Community resources
Community resources
Let's Build a Rovo Agent Workshop - August FAQs
đ Workshop recap
Mid-August the Technical Adoption Success (TAS) team ran our first Let's Build a Rovo Agent workshop, where we walked through how to automate the triage process using Rovo, Jira, Confluence, and the Automation platform. If you missed it, check out the recording here: [AI Workshop] Let's build a Rovo Agent.
đ Rovo permission updates
We had several questions around Rovo permissions across the workshops we ran, and have some updates to share! The ability to create agents can now be restricted to a group of users, allowing stricter governance over agents in your organization: đ IT'S HERE: Rovo Agent Governance + Create Permissions.
Stay tuned for more changes by following the Atlassian AI & Rovo and Rovo Articles pages on Atlassian Community!
đ Rovo Office Hours
Have other questions you didnât get to ask during the workshop? Register for Rovo Office Hours to discuss what youâve been building, work through blockers, and learn from your peers.
đŁ Q&A
During our three sessions we had more Q&A than we could get to in our time, so with the help of Loom, Rovo, and the TAS team we have compiled a list of all the questions (and our answers) from those sessions. Hope this helps the attendees and anyone else learning about Rovo!
Agent Creation, Ownership & Permissions
Who can create Rovo agents? What permissions are required?
Permissions to create Rovo agents have just been released (ROVO-76) you can now restrict permissions to create agents to a group of users if desired.
Can admins take ownership of agents if the creator leaves?
Yes, admins can take ownership of an agent that does not have an owner, for example if the original owner has been deactivated (Create and edit agents).
Can multiple users edit or own an agent?
Not currently, but this feature is planned (ROVO-130).
Can agent visibility be restricted to certain users or groups?
This is not currently possible, the feature request is being tracked (ROVO-142).
Do agents run with the creator's or user's permissions?
Agents operate with the permissions of the user who sent the prompt. In the context of Automation rules, the run as whoever configured the connection to Rovo.
Can service accounts be used for automation actions?
The current service user functionality can be used for automation actions, but not for the sake of using Rovo within an automation rule given that you have to be logged in as the user to set up the rule.
Based on feedback from this workshop the automation team is aware of this and looking into making this a possibility!
Agent Functionality and Automation
How do you get to the Agents?
To access Rovo Agents, navigate to the Rovo Chat interface within your Atlassian application (like Jira, Confluence, etc.). Then, select the "Agents" tab. From there, you can browse available agents by scrolling or searching, and start a chat with any agent by selecting it.
How efficient is the chat option to build the agent vs manual build?
Using a structured agent template works better when building agents that perform different functions or sets of actions based on context in their prompts. The chat option is faster and easier for building simple Rovo agents, while manual build offers more control and customization for advanced agents.
Can you create multiple agents and have agent-to-agent workflows?
Agents cannot reference other agents in the chat interface currently. You can work around this in automation to design workflows where agents interact with each other.
Are there other ways to automatically trigger a Rovo agent, other than using Jira actions?
Currently the only options to automatically call a Rovo agent would be to leverage the automation platform. There is no API to call Rovo agents currently.
Can we trigger an agent from within a workflow?
Yes, configure an Automation to have a specific workflow transition as the trigger, and then add the Agent to the rule. Configuring the workflow to call the agent as an action or automation step, depending on the platform's integration capabilities (Agents in automations).
Can Rovo create an issue or work item in Jira?
Yes, it's possible via Rovo Chat or by triggering it via an Agent and Automation action triggers (Agentic actions).
Can Rovo extract reports to Excel?
Rovo agents can help extract data and prepare it for export to Excel, but the final step may require you to download the data as CSV or similar format.
Can Rovo create a troubleshooting article based on the comments and solution of a closed Jira ticket?
Yes, Rovo can create a troubleshooting article from a closed Jira ticket by extracting relevant information and generating a new article.
Can Rovo suggest the best Workflow for the team and later create and associate it to the Jira project?
