You can now use parameter constraints (constrain positive, constrain 0 to 1, constrain A to B) in transition paths, much like you do in stationary general eqm or in calibration/estimation.
Behind the scenes the parameter constraints are handled quite differently though. In all the other commands it is by transforming constrained variables into unconstrained and then solving the unconstrained problem (e.g., if a variable x has to be positive then we can replace it with y=log(x), and now y can be unconstrained, and after we solve the y we can get back our original variable as x=exp(y) which is again positive). By contrast, in transition paths we simply look at whether the shooting update goes outside the bounds on the parameter, and if it does then we instead update halfway to the bound (instead of going outside the bounds; halfway is a hyper-parameter); something similar is being done with Anderson acceleration.