feeding those 1,500-line files into a LLM is a recipe for losing the broader context. instead of JUST pasting code, try generating a
dependency graph using something like
tokei
or
cloc
first to see the actual scale. once you have the structure, use an agentic workflow to map out the
side effects of those database calls specifically.
just don't trust it when it says the code is "clean" ⭐