#77 (comment)

Why do you think this is not a good idea?? I believe that this is a necessary option, and I was shocked when I found that this was missed.

Sorry, such inserts have a lot of unwanted side effects when used in larger or integrated projects.

We will close this, because this would introduce an anti-pattern:
Python’s import path is defined at interpreter startup, and PYTHONPATH is the standard way to add import locations. Console-script execution does not automatically treat the working directory as sys.path[0] the same way python -m ... does. That behavior is documented here: https://docs.python.org/3/library/sys_path_init.html

If any code change were acceptable, it would need to be explicit and opt-in, not a hidden global sys.path mutation. But in this case, no code change is needed; the build or test environment should make the extension importable before pybind11-stubgen runs.

For example, in CMake:

add_custom_target(stubs
    COMMAND ${CMAKE_COMMAND} -E env
            "PYTHONPATH=$<TARGET_FILE_DIR:mymodule>"
            ${Python_EXECUTABLE} -m pybind11_stubgen mymodule
            -o ${CMAKE_CURRENT_BINARY_DIR}/stubs
    DEPENDS mymodule
)

The same pattern works for add_test(...) as well (example).