A practical guardrail for this kind of scope creep: ask the model to name the files and behaviors it intends to change before editing, then compare the final diff against the requested acceptance criteria. Any extra change needs a reason of its own. A short non-goals list helps too. In your experience, does scope drift show up in the plan or only when you inspect the diff?
Macro8n
A practical guardrail for this kind of scope creep: ask the model to name the files and behaviors it intends to change before editing, then compare the final diff against the requested acceptance criteria. Any extra change needs a reason of its own. A short non-goals list helps too. In your experience, does scope drift show up in the plan or only when you inspect the diff?