I built a small nightly test that installs popular add-ons into a fresh headless Blender for each current version, enables them, polls every operator, then disables and re-enables them. Results, with the error output for every failure: https://lionit.github.io/blender-addon-check/ and the write-up: https://lionit.github.io/blender-addon-check/report.html
Short version:
- The Extensions platform works. 56 of the top 60 extensions pass on 5.2 LTS and 5.3 beta; the other four need a display or declare their own version limits.
- 4.5 LTS users: 14 of the top 60 extensions no longer install on 4.5 because they now require 5.0+.
- GitHub-hosted add-ons are the risk. 14 of 30 popular ones fail or misbehave, and 13 of those were already broken on 4.5. Blender 5 did not break them; nobody maintained them.
- Most common causes: compiled modules built for Python 3.11 (5.1 moved to 3.13), unregister() that does not undo register(), the removed bgl module, and the Sequence to Strip rename.
It is a loading test, not a feature test: green means installs, enables and re-enables cleanly.
If you develop an add-on, paid or free, you can add it to the nightly run for free here: https://github.com/lionit/blender-addon-check/issues/new?template=test-my-addon.yml. Paid add-ons stay private; only the result is published. I would like to know if this is useful, and what you would want it to check next.