
Image: CodeRabbit
Why it mattersA reviewer can see the test result inside the pull request. The hand-off to a QA team can wait until the agent has already tried the fix.
A pull request now comes with a test result. CodeRabbit released Verify in public beta on 18 November. Tejas Venugopal and Sahil M Bansal wrote the announcement. A developer types the comment @coderabbitai verify on a pull request. Verify reads the commits and the source code. It writes test cases from the change. It runs them in a sandbox the product sets up for the repository. The pass or fail result posts back to the pull request. CodeRabbit calls this agentic application testing.
The public beta is free for three actions. Those are PR test runs, full-repo journey runs, and fixes with CodeRabbit Agent. CodeRabbit has not stated how long the beta lasts. A developer who wants to try it can turn it on for one repository today.
What the product does on a pull request
CodeRabbit shows one worked example in the announcement. A developer changes the coupon code logic. Verify builds three test cases from that change. Two of the three report bugs that involve the simulated charge. One case for an invalid coupon and its removal passes. The result is three outcomes a reviewer can read inside the pull request. The reviewer never leaves the page.
Failed cases get a button labelled "Fix with CodeRabbit Agent". The agent reads the failing test and the diff. It writes a patch. The patch attaches to the same pull request. A reviewer can accept the patch or ignore it. The failure can also post to Jira or Linear as an automatic issue. The team hears about the bug in the tracker it already uses. Nobody has to copy the test output across tools by hand.
Where it fits in a working week
A full-repo journey run is the other half. Verify can generate a library of reusable scenarios from the whole repository. A team can tag those scenarios. It can schedule them against a branch and a test environment. It can run selected ones on demand. CodeRabbit saves the evidence of each run as a recording, a screenshot and a log. The post calls those "test evidence" and says it posts them with the result. A reviewer can replay the exact run that failed.
The claim CodeRabbit makes is smaller than a complete QA suite. It will not write an acceptance test a product manager needs to approve. It will not replace a tester who sits with a real user. It aims at the cases a reviewer would want to try before merging. Those are the cases most teams do not have the time to write.
Source
Introducing CodeRabbit Verify, CodeRabbit.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
