now it is good ?

The argument in the initial solution still is called input.

Mate..?

Please review and approve. Argument renamed to n and lambda seems to work.

Looks mostly good, but:
It looks like the argument is called n in all other languages and in the description, so I think it should be better to do the same.
define still works. Although there may be languages where test code is even easier for solutions to mess with than in C++, the name of the reference solution can still be obtained, so it would be preferable to use a local lambda.

Take a look please. Feels like I fixed reference solution name leakage. Added #includes to both solutions. Edge case 2147483646 tested.
I not shure about "cross-referencing"(or how name it) num variable in case of current code.

#define group_by_commas gbc passes the tests. The name of the reference solution can be seen in error messages because it's in the same line where user's solution is called from. One way to prevent this is to put the reference solution in a local lambda or you can do anything else you can think of.
The reference solution is called with the same argument num potentially after user's solution. The solution can be changed to accept it by reference and change the argument to bypass the tests (it's more relevant for harder katas, but still...).
Missing include for std::string in the solution. It probably should be in the initial solution too.
2147483646 is an edge case, so it should probably be tested just in case.

C++ translation kumited. Please review.

Seems good, approved.

Review please. It is good now?
Anyway, how translation can be approved?
UPD: I got how it works. Sorry for bothering.

There are still at least 2 fixed tests in both suites with 0s in them.

Also, randn(1,40) has way too high upper bound. It should only go up to 10 or so.

Can You check if I fix it right?
Not sure how it is work.

Solved this! It was a pleasant experience, thanks to you!
P.S. now i have to realize why test suite crash with my code 🤔
P.P.S. i was have to read the output)

No, it's not a KATA issue. Yes, your code is wrong, and makes the tests suite crash. Consider these inputs:

"george clooney" => "G.C"
"xbEfPNUivto MDBxMEDFevCRGVZVmQCVa" => "X.M"