Last updated: November 13, 2025

Context

Customers may face a situation where Learnings alone are not fixing recurring issues during pull request reviews. For example, CodeRabbit may comment that the code “does not compile” despite an existing Learning that instructs CodeRabbit to assume that PR code always compiles. This may appear to indicate that CodeRabbit is not consistently applying the Learnings.

These situations can occur when the model ignores or deprioritizes certain Learnings, possibly due to contextual overlap or conflicting instructions. In such cases, adding a “review instruction” rule can help resolve the issue by prompting the model to stop and reassess the Learnings before proceeding.

Answer

If Learnings are not being respected or seem to be ignored during automated reviews, you can apply the following workaround:

1. Review existing Learnings. Go to your project’s Learnings configuration and verify that all relevant Learnings are active and clearly phrased.

2. Add a « review instruction » rule. Introduce a new rule that explicitly tells the model to stop and reconsider the Learnings before continuing the review. For example:

“Before responding, review all Learnings to ensure none are ignored.”

This additional instruction helps ensure that the model prioritizes Learnings when generating review comments.

3. Save and re-test. Commit this change and observe the next few automated reviews. The model should now respect the Learnings more consistently.