The best terminal for Claude Code is one that knows what an agent is

Run Claude Code in a plain terminal and everything works — until you run three of them. Which tab was refactoring auth? Which one is blocked waiting for permission? Did the one in the worktree finish? The classic answer is tmux + git worktrees and a lot of discipline. It's Anthropic's own recommended pattern, and it's exactly what xNAUT builds in natively.

What "built for agents" means in practice

Live status pills. Every agent — claude, codex, gemini, or anything you add to the registry — shows working / done / blocked / waiting in the top bar, fed by real-time hooks. Click the pill, jump to the tab.

A worktree per agent. One modal creates an isolated git worktree and drops the agent into it, configured so git push and git status just behave. Parallel agents never fight over the same files.

Agents annotate their own diffs. The diff viewer shows inline notes agents leave — summary, rationale, confidence — anchored to the exact hunk, updating live while they work.

Agents can drive the terminal back. A localhost MCP server lets Claude Code create tickets, update project docs, and manage the vault — the terminal becomes a tool your agent calls.

Proof it happened. Every session can be recorded into a SHA-256 Merkle chain and exported as a client-ready report — hours you can invoice, cryptographically tamper-evident.

How the options compare

If tmux + worktrees already works for you, keep it — xNAUT is that workflow with the bookkeeping automated and a UI on top. Download for macOS or Windows, free and MIT-licensed.