Yes, Rovo can help your team design the best Jira workflow and beta features are available to assist in the creation of that workflow to your Jira project (see the Jira Workflow Editor Operations action currently in beta).
Can the agents automatically assign tickets to team members within a team?
Yes, you can use the agents to automatically assign tickets to team members within a team. You would use the same basic flow as described in this workshop, but configure your agent to return a user ID, and set the assignee field.
Can Rovo be connected with Slack, to perform actions like edit Confluence pages?
Yes, Rovo can be connected with Slack and used to perform actions like editing Confluence pages, but the ability to granularly edit a Confluence page is currently limited to appending content to the bottom of an existing page.
Can this agent communicate outside Atlassian space, like MS Teams?
The ability to send a chat or create a channel in Teams has recently entered beta, and should be rolled out to most if not all instances.
Can Rovo already analyze attachments like PDFs?
Yes, it can read PDFs, CSVs, DOCX, and others.
Can Rovo use installed Jira add-ons, e.g., "Microsoft 365 for Jira" or "Time to SLA"?
Rovo natively interacts with standard Atlassian data and features, but does not directly control or utilize third-party Jira add-ons without customization.
Will Rovo be integrated with 3rd party apps like Xray test management?
The app vendors themselves can theoretically release Rovo actions, but we have not seen that happening yet. Alternatively you could write actions in Forge for any third party app that has an API available.
Is Rovo able to process worklogs from Tempo Marketplace app?
Rovo does not natively process worklogs from the Tempo Marketplace app, similar to the above you could write actions in Forge that leveraged Tempoâs APIs to surface worklogs to Rovo.
Can Rovo support site admins when it comes to user creation, groups, and adjusting permissions?
Rovo currently does not allow site admins to directly create users, manage groups, or fully adjust user permissions across the Atlassian site.
Can Rovo build dashboards? Jira Dashboards?
No, it does not have the ability to create dashboards. Similarly to other answers in this section, you could leverage Forge and the REST API to build an agent with this ability.
That being said, itâs good to focus on what LLMs are good at - processing and analyzing text. For this type of automation a simpler script would likely suffice.
Can we make Rovo start a conversation with a user?
Currently, Rovo cannot proactively start a conversation with a user; it responds when a user initiates a chat or selects a conversation starter.
Is Rovo able to process and handle assets data objects?
Currently no, but this is in development (ROVO-34).
Is Rovo able to use Karma (the page builder in Confluence) to edit a page?
The ability to granularly edit a Confluence page is currently limited to appending content to the bottom of an existing page. Work is being done on improving this experience, but we donât have any details to share yet.
Using Confluence page can we ask Rovo to create a Draw.io diagram?
Rovo agents can read Confluence content and perform a variety of actions, but creating Draw.io diagrams from Confluence pages is not a built-in feature at the moment.
Can you use deep research with an agent?
This is not currently possible, but is in development.
Knowledge Sources, Data, and Integration
Will Rovo support a custom knowledge base? For instance, will it be possible in the future to upload Word documents or Excel files and have the Rovo agent use them as a knowledge base or template?
There are some third-party connectors that would make this possible, such as the Sharepoint connector: Knowledge sources for agents
Regarding the knowledge field, will the permissions of the Confluence space affect the results of user use?
Rovo and its agents can only retrieve and surface information from Confluence spaces that the user has permission to access: Knowledge sources for agents
What if the referencing Confluence page has custom permissions?
Rovo always respects the permission of the user running the prompt for both first and third party data: Rovo agent permissions and governance
Considering we have added the link to the Confluence page we need as a reference in the prompt, will the agent also look at other contents in the space selected?
If you provide a specific Confluence page URL, the agent will focus on that page's content. To consider other content, specify this in your prompt.
Does the permission include 3rd party data (e.g., Salesforce or SharePoint)?
Rovo synchronizes with access controls and permission settings from connected third-party apps and your Atlassian apps.
