FIELD NOTE / ENGINEERING REPORT

I Rebuilt My Vagrant Boxes for Rising OS CVEs, and Found Vagrant Cloud Shutting Down

Notes on kube-ready-box v0.3.0: a rebuild that took shipped Vagrant boxes to zero fixable CVEs, the HCP Vagrant shutdown discovered during upload, the move to GitHub Release + Pages, and a weekly scan for vulnerabilities that are not patched yet.

Vagrant · HashiCorp · Security · CVE · Trivy · GitHub Pages · DevOps · Kubernetes
Published
2026-11-01
Reading time
5 min
Source systems
1
Record type
Field note
PROVENANCE

Trace the lesson back to the systems that created the context.

The number of CVEs reported against Linux distributions has grown noticeably. My Kubernetes-ready Vagrant boxes, kube-ready-box (Ubuntu 24.04/26.04, Rocky 9, NixOS), were no exception, so I rebuilt all of them. When I went to upload the new boxes, Vagrant Cloud was in the middle of shutting down. This post records what changed and what I learned along the way.

1. CVE response: down to zero fixable vulnerabilities

I scanned the package databases inside the boxes with Trivy, counting only vulnerabilities the distribution has already fixed (--ignore-unfixed).

TargetFixable CVEs
Shipped Rocky 9 0.1.0 line438 (5 critical)
Unpublished Ubuntu 26.04 build1,141 (2 critical)
Every box rebuilt for v0.3.00

399 of the 438 Rocky CVEs came from the installer's old kernel. The build updated the kernel but never rebooted, so the old one stayed in the image. The rebuild now removes every kernel except the running one. I also bumped the bundled tools (gVisor, yq) and the pinned nixpkgs for NixOS.

Every rebuilt box was booted on VMware Fusion (NixOS on QEMU) and checked: root disk growth on first boot, swap off, zero failed systemd units, and the kernel matching the build record.

2. Found during upload: Vagrant Cloud is shutting down

Uploading the rebuilt boxes failed: vagrant cloud version create returned "resource creation is currently disabled". HashiCorp is ending HCP Vagrant (formerly Vagrant Cloud).

DateMilestone
2026-10-01No new boxes or registries
2026-11-02End of support and maintenance
2026-12-31End of operations; remaining deployments decommissioned

If your Vagrantfiles or CI reference Vagrant Cloud boxes as config.vm.box = "user/box", they need a new home before the end of the year.

3. Moving to GitHub Release + GitHub Pages

The .box files are now GitHub Release assets, and the catalog (the JSON metadata Vagrant reads) is served from GitHub Pages.

config.vm.box     = "dasomel/ubuntu-24.04-ext4"
config.vm.box_url = "https://dasomel.github.io/kube-ready-box/boxes/ubuntu-24.04-ext4.json"

The catalog lists versions, providers, architectures, download URLs and SHA-256 checksums, so checksum verification, vagrant box outdated and vagrant box update work just as they did on Vagrant Cloud. The format is in the Vagrant documentation.

What I found while moving:

  • Don't host the catalog on raw.githubusercontent or as a Release asset. Vagrant 2.4.9 treats a URL as a catalog only when the response Content-Type is application/json. GitHub Pages qualifies. A github.io URL that 301-redirects to a custom domain also worked in a real vagrant box add.
  • Each Release asset must be under 2 GiB (GitHub docs). The old boxes exceeded that. The root volume is now 16 GiB at install time and grows to the full disk (about 1 TB) on first boot, which brought the boxes down to 1–2 GiB. Along the way, Ubuntu 26.04's uutils dd printed a different error message than GNU dd and broke the zero-fill step.
  • Moving to S3, Azure Blob or IBM COS? See HashiCorp's migration guides.
  • To keep old versions, download them before the end of the year. HashiCorp published an export CLI, vagrant-registry-export.

Why the version is 0.3.0

The Ubuntu boxes reached 1.1.0 on Vagrant Cloud, but 1.1.0 was never a GitHub Release asset and disappears with Vagrant Cloud. This release therefore continues the 0.2.x line as 0.3.0. Vagrant ranks 1.1.0 above 0.3.0, so if you ever pulled 1.1.0, remove it once:

vagrant box remove dasomel/ubuntu-24.04-ext4 --box-version 1.1.0

Rocky (0.1.1) and NixOS (0.2.2) keep their own version lines.

4. Unpatched vulnerabilities: checked automatically every week

Not everything is fixed. The kernel vulnerability CVE-2026-98432 has no Ubuntu or Rocky fix yet. Items like this are tracked as "open" in the security response history.

Relying on someone to remember to check for fixes is fragile, so I added a scheduled GitHub Actions workflow:

  1. Every Monday it downloads the boxes currently shipped, exactly as the Pages catalog points to them, and verifies their checksums.
  2. It extracts the package databases (dpkg/rpm) from the disk images, read-only.
  3. Trivy counts fixable CVEs. Any critical or high finding fails the workflow, which is the signal to rebuild.

NixOS has no dpkg/rpm database, so Trivy cannot scan it. That box is reported as UNKNOWN, not as passing, and is maintained by bumping its nixpkgs pin.

Wrap-up

Vagrant keeps working after Vagrant Cloud is gone; only where the boxes are hosted changes. This round also left the project with a way to keep checking whether the artifacts it ships are still safe today.

  • Repository: dasomel/kube-ready-box
  • Release: v0.3.0
  • Follow-up: v0.3.1 (2026-11-02) rebuilds only the Ubuntu boxes to drop the installer's kernel arguments from GRUB and the leftover installer kernel. Fixable CVEs are still 0. On 2026-11-03 Ubuntu VirtualBox amd64 boxes were added to the same release, so Intel/AMD hosts are covered too.
  • Distribution: docs/box-distribution.md
NEXT PATH / VERIFY IN CONTEXT

Follow the lesson back into the system.

The article is one observation. The project record holds architecture, development pulse, proof and the wider engineering context.

Kube-Ready-Box