Overview
AI agent outputs are non-deterministic, so Tusk may sometimes fail to create a PR or create a PR with incorrect changes. For the sake of transparency, we wanted to share some of the main things that tend to trip Tusk up. We’re constantly working on making Tusk better at generating the highest-quality code for your codebase, so you’ll see the list below shrink as we make improvements.Reasons for PR Failure
Issue is too complex
Tusk is optimized to handle smaller issues (i.e., 1 story point) end to end. It works best when you assign it an issue that is a single user story. Tusk is unlikely to create a PR for an issue that:- Requires changes in more than 5 files, unless the changes are straightforward
- Requires concurrent changes across more than 1 repo
- Would take a human engineer more than 1 hour
- Would trouble a junior software engineer