I prefer the simple open() in this and similar cases. But it's no deal breaker for me. The rest looks good.

This seems like a straight up downgrade.
os-listdir (PTH208)

Only one other change of os.listdir --> Path.iterdir seems like an actual improvement. (in tests/mypy_test.py)

This seems like a straight up downgrade.
os-listdir (PTH208)

Only one other change of os.listdir --> Path.iterdir seems like an actual improvement. (in add_third_party_files above)

This could also be:

which reads clearer to me

I agree, to me as well.

Oh, haha, I know why this breaks. The suffix is .pyi\n 😅

I went with .strip() to keep the readability of .removesuffix(".pyi")

FWIW, this could also be:

    return {
        associated_package
        for distribution in stub_distributions
        for external_req in read_dependencies(distribution).external_pkgs
        for associated_package in _get_pkgs_associated_with_requirement(external_req.name)
    } | {
        # Exclude the distribution and file extension from path
        line.split("/", 2)[2].removesuffix(".pyi")
        for line in EXCLUDE_LIST.read_text().splitlines()
        # Skips comments, empty lines, and stdlib files, which are in
        # the exclude list because pytype has its own version.
        if line.startswith("stubs/")
    }

(not sure for line in EXCLUDE_LIST.read_text().splitlines() is really much better than with EXCLUDE_LIST.open() as f: for line in f:, it's was more as a personal exercise to spot where comprehensions could be used. And could doesn't mean should XD)

I wouldn't want to do fairly extensive changes like this in a PR that's mostly about introducing Path. (Also, the code above is a bit too clever for my taste.)

Yeah agreed. I don't wanna push this code-change.

Speaking of changes not directly related to introducing Path (or rather, enforcing it). There's still a few changes in the pytype script that were more a result of looking at it more closely. I could split those off as well if you prefer.