Automatic dependency submission

Configure network access for self-hosted runners

If your self-hosted runners operate behind a firewall with restricted outbound internet access, you must add certain URLs to the allowlist for automatic dependency submission. The required URLs depend on the ecosystems your repositories use.

Required URLs for all ecosystems

These URLs are required for all automatic dependency submission workflows:

• https://github.com—Required for accessing GitHub and downloading actions.
• https://api.github.com—Required for GitHub API access.
• https://*.githubusercontent.com—Required for downloading action source code and releases (including raw.githubusercontent.com, github-releases.githubusercontent.com, and objects.githubusercontent.com).

Ecosystem-specific URLs

Depending on the ecosystems you use, you may need to allowlist additional URLs.

Go

• https://go.dev—For downloading the Go toolchain.
• https://golang.org—Alternate domain for Go downloads.
• https://proxy.golang.org—Official Go module proxy for downloading Go modules during dependency detection.

Java (Maven and Gradle)

• https://repo.maven.apache.org—Maven Central repository for downloading dependencies.
• https://api.adoptium.net—For downloading Adoptium/Temurin JDK distributions (default distribution used by actions/setup-java).

For Gradle projects, you also need access to the Gradle Plugin Portal, where the autosubmission action downloads the github-dependency-graph-gradle-plugin by default:

• https://plugins.gradle.org—Serves the plugin marker and maven-metadata.xml.
• https://plugins-artifacts.gradle.org—Serves the plugin's POM, module metadata, and JAR. The Gradle Plugin Portal redirects artifact downloads to this host, so allowlisting only plugins.gradle.org resolves the metadata and then fails when the plugin artifact is downloaded.

If your runners cannot reach the Gradle Plugin Portal, or you want to avoid depending on hosts that can change over time, resolve the plugin from an internal repository you control instead. For more information, see Gradle projects.

If you use a different JDK distribution, you may also need:

• https://aka.ms and https://download.microsoft.com—For Microsoft Build of OpenJDK (note: aka.ms is also used for .NET downloads).
• https://download.oracle.com—For Oracle JDK.
• https://api.azul.com—For Azul Zulu OpenJDK.

.NET (C#, F#, Visual Basic)

• https://aka.ms—Microsoft URL shortener that redirects to .NET download locations.
• https://builds.dotnet.microsoft.com—Primary feed for .NET SDK and runtime downloads.
• https://ci.dot.net—Secondary feed for .NET builds.

Python

• https://python.org—For downloading Python interpreters.

Use GitHub-hosted larger runners for automatic dependency submission

GitHub Team or GitHub Enterprise Cloud users can use larger runners to run automatic dependency submissions jobs.

1. Provision a larger runner at the organization level with the name dependency-submission. For more information, see Adding a larger runner to an organization.
2. Give your repository access to the runner. For more information, see Allowing repositories to access larger runners.
3. Under "Dependency graph", click the dropdown menu next to “Automatic dependency submission”, then select Enabled for labeled runners.

Troubleshoot automatic dependency submission

Automatic dependency submission makes a best effort to cache package downloads between runs using the Cache action to speed up workflows. For self-hosted runners, you may want to manage this cache within your own infrastructure. To do this, you can disable the built-in caching by setting an environment variable of GH_DEPENDENCY_SUBMISSION_SKIP_CACHE to true. For more information, see Store information in variables.

Manifest deduplication

A repository can use multiple methods for dependency submission, which can cause the same package manifest to be scanned multiple times, potentially with different outputs from each scan. Dependency graph uses deduplication logic to parse the outputs, prioritizing the most accurate information for each manifest file.

Dependency graph displays only one instance of each manifest file using the following precedence rules.