Would the agent provide the insight to the user, that resources are inaccessible, if this is the case?
If the resource is inaccessible, Rovo will notify the user that it can't provide a response due to lack of permissions.
If we use the asset gadget in Confluence can it read from there, let's say Application list?
Currently, Rovo agents cannot natively read or query data from the Confluence Asset gadget out-of-the-box.
For triage - can we reference a set of existing tickets in Jira as the source of 'what good looks like'?
Yes, you can use a set of existing Jira ticketsâincluding their details and manual outcomesâas a knowledge source for Rovo triage agents. This would leave more up to the interpretation of Rovo, but would still theoretically work.
A hybrid approach would be leveraging your existing tickets to create a source of truth in Confluence that Rovo can base its routing off of.
How could we use the project itself as extended knowledge for the agent and let it figure out from the project's historical data?
You can configure the agent to retrieve and analyze the project's historical dataâsuch as issues, comments, worklogs, and outcomesâso it can learn from past patterns and decisions.
About how many custom fields details it can pull the data? If I try to pull 5 custom fields will it pull?
Rovo can retrieve multiple custom fields from Jira (ROVO-137).
Agent Usage, Monitoring, and Performance
My organisation already created a lot of chat agents and it is very difficult to find the right agent when I am in the Rovo chat interface. Is there any best practice on that?
Use search, filters, and favorites to quickly access agents.
Name and describe agents clearly.
Regularly review and refine your agent list.
Watch for upcoming improvements to agent discovery in Rovo.
Can we measure performance of Rovo, if it is doing well after deployment?
You can measure the performance of Rovo after deployment by tracking metrics such as usage statistics: Access Rovo insights.
Improvements to Rovo observability have recently entered EAP and should be widely released soon!
Is there an analytics dashboard for agents?
Yes, Rovo provides analytics dashboards for customers to track agent usage and performance (Access Rovo insights).
Improvements to Rovo observability have recently entered EAP and should be widely released soon!
Are there insights available on the usage of Rovo - for example - new agents created, times an agent was used etc.?
Improvements to Rovo observability have recently entered EAP and should be widely released soon!
Security, Privacy, and Compliance
Where does Rovo 'live', where does the data reside, and is that searchable outside our instance?
Rovo supports data residency with limited availability (Rovo Data Privacy and Usage Guidelines).
By default the data is not searchable, but our Remote MCP Server allows for creating a connection to the Teamwork Graph for internal LLMs to parse.
Can Rovo AI be hacked by employees, for example through prompt engineering?
Rovo is designed with security in mind and follows Atlassian's security best practices. We have not seen examples of being able to improperly access data through prompt engineering.
As Rovo is granted autonomy when it performs certain actions, can we trace back if this is a legit human action or an agentic action?
Organization admins can source information about tracing Rovo agent actions and distinguish between human and agentic actions by leveraging the audit log. Currently Rovo actions are only being logged for Guard Premium customers, but efforts are being made to expand that to all Rovo users.
Pricing, Credits, and Quotas
With regards to usage, we understand that it is around 10 Credits for a General Search using Rovo and can be 100 Credits for any Deep Research using Rovo. If we have a Rovo Agent that is designed to fulfil a critical part of our workflow, and we have exhausted our total credits, I'd imagine that the Agent would stop operating; am I correct in thinking in that way?
Currently Rovo does not enforce usage quotas strictlyâif you exhaust your Rovo credits, the agent does not immediately stop operating. In the future, usage beyond quotas will incur additional charges (Rovo Usage Limits).
How should we understand the Rovo pricing plan?
More clarity is coming around pricing and credits soon. The best sources of information now are Rovo Usage Limits and Double Clicking on Rovo Usage Quotas - Update to give you more flexibility and control.
Best Practices, Guidance, and Troubleshooting
How important is the Description in Identity?
The description of the agent can help provide clarity to users, helping them quickly understand what the agent does and set expectations. It does not impact the behavior of an agent at all.
