After reading the claude-code source code, I found that its tool calling optimization relies not on a single trick, but on multiple layers of mechanisms working together.

I prefer agent memory on demand, not by default. The real difficulty is deciding what to keep, when to keep it, and how to stop memory from turning into noise.

Claude Code is great for interactive exploration. Once you need long-running, recoverable, auditable agent execution, code-level control becomes much harder to avoid.

Yes. It is git worktree used as isolation for agent workflows: one repository, multiple directories, and far fewer workspace collisions.