1. User submissions take the highest priority, because they are usually created during artifact builds they have the most complete information.
   • If there are multiple manual snapshots from different detectors, they are sorted alphabetically by correlator and the first one used.
   • If there are two correlators with the same detector, the resolved dependencies are merged. For more information about correlators and detectors, see REST API endpoints for dependency submission.
2. Dependabot graph jobs have the second-highest priority. For ecosystems where Dependabot graph jobs are available (currently Go and Python), they take precedence over automatic dependency submission.
3. Automatic submissions have the next priority since they are also created during artifact builds, but are not submitted by users.
4. Static analysis results are used when no other data is available.

Package ecosystem-specific information

Maven projects

For Maven projects, automatic dependency submission runs an open source fork of the Maven Dependency Tree Dependency Submission. The fork allows GitHub to stay in sync with the upstream repository plus maintain some changes that are only applicable to automatic submission. The fork's source is available at advanced-security/maven-dependency-submission-action.

If your repository's dependencies seem inaccurate, check that the timestamp of the last dependency graph build matches the last change to your pom.xml file. The timestamp is visible on the table of alerts in the repository's Dependabot tab. Pushing a commit which updates pom.xml will trigger a new run of the Dependency Tree Submission action and force a rebuild of that repository's dependency graph.

Gradle projects

For Gradle projects, automatic dependency submission runs a fork of the open source Gradle actions from gradle/actions. The fork is available at actions/gradle-build-tools-actions. You can view the results of the autosubmission action under your repository's Actions tab. Each run will be labeled "Automatic Dependency Submission (Gradle)" and its output will contain the JSON payload which the action submitted to the API.

Resolving the submission plugin from an internal repository

By default, the action downloads the github-dependency-graph-gradle-plugin from the Gradle Plugin Portal (https://plugins.gradle.org). For self-hosted runners on a restricted network, hosting the plugin in an internal repository that you control, such as a private Artifactory or Nexus instance, is more reliable than allowlisting the portal, whose hosts and CDNs can change over time. You can point the action at your internal repository with these environment variables:

• GRADLE_PLUGIN_REPOSITORY_URL—The URL of the internal repository to resolve the plugin from.
• GRADLE_PLUGIN_REPOSITORY_USERNAME and GRADLE_PLUGIN_REPOSITORY_PASSWORD—Credentials, if the repository requires authentication.

Automatic dependency submission runs a workflow that GitHub manages, not one you author in your repository, so you cannot add an env: block to it. You can set these variables on the runner, but every job scheduled on that runner, not only automatic dependency submission jobs, inherits them.

If your internal repository allows anonymous read access, you only need to set GRADLE_PLUGIN_REPOSITORY_URL and can omit the credential variables entirely, avoiding this concern. If the repository requires authentication, use read-only credentials and a dedicated runner. For organization- or enterprise-level runners, also restrict runner-group access to only the repositories that need these credentials. See Managing access to self-hosted runners using groups.

Resolving the plugin from an internal repository is separate from configuring how your build resolves its own dependencies, for example an init.gradle file that points at an internal registry. These variables control only where the dependency-submission plugin is downloaded from.

For the latest configuration details, see the actions/gradle-build-tools-actions documentation.

.NET projects

The .NET autosubmission action uses the open source component-detection project as the engine for its dependency detection. It supports .NET 8.x, 9.x, and 10.x. .NET autosubmission runs if the repository's dependabot.yml defines nuget as a package-ecosystem or when there is a supported manifest file in the root directory of the repository. Supported manifest files include .sln, .csproj, packages.config, .vbproj, .vcxproj, and .fsproj.

Python projects

Python repositories with the dependency graph enabled use Dependabot graph jobs to generate dependency graphs. Dependabot graph jobs take precedence over automatic dependency submissions, and the autosubmission workflow described below will no longer run for Python projects. For more information, see How the dependency graph recognizes dependencies.

Automatic dependency submission for Python uses the open source component-detection project as its underlying graph generation engine. The autosubmission action will only run if there is a requirements.txt file in the root directory of the repository. Python autosubmission does not currently support private packages; packages referenced in requirements.txt which are not publicly available will cause the autosubmission action to fail.