Repository navigation
The release check no longer reports screens it reached as unreachable. A run records each
screen with a fingerprint of that screen's contents, so two states of one screen count
separately towards coverage. The comparison against a baseline matched name and fingerprint
together, and the fingerprint changes between runs — so every baseline screen looked unreached
and each produced a high finding. On tapp's own fixture it raised three. A genuine loss of
coverage is still reported, once per screen.
The production dependency advisories are cleared. Six known weaknesses, two high, all
reaching the package indirectly through the Model Context Protocol SDK. Seven transitive
packages moved; no direct dependency changed.
Together these are why the public repository's own test run had failed on every release since
24 October. The generated status page now prints that run's result, alongside every place tapp
is published, so the next gap is read rather than discovered.