projects · security-scanner

The platform audits itself.

A home lab accretes containers, dependencies, and config faster than any human reviews them. security-scanner turns that review into a scheduled job: every week it scans every image, repo, and the host itself, then renders the findings as a dashboard with a trend line — so security posture is a number that moves, not a feeling.

The product

Findings, triaged by trend

The severity tiles answer "how bad is now"; the fourteen-scan trend answers the better question — "which direction are we heading, and what caused that spike?"

dashboard — severity counts and the last 14 scans
security-scanner dashboard: severity count tiles and a critical+high trend line over the last 14 scans
The dip and rebound in the trend line is a base-image upgrade landing and new CVEs being published against it — the normal breathing of a patched fleet.

Lessons

What it taught the platform

Bursty ≠ broken

A scanner that pegs the CPU once a week is healthy; a quiet service that averages two cores is not. Telling those apart properly produced the lifetime-average measurement in the monitoring pattern.

False positives need a ledger

An unexplained red tile gets ignored within a month. Every accepted finding carries a written reason, so the dashboard stays credible enough to act on.

Scanning your own stack is humbling

The first full sweep of any honest fleet finds four digits of findings. The discipline isn't zero — it's knowing which seventeen actually matter this week.

Copied