They say that, in relation to the text of a PR, commit, forum post, or similar, "the output must be your own words, explanation, description, etc., not a verbatim dump of an LLM's output". If you must, it sounds like you could still use an LLM to identify any aggression or confrontation in your text, but then fix it yourself. The other advantage of that approach is that you'll become more aware of your communication patterns, and how to remedy harmful ones.