The base idea is very cool, but the process of solving a level is long and tiring.
One big issue I see is how finicky the platforming can be. In particular, the fact that you can affect the trajectory of past selves (with weight and collision) is, during the first level at least, only causing mistakes and errors. For example, it's very difficult to estimate how much I should be moving in a given direction and for how long since I don't know exactly how my trajectory will be affected in the future. This becomes even more of an issue if I need to go somewhere else to help another future self afterwards. This causes a lot of frustrating trial-and-error. I think you should not be able to change the trajectory of path selves except in very specific scenarios where the player has a lot of control over it and doesn't need to go back and forth to minmax it.
(Edit for clarification: I think it's less an issue of the platforming itself being bad, and rather that the kind of character controller parameters you're using are not well suited to precise, errorless movement that you will rely on with future selves. Specifically, jump and fall speeds are too fast, I think)
Additionally, it feels like yellow sometimes just has to wait for some time because future selves have to move first, and once again it's very hard to estimate how long you should wait for and causes more trial-and-error. Maybe you could either avoid those kinds of puzzles entirely, or add a better way to estimate time passing (for example, a visible clock or timer, which each step of the solution taking roughly the same amount of time)
Still, I do think the idea has a lot of potential!