Inspiration

Uptime Kuma already knows maintenance is not downtime. A bad change can erase that: status pages drop and on-call gets paged for planned work. We wanted a supervised layer that understands a code change, proves it with a real failing test, and only then builds release evidence a human can trust.

What it does

GuardianLoop sits on a public Uptime Kuma fork. For a merge request it:

Maps impact with a criticality policy of real server paths
Posts a risk review (including “do not promote” while a P0 is open)
Selects a focused regression that fails on a controlled maintenance-window bug
Shows red CI before repair and green CI after a separate fix commit
Packages an image, runs staging smoke, and writes release-evidence.json from real job artifacts
Keeps production behind a manual human gate

How we built it

Node tooling under guardianloop/ (policies, impact, review, test selection, evidence, decision), GitLab CI (.gitlab-ci.yml), Duo flow stubs under .gitlab/duo/flows/, and infra compose for staging. Demo fixture is isolated — the live server does not import it.

Challenges

Getting SaaS docker-in-docker staging smoke reliable; keeping audit triage honest (no fake blockers); and proving red→green with real public job URLs instead of a mocked demo.