Building a Software Factory to Reduce Astro's GitHub Issues to Zero

The Astro team built an automated issue triage pipeline that runs isolated AI subagents within GitHub Actions. This system reads bug reports, reproduces them in a sandbox, diagnoses the root cause, applies fixes, and deploys a preview release for the reporter to verify.

The pipeline is built on Flue, a custom open framework. It is designed so that maintainers can reuse the same triage skills they test locally with a coding harness within GitHub Actions.

The automated issue handling process consists of the following four steps:

• Reproduce: Clone the reported reproduction repository to confirm the issue.
• Diagnose: Add instrumentation and logging to the code to identify the root cause of the bug.
• Verify: Review tests, code comments, and documentation to determine if it is an actual bug or intended behavior.
• Fix: Convert the reproduction case into a failing unit test and apply a solution according to architecture guidelines.

As a result, open issues in the Astro repository decreased from over 200 to approximately 30, and the team expects to reach zero issues within the next month. Instead of bulk-closing old issues or ignoring reports, they chose to automate the maintenance work itself.