Stylizer

Inspiration

As engineers, we write in plaintext and bullet points. This makes our life easy but the readers life harder so why not leverage Rovo to get the best of both worlds?

What it does

In the context of a page you can ask Stylizer to help you to add formatting to a page. The agent will confirm the type of styling that you are looking to apply (e.g. Technical, Creative, or Formal) and then propose an amended version of the page.

Stylizer can help improve the formatting of the page using all native Confluence functionalities like info panels, as well as adding relevant headings, and captioning images.

The user can go back and forth with the agents about specific requests and once a final draft has been reached the user can ask the agent to update the current Confluence page.

How we built it

We build the Rovo agent by providing some fundamental actions: The ability to read content from a Confluence page, the ability to update a Confluence page, and an image to text capability to help with captioning images.

After this we developed a detailed prompt to walk the user through the stylization process.

Challenges we ran into

Initially we thought that the Rovo agent itself might be able to caption the images sensibly but some testing showed that it was just inferring a caption from surrounding text which was not always accurate. This lead to the inclusion of the additional image-to-text capability.

We found it difficult to get the Rovo agent to consistently output the page in a more structured format according to Atlassian's Confluence Storage Format. This made the page update functionality unreliable and required significant careful prompting to create a more repeatable process.