This'll be a controversial answer, but this sounds like exactly the sort of thing an LLM should be able to do reasonably well, whether by the human writing the docs and the LLM updating the code accordingly or by the human writing the code and the LLM updating the docs accordingly.