Could we create an agent that reviews all of the existing issues under features in a project and then evaluates whether we have the correct tasks to ship the features? If we don't, can it suggest child epics and tasks to build, test and ship the features?
Yes, but you would need to understand the structure of your data. For more details on building such agents and integrating them with workflows, see the Rovo documentation on Agents in automations.
If anyone can make an agent in the org, how do you prevent the list of available agents getting out of control and will there be tools to detect abandoned agents or unused agents and remove them at an admin level?
Admins can control this using the newly shipped Create Agent Permissions feature. Rovo agents permissions and governance
How does Rovo handle if it finds multiple answers to a question?
When Rovo finds multiple answers to a question, it uses several factors to determine which answer to provide, with one key factor being the recency of the information - pages that have been updated more recently are ranked higher. Rovo scores and ranks potential answers, aiming to select the most relevant and up-to-date response, but the quality of the answer ultimately depends on the quality and freshness of the underlying data it is referencing.
I would be interested in creating an agent that is an expert on many technical manuals that we keep in a SharePoint library. Any gotchas to look out for?
Start in a sandbox and configure a leaner knowledge source (such as a specific set of technical manuals). Volume could be an issue, depending on how many manuals are involved. Having some sort of index page that could be used as a source of truth could help work around this.
These prompts inside general configuration should be same as you have written, other than the entered links, or can be different?
You can change the prompt, tweak and experiment with it to make it work for you. The example provided is what was used in testing for this workshop.
Can you talk to us more about aligning our instructions with the "actions" portion?
When writing a prompt, if you would like the agent to use a specific action call it out by name such as using the Send Slack message action, send the following message in the channel #updates: [...]
Was this system prompt influenced by Claudeâs style?
The TCREI framework (Task, Context, References, Evaluate, Iterate) reflects best practices in prompt engineering common across many AI platforms, including Claude, ChatGPT, and others.
I didn't catch what that prompt engineering framework you recommended was.
The TCREI framework. Feel free to have a look at some use cases and prompts for agents here (Rovo - use cases).
How do you get to the Automation?
Depending on permissions, Automation is available at the project/space level, or from the Global Automation option for administrators.
Is there a way for Rovo to assess its confidence? And based on that to just say "I'm not sure" instead of taking actions?
Rovo agents can assess their confidence in responses, but the ability to act on low confidence depends on how the agent is configured.
Could Rovo be used to enforce regex on a non-mandatory short text field before case creation?
This can easily be solved using a validator in the workflow, Rovo would not contribute anything meaningful here.
Is it planned to have this kind of webinar but using JSM instead of Jira Software to see use cases for using Rovo?
We are planning on putting out a monthly series of webinars with a variety of Rovo related content! That being said, this use case of triage automation would apply equally to JSM as it would for Jira Software.
Will the markdown in the definition ever be rendered?
Markdown in the agent definition or instruction set is not rendered as formatted text within the Rovo agent administration interface.
In the instruction set - does the spacing matter? Should it be spaced exactly like the template?
Spacing does not need to match the template exactly. Structure and clarity are most important.
Do we need to keep the syntax in the instruction as seen here (e.g., using # and * to comment off text)?
Not needed for the functioning of the agent, but it can help with readability of the instructions. LLMs parse markdown syntax well, so it can provide a structure that is easier to parse for both the LLM and human user.
Why custom field for a response team is used instead of the actual team?
This was just for the purposes of the webinar so we could align on a single field with values available to all attendees. It's definitely possible to use the actual Team field!
A lot of customers look to setup multilanguage Wiki's, We know Rovo is good at translation - however automating this, loses the page formatting completely.
This is a use case the TAS team would love to chat about more! To preserve formatting when using automation, try wrapping your {{agentResponse}} value in a code block with the Smart Values language selected as per Use Smart Values with rich text editing.
Was this helpful?
Thanks!
Alex Gallien
3 comments