# Suppress vulnerabilities applying Kubernetes context to scans

> Source: <https://github.com/alegrey91/vex8s>
> Published: 2026-09-17 13:39:50+00:00

[(this logo is not AI generated)](https://github.com/alegrey91/vex8s/blob/main/vex8s.png)

Vex8s generates [VEX](https://www.ntia.gov/files/ntia/publications/vex_one-page_summary.pdf) documents by correlating container vulnerabilities with Kubernetes settings to determine which CVEs are actually exploitable in your cluster.

Please note, this is an experimental project. Things might change quickly.

The project aims to assess the exploitability of known CVEs within Kubernetes workloads by combining vulnerability classification and [`securityContext`](https://kubernetes.io/docs/tasks/configure-pod-container/security-context/) analysis.

It is based on the following concept:

- Each CVE is categorized into one or more vulnerability classes ([CWE](https://cwe.mitre.org/index.html) )
- CVE description is processed by an embedded ML [model](https://github.com/alegrey91/vex8s-model) to predict its[*exploitation category*](https://github.com/alegrey91/vex8s-model?tab=readme-ov-file#classification) .
- Both the CWEs and the predicted *exploitation categories* are combined to determine if the CVE is mitigable.
- Each *exploitation category* , maps to a set of Kubernetes settings that can block or reduce the impact.
- By parsing a Kubernetes manifest, we can inspect the container settings to evaluate whether the relevant settings are in place.
- Combining both analyses allows the system to determine if a CVE is exploitable in a given workload configuration.
- If it results in a CVE mitigation, we add this to the final VEX document.

For a more in-depth reading you can consult this paper: [Environment-Aware Vulnerability Suppression Using Kubernetes Security Contexts and VEX](https://github.com/alegrey91/vex8s/blob/main/docs/environmet_aware_vulnerability_suppression_using_kubernetes_security_context_and_vex.pdf)

You can download the latest binary from the [release](https://github.com/alegrey91/vex8s/releases) page.

Or you can build it manually:

```
make build
```

`vex8s` currently supports 2 ways to generate VEX documents:

- 
**passive-mode** : passing an already generated vulnerability report created by`trivy` or`grype` .
- 
**active-mode** : actively scanning the images using`trivy` or`grype` engines and then gereating the document based on the results.

Using `trivy`:

```
# generate vulnerability report.
trivy image --format json --output nginx.trivy.json nginx:1.21.0

# generate VEX document by processing vulnerability report.
vex8s generate --manifest examples/nginx.yaml --report nginx.trivy.json --output nginx.vex.json

# scan again with VEX document to suppress vulnerabilities.
trivy image --vex nginx.vex.json --show-suppressed nginx:1.21.0
```

The same can be applied using `grype`:

```
# generate sbom report.
grype --output cyclonedx-json --file nginx.grype.json nginx:1.21.0

# generate vulnerability report.
grype sbom:./nginx.grype.json --output json --file nginx.grype-vr.json

# generate VEX document by processing vulnerability report.
vex8s generate --manifest examples/nginx.yaml --report nginx.grype-vr.json --output nginx.vex.json

# scan sbom with VEX document to suppress vulnerabilities.
grype sbom:./nginx.grype.json --output table --vex nginx.vex.json --show-suppressed
```

Using `trivy`:

```
# scan the image and automatically generate VEX document.
vex8s generate --manifest examples/nginx.yaml --scan.engine trivy --output nginx.vex.json

# scan again with VEX document to suppress vulnerabilities.
trivy image --vex nginx.vex.json --show-suppressed nginx:1.21.0
```

The same can be applied using `grype`:

```
# generate sbom report.
grype --output cyclonedx-json --file nginx.grype.json nginx:1.21.0

# scan the image and automatically generate VEX document.
vex8s generate --manifest examples/nginx.yaml --scan.engine grype --output nginx.vex.json

# scan sbom with VEX document to suppress vulnerabilities.
grype sbom:./nginx.grype.json --output table --vex nginx.vex.json --show-suppressed
```

Each CVE is classified into one or more *exploitation classes*, which drive the
mitigation decision. `vex8s` supports two classifier engines via `--classifier`:

- **`embedded`** (default): an offline ONNX ML[model](https://github.com/alegrey91/vex8s-model) bundled in the binary. No network access required.
- **`gemini`** : uses Google's Gemini LLM to classify the CVE description.
Requires the`GEMINI_API_KEY` environment variable (optionally`GEMINI_MODEL` ).

```
export GEMINI_API_KEY="your-api-key"

vex8s generate --manifest examples/nginx.yaml --report nginx.trivy.json \
  --output nginx.vex.json --classifier gemini
```

See the [documentation](https://github.com/alegrey91/vex8s/blob/main/docs) — in particular the
[User Guide](https://github.com/alegrey91/vex8s/blob/main/docs/user-guide.md) — for a full walkthrough, the Gemini
classifier setup, and a complete flag reference.

This project was inspired by Akihiro Suda's project [vexllm](https://github.com/AkihiroSuda/vexllm).