Inspiration

Confluence can become a junkyard of unorganised documents. We wanted a way to easily create templates for pages.

What it does

Generates templates based on 1 or more existing pages, including interactive refining of the template. Can search for pages based on their title, or help the user find the pages they want in the current space. After guiding the user through the process, it creates a space or global template in Confluence.

How we built it

We wrote actions for a Rovo agent, allowing it to fetch Confluence content in various ways including via CQL, by title, and by ID. Then gave it an action for creating templates, using the agents existing knowledge of Confluence storage format to convert the template suggestions to an actionable format.

Then we wrote a detailed prompt informing the agent how to perform these tasks and what guidance to offer the user about the process.

Challenges we ran into

Development cycle for Rovo is a little painful as most changes involve updating the manifest which means deploying every time. Getting the agent to act in a reliable and predictable way was difficult, especially enabling it to use it's actions properly. Debugging agents is quite challenging as there are no logs or way of knowing why/how they are doing certain things.

Accomplishments that we're proud of

Building a useful Rovo agent in only 2 days - quick onboarding process for Forge and lack of infrastructure enabled this. Getting the agent to use CQL search including pagination to find Confluence pages.

What we learned

The importance of prompt structure in guiding the agent to perform reliably. How to make a Rovo agent - this is our first time learning to make one.

What's next for Standardizer

Releasing the app to the marketplace.

Built With

Share this project:

Updates