Why are there differences between the agents behaviors depending on where they are created? For example, UI created agents can access content on Jira, confluence etc… But a forge rovo agent can not, it will need to build an action to do this. In reverse, I have also noticed limitations on ui created agents but they limitations don’t exist when created in forge.
1 Like
Hi @Zishan ,
Forge based Agents must abide by the same data access rules as any other Forge app functionality and the most straight forward way to implement this is by having the Forge app retrieve the data as it would for any other interaction. For example, Confluence page update notifications to Forge apps don’t include the page content, but they do include the identifiers required to retrieve the page content. Similarly, when a Forge Agent’s action is invoked from a Confluence page chat context, the page content is not sent to the app, but rather the page ID is so that the page content can be retrieved by the app. As for all interactions, it is generally preferred for Forge apps to retrieve data using the current user’s credentials rather than the app’s credentials - i.e. api.asUser() rather than api.asApp(). This avoids the risk of privilege escalation.
Note also that a Forge app will only be able to retrieve data from the current Jira or Confluence workspace and not an alternate workspace, unlike user-created Agents which can access data in any workspace within the site it is installed in. This is also consistent with the way in which Forge apps generally behave - they can only interact with the current workspace as there is no support for cross-product Forge app interactions.
Regards,
Dugald
1 Like