$ man releaselab

About the lab

Most release notes tell you what changed. They rarely tell you what happens when you upgrade a real install. That's the gap releaselab.dev tries to fill.

What happens to a release

  1. A watcher checks the projects we cover for new stable releases.
  2. An AI agent (Claude, by Anthropic) reads the release notes and decides whether the release is worth a page. Re-tags and trivial patches get skipped.
  3. If it is, the agent creates a fresh machine in the lab, installs the previous version the way the project's docs recommend, adds some data, and upgrades. Then it checks that the service runs, the web UI or API answers, and the data survived.
  4. It writes up what it did and what happened, quoting real command output, and proposes the page.
  5. A human reviews every page before it's published.

Rules the agent works under

The lab

Tests run on a small Proxmox VE cluster at home. Each test gets a throwaway container or VM on an isolated network. It can reach the internet to download software, and nothing else. Machines are destroyed after every run, and anything older than two hours is destroyed automatically.

That also sets the limits: one modest machine per test, a one-off install rather than your exact setup, and no long-running traffic. If your install differs a lot from a default one, treat our result as one data point, not a guarantee.

Verdicts

Something wrong?

If a page is wrong or your upgrade went differently, we want to know, and we'll re-test. Write to hello@releaselab.dev with the page and what you saw.