2 min read AI-generated

Claude Code 2.1.206: /doctor Now Trims Your CLAUDE.md

Copy article as Markdown

27 changes, but no big stage — version 2.1.206 is a classic stability update. The most interesting detail: /doctor now suggests removing redundant lines from your CLAUDE.md.

Featured image for "Claude Code 2.1.206: /doctor Now Trims Your CLAUDE.md"

Claude Code 2.1.206 is out — and it’s exactly the kind of release I like: no new model announcement, no big stage. Instead, 27 changes that mostly clean up the spots where you get stuck in daily use. 9 improvements, 18 bug fixes, no breaking changes.

The highlight: /doctor declutters your CLAUDE.md

The actual new feature is small but clever. When you run /doctor now, Claude Code inspects your committed CLAUDE.md and suggests which lines you can cut — namely everything Claude could already figure out just by reading the code.

Typical candidates: directory structures, the list of dependencies from package.json, simple build and test commands, formatting rules that already live in your linter config. None of that needs to be duplicated in CLAUDE.md.

Important: it deletes nothing automatically. These are only suggestions — the call stays yours. And whatever Claude can’t infer from the code — the why behind decisions, historical baggage, project-specific gotchas — should stay in.

The benefit is real: CLAUDE.md gets loaded into context every single turn. Less ballast means fewer tokens burned — and important instructions no longer drown in pages of text.

The rest: friction removed

The bulk of the release is fixes to things that are genuinely annoying when they break. An expired login used to throw the misleading error “issue with the selected model” across all models, instead of simply prompting /login — fixed. claude --resume and --continue were unresponsive to keyboard input at startup — fixed. MCP servers that set their own request_timeout_ms were still hitting the 60-second default timeout — fixed.

On top of that: background agents now update in the background immediately after an update, instead of running a slow upgrade when you reattach. And EnterWorktree now asks for confirmation before entering a git worktree outside .claude/worktrees/ — a sensible seatbelt.

My take

Releases like this say more about a tool’s maturity than any new feature show. If you use Claude Code all day, you don’t feel the big announcement — you feel the twenty small moments where something stops getting in your way.

The /doctor check on CLAUDE.md is my favorite here. I’ve seen CLAUDE.md files that grew into novels over months — half stale, half redundant. A tool that tells you “this can go” is overdue. Worth installing the update and running /doctor once.


Sources: