If you really want to solve your issues, you may need to post a file. We'd only need the bone layer, with its actions, to diagnoses the problem.
If it's commercial work, there are plenty of trustworthy people here you could PM.
Select both foot target bones and rejoin the bone translation channel. Delete all the keyframes, and now you can copy or reference the action and it behaves as expected.
If you copy the action to the mainline, you can then resplit the channel dimensions, if needed.
There does seem to be a bit of a bug here, but I'm not sure of the cause yet.
yeah - I think there's a bug to do with split dimensions in the target bone TAR_F_leg translation. In the action the dimensions are not split, but if you split them you get keys for x and y at each keyed point; however inserting (copy) the action into mainline does not correctly copy the y translation at the intermediate frames.
I **think** the base file mainline is somehow corrupt. The keys for TAR_F_leg stored in frames 1-60 are not the same as in the action; and it also holds split keys which don't have enough y keys. Manually erasing the "split" keys and overwriting the mainline translation keys with the values from the action does seem to work.
there's another issue that I don't think has bearing on the above - but there's a lot of "freeze" keys - and those freeze points also include keys for the smartbones for the head and neck actions -- so there will be conflicts with those actions over the walk.
again probably unrelated, but possibly an anomaly -- the wing base bone jumps on frame 1 -- it has a small animation offset
finally (for now) I note that the file was originally created back in Nov 2016 - which version of moho was that?