unless the user uses it

Skills

• Get page views
• Get page
• Find source of truth
• Find owner of topic

Knowledge

• All organizational knowledge

Conversation starters

• What Jira Training Recommendations do you have?
• What User Guides are available in the <<Insert company name here>> Atlassian Team Confluence space?
• How to Delete Jira Cards?
  • This is a good one specifically for us as we have a 3 cool down on work item deletion with Jira automation so some users are like how do I delete a card.

Web search 

This is enabled. 

---------------------------------------------------------------------

Also I am thinking about making a Jira Space Analyzer, basically a bot that can review your Jira space, boards, backlog, sprint, in progress work, and provide like a summary of that. 

Reworded by AI: 

This agent focuses on providing a clear, high level view of a Jira project by analyzing work across the backlog, active sprints, and in progress items to surface key insights. Instead of manually reviewing multiple boards, filters, or views, the agent summarizes priorities, active work, blockers, risks, and recommended next steps in one place.

[Sang Park]

Thank you!

[Keyna Hendrickson]

We have two that we use regularly: a Pre-Bug that we use to sus out UI "issues" that are really user issues, and not software bugs. It helps us rule out user error before we raise a bug ticket. And we have a Release Notes writer that produces release notes per deployment in a consistent manner. These are real time savers for both our dev team and our quality team. 

[Manuel Alexandre Sousa Pereira]

You should check out this post - 🏆 Vote for your favorite REAL Rovo demo, and help... - Atlassian Community

I've developed several tools to help our team work more efficiently:

- Jira Flow Watcher: Helps the team identify stale work.
- User Story Formatter: Assists the team in creating tickets in the proper format.
- Sprint Agent: Generates daily reports on sprint progress.
- Industry Pulse Agent: Provides real-time pulse checks on various industries and niches.
- Ticket Timekeeper: Aids the team in managing and logging time spent on tickets.

I believe the game changer is using Rovo Agents with Jira automations, sending notifications in Teams/Slack and saving outputs in Confluence for future memory.

[Jason Krewson]

Awesome, thanks so much for sharing what you have created and that link. 

I am really interested in "Sprint Agent" and know a team that would start using that the moment I create it.

Can you share your details on this one?

[Manuel Alexandre Sousa Pereira]

Sure, happy to help:

Instruction: 

You are an assistant focused on preparing friendly and concise recaps for daily standup meetings in the (Project Name). Link - (Jira project link)

Your main goal is to help users quickly review and share what was discussed, including key updates, blockers, and next steps, in a way that's easy for management to understand.

Organize information by team member or topic as appropriate.

Highlight any blockers or urgent issues clearly.

Use a friendly, approachable, and concise tone.

Keep summaries brief, clear, and suitable for quick updates to management.

Skills

Message chat
Message channel
Search fields
Search Jira projects

Knowledge

Project Jira Space
Project Confluence folder

Conversation starters

What were the key updates from today’s (Project Name) standup?
Can you summarize any blockers discussed in the (Project Name) meeting?
Give me a quick recap of today’s (Project Name) standup.

Web search - Off

Jira Automation

[Jason Krewson]

Wow thank you so much!

[Jason Krewson]

I’m seeing many users create agents just to generate work items, even though Rovo Chat already does this very well. If you’ve built one of these agents, what’s the added value versus directing users to use Rovo Chat instead?

[Manuel Alexandre Sousa Pereira]

Hi again, Imo the added value is to format any ticket/issue based on your own template - this template can be in Confluence for example.

I've seen an Rovo/Atlassian webinar where they do this but with a Rubric Map(w/ flags), stored in Confluence. So you would invoke the general Rovo Agent, ask him to make a review on your ticket based on that rubric. The output is a assessment of what is good, bad, needs improvement, so that the human can make the necessary changes.

webinar link - Beyond the Agent: Design Patterns that scale AI across your Organization - Zoom

[Jason Krewson]

That make sense to me, thank you!

[Frank Sherwood]

I've been creating a pair of pretty ambitious agents for techwriting in a software company. The goal is to get draft language for doc updates from the Jira issues defining the software changes.  If Product Managers and Developers are disciplined enough at describing the work accurately in Jira, then this should be a productive approach. 

The task currently is split into two agents with a markdown handover document, rather than being subagents - because the work scope challenges context limits. 

Though that brings up a sore point with Atlassian: It's been impossible to find clear answers on LLM models or context capacity.  It's all vague "we choose from multiple models with varying capacities depending on the task" kind of answers. Which makes it VERY difficult to plan or assess.  All I can do is guess by results behavior.  If instructions get randomly skipped a lot, that's a sign of context rot.  Condense or partition the work.  Also I have no accounting metrics to suggest how much each of these runs is costing.  

The first agent creates a Document Change Assessment, by reading a Jira Epic and its children, summarizing the functional impacts to users or implementation engineers, then matching those to relevant existing user and implementation documents to suggest what topics within which documents need change or additions. 

I've uploaded my existing PDF documents to Confluence pages in a folder. This gets the content into Atlassian's RAG index of tenant content, so the agent can identify related doc content without having to read the entire documents. This is key, because otherwise none of this would be possible within any realistic context capacity. 

The output of this agent is a markdown Document Change Summary, organized by document and change topics. Each suggested change has a tag for human-in-loop to approve or reject the change. Each change lists the impacted document and topic, the specific Jira issues defining the change, and a brief summary of the change.  Sufficient for a human reviewer to assess.

After editing for approve/reject, that document is fed in to the second agent to suggest actual draft update language in context and following existing document style.  That Doc Draft agent is pointed to the same Epic and to the Confluence folder of docs.  Based on the approved change summaries, it rereads the targeted documents in finer detail, rereads the specific Jira issues, and proposes draft update language targeted to specific context within the docs. 

The results are sometimes awesome. I can just copy and paste into my existing docs, in the targeted locations. And sometimes they're not. My struggle right now is to get consistent reliability.

The instructions for each agent are quite lengthy. I'm spending a lot of time with various AI's including Rovo to assess and improve those instructions.  It's being a great lesson in asking AI to improve your prompt for AI.

Most recently I've applied a "skill" prompt called "Grill Me". It assesses a plan and "interviews" you (relentlessly) to probe all decision points and ambiguities. It thinks of things I'd never have considered by myself.  OTOH though, the end result has grown my instruction prompt too much, I fear. And I've spent literally all day considering questions and approving/rejecting suggestions. I think now I need to make a pass with an LLM to aggressively trim what isn't important.  I'm not sure whether I'll use GrillMe again.

This is still very much a work in progress.

[Jason Krewson]

That sounds awesome, hope you can get it reliable like you want!