I can see some ways to use this and easily check that the LLM is not hallucinating parts of it, because you can ask the LLM to unminify (or deobfuscate) some component, then request unit tests to be written by the LLM, then humanly check that the unit tests are meaningful and that they don't miss things on the unminified code, then run the tests on the original minified version to confirm the LLM's work, maybe set up some mutation testing if it is relevant.