GitHub Action
Harden-Runner
🔒 Harden-Runner GitHub Action installs a security agent on the GitHub-hosted runner (Ubuntu VM) to
- Prevent exfiltration of credentials 🔑
- Detect tampering of source code during build 🕵️
- Detect compromised dependencies and build tools 🚨
Compromised dependencies and build tools typically make outbound calls to exfiltrate credentials, or may tamper source code, dependencies, or artifacts during the build.
Harden-Runner GitHub Actions installs a daemon that monitors process, file, and network activity to:
Countermeasure | Threat | |
---|---|---|
1. | Block outbound calls that are not in the allowed list to prevent exfiltration of credentials | To prevent Codecov breach scenario |
2. | Detect if source code is being overwritten during the build process to inject a backdoor | To detect SolarWinds incident scenario |
3. | Detect compromised dependencies that make unexpected outbound network calls | To detect Dependency confusion and Malicious dependencies |
Read this case study on how Harden-Runner detected malicious packages in the NPM registry.
-
Add
step-security/harden-runner
to your GitHub Actions workflow file as the first step in each job.steps: - uses: step-security/harden-runner@128a63446a954579617e875aaab7d2978154e969 # v2.4.0 with: egress-policy: audit
-
In the workflow logs, you will see a link to security insights and recommendations.
-
Click on the link (example link). You will see a process monitor view of file and network activities correlated with each step of the job. These insights hold significant value for forensic investigations, proving crucial in the event of an incident.
-
Under the insights section, you'll find a suggested policy. You can either update your workflow file with this policy, or alternatively, use the Policy Store to apply the policy without modifying the workflow file.
For details, check out the documentation at https://docs.stepsecurity.io
Once allowed endpoints are set in the policy in the workflow file, or in the Policy Store
- Harden-Runner blocks egress traffic at the DNS (Layer 7) and network layers (Layers 3 and 4).
- It blocks DNS exfiltration, where attacker tries to send data out using DNS resolution
- Blocks outbound traffic using IP tables
- Wildcard domains are supported, e.g. you can add
*.data.mcr.microsoft.com:443
to the allowed list, and egress traffic will be allowed toeastus.data.mcr.microsoft.com:443
andwestus.data.mcr.microsoft.com:443
.
Harden-Runner monitors file writes and can detect if a file is overwritten.
- Source code overwrite is not expected in a release build
- All source code files are monitored, which means even changes to IaC files (Kubernetes manifest, Terraform) are detected
- You can enable notifications to get one-time alert when source code is overwritten
GitHub-hosted runner uses passwordless sudo for running jobs.
- This means compromised build tools or dependencies can install attack tools
- If your job does not need sudo access, you see a policy recommendation to disable sudo in the insights page
- When you set
disable-sudo
totrue
, the job steps run without sudo access to the Ubuntu VM
Install the StepSecurity Actions Security GitHub App to get security alerts.
- Email and Slack notifications are supported
- Notifications are sent when outbound traffic is blocked or source code is overwritten
- Notifications are not repeated for the same alert for a given workflow
Private repositories are supported if they have a commercial license. Check out the documentation for more details.
Install the StepSecurity Actions Security GitHub App to use Harden-Runner GitHub Action for Private
repositories.
- If you use Harden-Runner GitHub Action in a private repository, the generated insights URL is NOT public.
- You need to authenticate first to access insights URL for private repository. Only those who have access to the repository can view it.
- StepSecurity Actions Security GitHub App only needs
actions: read
permissions on your repositories.
Read this case study on how Kapiche uses Harden Runner to improve software supply chain security in their open source and private repositories.
If you have questions or ideas, please use discussions.
- Support for private repositories
- Where should allowed-endpoints be stored?
- Cryptographically verify tools run as part of the CI/ CD pipeline
- Harden-Runner GitHub Action only works for GitHub-hosted runners. Self-hosted runners are not supported. We have started work on supporting Kubernetes-Based Self-Hosted Actions Runners.
- Only Ubuntu VM is supported. Windows and MacOS GitHub-hosted runners are not supported. There is a discussion about that here.
- Harden-Runner is not supported when job is run in a container as it needs sudo access on the Ubuntu VM to run. It can be used to monitor jobs that use containers to run steps. The limitation is if the entire job is run in a container. That is not common for GitHub Actions workflows, as most of them run directly on
ubuntu-latest
.
I think this is a great idea and for the threat model of build-time, an immediate network egress request monitoring makes a lot of sense - Liran Tal, GitHub Star, and Author of Essential Node.js Security
Harden-Runner strikes an elegant balance between ease-of-use, maintainability, and mitigation that I intend to apply to all of my 300+ npm packages. I look forward to the tool’s improvement over time - Jordan Harband, Open Source Maintainer
Harden runner from Step security is such a nice solution, it is another piece of the puzzle in helping treat the CI environment like production and solving supply chain security. I look forward to seeing it evolve. - Cam Parry, Staff Site Reliability Engineer, Kapiche
Some important workflows using harden-runner:
Repository | Link to insights | |
---|---|---|
1. | nvm-sh/nvm | Link to insights |
2. | jsx-eslint/eslint-plugin-react | Link to insights |
3. | microsoft/msquic | Link to insights |
4. | ossf/scorecard | Link to insights |
5. | Automattic/vip-go-mu-plugins | Link to insights |
Harden-Runner GitHub Action downloads and installs the StepSecurity Agent.
- The code to monitor file, process, and network activity is in the Agent.
- The agent is written in Go and is open source at https://github.com/step-security/agent
- The agent's build is reproducible. You can view the steps to reproduce the build here