Your framing of Claude Code vs Cursor is exactly right - one-shot solutions vs iterative refinement. I've landed on a similar split in my workflow.
The nested Claude.md tip is gold. I started with a single file but it got unwieldy at 200+ lines. Now I have domain-specific instruction files that Claude pulls in based on context.
The Explore → Plan → Code flow you describe has become non-negotiable for me. Before I learned this, I'd give Claude a task and watch it go in circles. Now plan mode is where I spend most of my prompting effort.
Your framing of Claude Code vs Cursor is exactly right - one-shot solutions vs iterative refinement. I've landed on a similar split in my workflow.
The nested Claude.md tip is gold. I started with a single file but it got unwieldy at 200+ lines. Now I have domain-specific instruction files that Claude pulls in based on context.
The Explore → Plan → Code flow you describe has become non-negotiable for me. Before I learned this, I'd give Claude a task and watch it go in circles. Now plan mode is where I spend most of my prompting effort.
I wrote up my honest assessment of what actually works with Claude Code: https://thoughts.jock.pl/p/claude-code-review-real-testing-vs-zapier-make-2026