A pull request is rarely finished the moment it is opened. CI might fail on something the run could not check, or a reviewer asks for changes. Instead of fixing that by hand or starting from scratch, let CodeLynq work the feedback into the same pull request. It updates the existing branch; it never opens a second pull request for it.
The feedback loop works for repositories on GitHub, GitLab and Bitbucket.
See the feedback
Open the task, go to the Runs tab and select the run. Under Pull request feedback you see:
- the state of the pull request, and whether it is Mergeable or has Conflicts,
- the CI checks, each passed, pending or failed, with the reason for a failure,
- the Review comments.
Click Refresh to fetch the latest state from your git host.
Address it on request
Click Iterate on feedback. CodeLynq reads the current CI results and review comments, makes the changes on the pull request's branch, runs your verify command again and pushes the update. The run shows Iterated × with the number of rounds so far.
The button is available while the run has an open pull request and is not busy. A merged or closed pull request cannot be iterated on, and there is a maximum number of rounds per run.
Tip: for review comments, be as concrete as you would be for a colleague. "Use the existing formatDate helper" works better than "this could be cleaner".
Address it automatically
With Auto-iterate until green, CodeLynq starts a new round by itself whenever CI fails or a reviewer requests changes on one of its pull requests - until the checks pass or the maximum number of rounds is reached.
To switch it on:
- Set up the CI webhook on each repository, so your git host tells CodeLynq about CI results and reviews. See Connect a repository.
- Open the project, go to Settings → Context & conventions and turn on Auto-iterate until green. This requires being an organization admin or the project's owner.
- Click Save Settings.
Automatic rounds count as runs, just like any other: they count towards your plan's limits and use model credit, and they are not started when a limit is reached.
Feedback that is not about the pull request
Do you want to change the approach, rather than respond to CI or a review? Use Request a change on the task's Overview: the AI works your request into a new version of the spec, and once you approve it the change is built on the same branch, updating this pull request. See Request a change to a built task.
Several branches or repositories
With a task on several branches or repositories, each pull request is followed up on its own branch.
Related
- Quality gates
- Analytics and costs - iterations are shown as a separate figure.