The core thesis is that agentic discovery is a waste of a model's "intelligence budget." By letting the engineer pick what and where, the AI can focus on just writing the code correctly. This mostly involves: - Robust parsing to ensure the LLM's intended changes actually get applied - UX/workflow convenience so you're wasting minimal time interfacing with the tool - automatic validation, retries, backup, etc. to avoid wasting time bookkeeping

Happy to answer any questions. I tried to add some onboarding before I pushed but it's very much built for my own workflow.