Nanako0129/pilotfish — reverse-engineered prompt

I want you to install this pilotfish setup into my global Claude Code config so my main session can stay on the best planning model, while cheaper models handle the routine coding work, searches, docs, and bulk edits, with a separate verifier checking important results before anything is treated as done.

Please read the local install instructions in this repo and follow the recommended path. First inspect my existing Claude Code settings and agent files, then show me a clear merge plan and ask for approval before writing anything. Keep anything I already have that should be preserved, and make the changes in the safest way instead of overwriting stuff blindly.

When you are done, give me a simple summary of what changed, what roles were added, how fallback works if the top model is unavailable, and how I can confirm the setup is working in a normal project. If anything is unclear, check the current docs online and use the repo guidance.