Note: Code scanning is currently in beta and subject to change. For more information about taking part in the beta, sign up here.
In repositories where code scanning is configured as a pull request check, code scanning checks the code in the pull request. By default, this is limited to pull requests that target the default branch, but you can change this configuration within GitHub Actions or in a third-party CI/CD system. If merging the changes would introduce new code scanning alerts to the target branch, these are reported as check results in the pull request. The alerts are also shown as annotations in the Files changed tab of the pull request. If you have write permission for the repository, you can see any existing code scanning alerts on the Security tab. For information about repository alerts, see "Managing code scanning alerts for your repository."
If code scanning has any results with a severity of
error, the check fails and the error is reported in the check results. If all the results found by code scanning have lower severities, the alerts are treated as warnings or notices and the check succeeds. If your pull request targets a protected branch that has been enabled for code scanning, and the repository owner has configured required status checks, then you must either fix or close all error alerts before the pull request can be merged. For more information, see "About required status checks."
There are many options for configuring code scanning as a pull request check, so the exact setup of each repository will vary and some will have more than one check. The check that contains the results of code scanning is: Code scanning results.
If the repository uses the CodeQL analysis workflow a CodeQL / Analyze (LANGUAGE) check is run for each language before the results check runs. The analysis check may fail if there are configuration problems, or if the pull request breaks the build for a language that the analysis needs to compile (for example, C/C++, C#, or Java). As with other pull request checks, you can see full details of the check failure on the Checks tab. For more information about configuring and troubleshooting, see "Configuring code scanning" or "Troubleshooting code scanning."
When you look at the Files changed tab for a pull request, you see annotations for any lines of code that triggered the alert.
Some annotations contain links with extra context for the alert. In the example above, from CodeQL analysis, you can click user-provided value to see where the untrusted data enters the data flow (this is referred to as the source). In this case you can view the full path from the source to the code that uses the data (the sink) by clicking Show paths. This makes it easy to check whether the data is untrusted or if the analysis failed to recognize a data sanitization step between the source and the sink. For information about analyzing data flow using CodeQL, see "About data flow analysis."
For more information about an alert, click Show more details on the annotation. This allows you to see all of the context and metadata provided by the tool in an alert view. In the example below, you can see tags showing the severity, type, and relevant common weakness enumerations (CWEs) for the problem. The view also shows which commit introduced the problem.
In the detailed view for an alert, some code scanning tools, like CodeQL analysis, also include a description of the problem and a Show more link for guidance on how to fix your code.
Anyone with write permission for a repository can fix a code scanning alert that's identified on a pull request. If you commit changes to the pull request this triggers a new run of the pull request checks. If your changes fix the problem, the alert is closed and the annotation removed.
If you don't think that an alert needs to be fixed, you can close the alert manually. For example, an error in code that's used only for testing, or when the effort of fixing the error is greater than the potential benefit of improving the code. The Close button is available in annotations and in the alerts view if you have write permission for the repository.
If you close a CodeQL alert as a false positive result, for example because the code uses a sanitization library that isn't supported, consider contributing to the CodeQL repository and improving the analysis. For more information about CodeQL, see "Contributing to CodeQL."