CP.22 suggests a recursive mutex can solve a broken design

The guideline about not calling unknown code while holding a lock is good, but the example that says "Such problem can often be solved by using a recursive_mutex" is not good. Recursive mutexes are a code smell, not a solution.

I don't have a suggestion for improvement.

Separate to the recursive mutex aspect, the first example could be extended to cover the case of templates, where the template argument can run arbitrary code in constructors and destructors. These aren't usually thought of a "callbacks" but they are certainly "unknown code", so worth calling out explicitly.