Custom Rovo agent access to custom content

[Habib ZOUARI]

I've created a custom Rovo agent in Forge app installed in confluence workspace. As I know Atlassian built-in Rovo agents can access and retrieve confluence page/space content automatically, so I want to know if it's possible that my custom Rovo agent would have the same or not , or if there are specific configurations, permissions or whatever needed to reach that. I'm asking that because I tried to use my custom agent with some confluence tasks and mostly the agent didn't respond properly.

Also, I want to know if this custom agent is able to access custom content , means that I need the agent to manipulate some tasks inside my custom content app

[Dr Valeri Colon _Connect Centric_ · Accepted solution]

Welcome to the community @Habib ZOUARI Custom Rovo agents don’t get automatic access to all Confluence content. They can only use the knowledge sources you attach and whatever your Forge app passes in, and your app must have the proper scopes to read those spaces or pages. They also can’t access custom app data unless your Forge app retrieves it and includes it in the prompt. If responses seem incomplete, it usually means required spaces weren’t added as knowledge or the app lacks read permissions.

[Habib ZOUARI]

Thank you for your response! @Dr Valeri Colon _Connect Centric_ 

 How do I attach knowledge sources (Confluence spaces/pages/or even customcontent) to my custom Rovo agent? Is this done through:

1. The manifest file configuration?
2. The Rovo admin panel at admin.atlassian.com?
3. Or programmatically through the Forge API?

Could you please point me to documentation or provide an example of how to attach Confluence spaces as knowledge sources to a custom Forge Rovo agent?

[Dr Valeri Colon _Connect Centric_]

Great questions, @Habib ZOUARI For custom Rovo agents built in Forge, knowledge sources aren’t attached in the manifest or via API today. They’re added the same way you configure knowledge for any other agent. Go to admin.atlassian.com → Rovo → Agents → Knowledge and attach the spaces or pages you want. Custom app data isn’t indexed automatically, so your Forge app must retrieve that content and pass it into the agent.

https://developer.atlassian.com/platform/forge/extend-atlassian-products-with-a-forge-rovo-agent/

[Habib ZOUARI]

Hello again @Dr Valeri Colon _Connect Centric_ hope you're doing well.

 I have some several questions if you have some answers I would be grateful.

When a Forge custom Rovo Agent is asked to generate a large volume of structured content (e.g. a full course with multiple sections and components) in a single conversation turn, the LLM that powers the Rovo Agent silently truncates its output mid-generation. Because action inputs are constructed from the LLM's raw output, the truncated content results in malformed (non-parseable) JSON being passed to the action handler ,causing the action to fail or receive incomplete data. So i need to know if :

1. Is there an official, documented token/character limit on the LLM output that populates action inputs for Forge Rovo Agents?
2. Is there a plan to expose a structured truncation signal to action handlers?
3. Is there an alternative solution to maintain the generated large content with the same number of components?

Your help is appreciated !

[Dr Valeri Colon _Connect Centric_]

@Habib ZOUARI again great questions—this is a real limitation people are running into.

1. There isn’t a clearly documented hard token/character limit for Rovo action inputs. Limits come from the underlying LLM + platform, and truncation can happen without warning.
2. There’s currently no structured truncation signal exposed to action handlers.
3. For large outputs, the reliable pattern is to chunk the generation (section by section) or generate/store content first, then pass smaller payloads to actions.

Single-turn “large payload” generation isn’t reliable today.

[Habib ZOUARI]

@Dr Valeri Colon _Connect Centric_ thanks for the clarification. I’ve already tried batching as a workaround, but the agent’s behavior is inconsistent — sometimes it works as expected, but other times it impacts the user experience by prompting for confirmation or splitting the process unexpectedly.

A few follow-up questions:

• Is there any way to make the batching experience more seamless for users, or to control how/when the agent asks for confirmation?
• Are there plans to improve large content generation in a single turn, or to provide more predictable handling for these cases?
• Is there any recommended best practice for handling or detecting output truncation, beyond trial and error?

Thanks again for your help.

[Dr Valeri Colon _Connect Centric_]

@Habib ZOUARI —this is where things get tricky today.

For batching, there’s no fine control over when the agent asks for confirmation—that’s part of Rovo’s interaction model. Keeping steps small and clearly scoped helps reduce interruptions.

On large content, no firm timeline, but this is an area Atlassian is actively improving.

For truncation, since there’s no explicit signal, teams usually rely on validation patterns (e.g., checking outputs before passing to actions).

You’re hitting real platform limits here—not a misconfiguration.