Prerequisites
-
A repository connected through Review setup, with its provider sending pipeline events:
Review fetches jobs and logs from these CI systems only, so runs from other CI tools, such as check runs another tool posts on GitHub, aren’t analyzed. See Configure webhooks. AWS CodeCommit repositories don’t support pipeline analysis.
- The workspace Admin role to turn analysis on or start one by hand. Everyone else can read runs and findings. See roles.
Turn on pipeline analysis
Pipeline analysis is a workspace-wide setting and is off by default. There’s no per-repository switch: once it’s on, every failed run from a connected repository that sends pipeline events is analyzed.1
Open Review settings
Go to Review → Settings.
2
Switch it on
In the Review section, turn on Pipeline analysis. You can also click Turn on in the banner at the top of Review → Pipelines.Success state: the toast reads “Pipeline analysis enabled”, and new CI runs from your connected repositories start to appear on Review → Pipelines.
How a failed run becomes findings
- Your provider sends a pipeline event, and Review records the run with its status and commit.
- Only failed runs are analyzed. Successful, running, pending and cancelled runs are listed but never analyzed, and a run is never analyzed twice.
- Review fetches the failed jobs and the last 50 KB of each job’s log through your Git connection. When the run belongs to a reviewed MR or PR, Review links it so the analysis can use that change’s review context.
- CloudThinker analyzes the failed jobs in a Chat conversation and writes root-cause findings with a suggested fix. Workspace admins get a Pipeline analysis notification when it finds issues. Choose where it goes in notification settings.
Only the last 50 KB of each failed job’s log is read. If the real error comes earlier, for example a compile error followed by a long test dump, the analysis can miss it. Fail fast, or trim verbose output from the job, so the error lands near the end of the log.
Check a failed run
1
Open the run list
Go to Review → Pipelines. Each row shows the run Status, the Pipeline (repository, number and commit), and the Analysis state. Narrow the list with Search runs, Any status and Any analysis.
2
Start an analysis if one hasn't run
A failed run marked Not analyzed shows an Analyze button. Click it to start without waiting.Success state: the toast reads “Pipeline analysis started” and the row moves to Analyzing.
3
Read the findings
Open the run. Pipeline jobs shows each stage and job with its status. Findings lists the root causes with Severity, Issue, Location and Status. Status reads Open until someone settles the finding, then Resolved, Acknowledged or Withdrawn. Open a finding for its Description and Suggested Fix.
4
Hand the fix off
On a finding, click Copy prompt to copy a ready-made fix request for your coding agent, or View file to jump to the failing code on your provider. On the run list, click an Analyzed badge to open the full analysis in Chat.
Troubleshooting
Review → Pipelines says No pipeline runs yet
Review → Pipelines says No pipeline runs yet
Check that Pipeline analysis is on and that your provider’s webhook is set up to send pipeline events. Then push a commit that runs a pipeline.
A failed run shows Analysis failed
A failed run shows Analysis failed
Review couldn’t fetch the failed jobs from your provider or couldn’t start the analysis. A run is analyzed only once, so check that your Git connection is healthy in Review setup, then rerun the pipeline on your provider to get a new run.
Next steps
Releases
Check whether a release is safe to ship before you deploy
Review Insights
Track review quality and get weekly repository reports
Mention commands
Talk to Review from MR and PR comments
Notifications
Choose where Review alerts and digests go