Hmbown/CodeWhale — reverse-engineered prompt

Build me a terminal coding agent called CodeWhale for DeepSeek V4. I want to run it with a codewhale command and get a clean full screen terminal interface where I can ask it to understand a project, plan changes, edit files, run commands, and show its reasoning and progress as it works.

Please make it feel safe by default. It should have a read only planning mode, a normal agent mode that asks before risky changes, and a YOLO mode for folders I trust. Before each turn it should save a hidden snapshot so I can restore or undo the work later.

Add an auto mode that picks the model and thinking level for each turn. It should keep a clear constitution style prompt so the agent follows user intent, checks evidence, and does not claim success without running checks. Include live cost tracking, post edit diagnostics when language tools are available, parallel sub agents for bigger tasks, and simple install options through npm, Cargo, Homebrew, Docker, and direct release binaries. Look up current docs online if you need to.