amal66 puts Mike's riskiest parts under test
This fork is treating an automated test suite as a shippable feature, not an afterthought.
amal66 has built up a growing battery of automated checks - past 118 of them - aimed squarely at the places where a legal-AI tool quietly goes wrong. The coverage targets the sensitive edges: encrypted user keys, the tokens that gate secure file downloads, and the routing that decides which AI provider handles a given request, including making sure an unrecognised model fails safely instead of guessing.
The team also reworked how the app starts up so the whole thing can be exercised end-to-end, not just in pieces. The practical payoff is that these tests double as documentation - they pin down how the trickiest helpers are actually meant to behave, so future changes don't silently break them.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?