If a package failed to build in the test collection, or manual examination of the packaging instructions show that there are configurable build options that we need, the next step is to work in the builder repository to fix those builds.

The input would be:

• Package Names
• Versions for build customizations

The output (generated by AI using our internal knowledge) would be:

• Package settings
• Package plugins
• Package patches
• New requirements and/or constraints

A preliminary MR would be created from a bot, doing 75% of the work. Realistically, it'll require some manual intervention but we can make it better over time by improving our documentation and agents.