Opening this again as https://feedback.vrchat.com/open-beta/p/add-a-public-vrcsdk-api-method-for-silently-invoking-the-constraint-upgrader is not the same issue and was probably inappropriately merged with Fuka's post.
In 3.7.0-beta.1 there is an auto fix for converting existing constraints to VRC Constraints. This enumerates animation clips in the animator controllers attached to the avatar descriptor, and rewrites their curves to match the new constraints. However, it will not be able to identify animations merged using nondestructive utilities. Thus, when the user runs the auto fix, they'll end up with components converted, but animations left unchanged (or vice versa).
While Senky's post is a useful one (we could invoke the converter at build time to help address this issue), being able to invoke the conversion process from our tools does not resolve the issue of the auto fix action, when invoked manually by the end-user from the VRCSDK UI directly, missing animation clips that need adjustment. As such, an API is needed to allow nondestructive tools help the VRCSDK discover any additional animation clips that need to be processed during this manual auto fix invocation.
Note also that both VRCFury and Modular Avatar perform animation path rewriting, so your API will need to be amenable to this as well. MA currently only does prefix rewriting, but I have plans to add more advanced forms of path rewriting in the future.