On Monday morning the security tab of your GitHub organisation shows thousands of Dependabot and code scanning findings. Nobody can fix them all this week, so the operational question is: which ones first? GitHub Advanced Security alerts tell you what is vulnerable in the code. They do not tell you whether the affected image is running anywhere, or reachable from the internet.
Dynatrace can fill that gap for Kubernetes workloads. This article follows the order in which you will ask the questions: what you need, how an image gets tied to its repository, how to check that data flows, and how to filter in GitHub. For an overview of the feature, read Dynatrace Application Security first. Facts come from the Dynatrace documentation and the vendor’s blog post, read on 12 November 2026. I did not install the extension, and anything unverified is flagged as such.
For the analysis side inside Dynatrace, see the article on runtime vulnerabilities and process exposure. Here we look at the opposite direction: sending that context back to GitHub.
The problem: a backlog without context
A dependency alert on a library that is never deployed is not as urgent as one on a service exposed in production. Without deployment information every alert looks alike, and teams sort by severity alone or at random. The cost is time: each needless fix delays a useful one.
The idea is simple. Dynatrace finds the container images that actually run, matches them to their GitHub repository, then writes a deployment record on the GitHub side. If Runtime Vulnerability Analytics (RVA) is active, the record also carries internet exposure and sensitive data reachability.
GitHub Advanced Security alerts: what you need first
This applies to Dynatrace SaaS: the runtime context relies on Grail and a platform token. The documentation I read does not cover Dynatrace Managed, so I do not claim it for Managed.
- An ActiveGate version 1.310 or later that can run Extensions 2.0 and reach the GitHub API.
- A token with the
openpipeline.events_securityscope for ingestion. - For runtime context: a platform token with
storage:smartscape:read,storage:buckets:readandstorage:security.events:read. - Optional but useful: RVA, for internet exposure and data access.
On the GitHub side, the documentation recommends a GitHub App over a personal access token: fine-grained permissions, organisation-level audit logs and higher API limits. A classic personal token (scopes repo and audit_log) is fine for a quick test, but you must manage its expiry by hand.
Two permissions matter for runtime context: Artifact metadata with read and write access (it lets Dynatrace write the deployment records) and Attestations read-only. Install the app on each organisation you want to monitor.
Linking an image to its repository: attestations or naming
This step decides how good the result is. Dynatrace queries Grail for Kubernetes workload images, then looks for the matching repository among those the configured user can read.
- Artifact attestation (recommended): an attestation step in the build workflow ties the image to its repository and proves its provenance. It works whatever the registry.
- Image name (fallback): without an attestation, Dynatrace compares the image name with accessible repositories. The documentation states that this fallback only works for images hosted on
ghcr.io.
In practice: if your images live in another registry and carry no attestation, they will not be matched. Add the attestation to your workflows before expecting a useful filter.
How to check that it works
After installing from the Hub (search for « GitHub Advanced Security », then Install), check two things.
First, ingestion, in Notebooks. The values below come from the documentation.
fetch logs
| filter log.source == "GitHub Advanced Security"
fetch security.events
| filter event.provider == "GitHub Advanced Security"
| filter event.type == "VULNERABILITY_FINDING"
Second, runtime context, in GitHub: open Packages > Linked artifacts in the organisation. A deployment record means the upload worked. Possible exposure values are PUBLIC_NETWORK, ADJACENT_NETWORK, NOT_DETECTED and NOT_AVAILABLE; for data assets, REACHABLE, NOT_DETECTED and NOT_AVAILABLE.
Prioritising in GitHub with runtime filters
In the organisation’s Dependabot or code scanning list, start from repositories that have a deployment, then narrow down.
The first filter drops everything that does not run. The second isolates vulnerabilities in internet-exposed images. The third targets those that reach sensitive data. Once the list is short, you can use it to create a GitHub security campaign and fix findings in a batch.
For SRE the benefit is direct: less noise, a repair time (MTTR) focused on what touches production, and less toil sorting by hand. To use the same data inside Dynatrace, an article on DQL and Smartscape topology helps you read the workloads involved.
Limits to know
- No findings are ingested until a scan has run, even if the project has open issues.
- For code artifacts,
component.versionmay hold only a bound (for example<1.4), which blocks matching with runtime components. - The image-name fallback only covers
ghcr.io. - Secret scanning alerts default to HIGH risk, which you can change in the advanced options.
- The vendor’s post mentions namespace filtering; the documentation’s example query says that filter is a planned update. I therefore do not take it for granted.
Best practices
- Start with a pilot organisation and a GitHub App installed only on the repositories that matter.
- Add the build attestation first: without it, matching depends on the image name and on
ghcr.io. - Keep severity as a second criterion: a runtime filter does not replace judgement on a critical flaw that is not yet deployed and should be fixed before its first deployment.
- Plan token renewal if you do not use the GitHub App.
Frequently asked questions
Do I need RVA to send context to GitHub?
Deployment records rely on the images Dynatrace detects. RVA is optional, but it adds exposure and sensitive data access, which power the runtime-risk filters.
My images are not on ghcr.io: is it lost?
No, provided you add a build attestation: it works whatever the registry. Without one, name-based matching only works for ghcr.io.
Does it work with Dynatrace Managed?
The documentation I read does not say. The mechanism described relies on Grail and a platform token, so check compatibility in the documentation before any Managed rollout.
Conclusion
Runtime context turns an alert backlog into an ordered list: deployed, exposed, close to sensitive data. Set up attestations, install the GitHub App with the right permissions, run both checks, then filter. Sources: the vendor’s blog post and the Dynatrace documentation.
span.observer is an independent community blog, not affiliated with Dynatrace LLC or GitHub. Trademarks belong to their owners. Features change quickly: check options in the official documentation.
Originally published on span.observer. French version: https://span.observer/fr/alertes-github-advanced-security-runtime/.



Top comments (0)