For that you'd be better off having the LLM write TODO stubs in the codebase and search for that. In fact, most of the recent models just do this, even without prompting.