Anthropic's coding assistant Claude Code has reached version 2.1.261. The headline addition is a new slash command, /skill-doctor. It lists the skills that were loaded but never called, along with how much context each one occupies. If your setup has grown heavy from too many skills, you can now trim it based on measured numbers.
/skill-doctor names the skills you never use
Skills in Claude Code let you preload a set of instructions for a particular kind of work. They are convenient enough that people keep adding them, but every loaded skill takes up context. The more context they occupy, the less room is left for the actual task.
Version 2.1.261 turns that hidden cost into numbers. Running /skill-doctor shows which of your loaded skills go unused and what each one costs in context. The real value here is being able to pick pruning candidates from measurements instead of guesswork.
/skill-doctor
The benefit scales with how many plugins and skills you have installed. Simply running it once and checking whether anything unfamiliar shows up is a reasonable starting point.
Let Claude read more output before it spills to a file
On the workflow side, output limits were raised. Two new settings, bashOutputMaxChars and taskOutputMaxChars, control how much command and background-task output Claude receives inline before it gets written out to a file. The ceiling is 128,000 characters.
When you want Claude to read logs or test results directly, handing them over on the spot is faster than saving them and reading them back. For anyone who regularly deals with long output, this is a quiet but useful change.
The release also adds --append-subagent-system-prompt-file, which reads a subagent's system prompt from a file. It is meant for prompts too large to pass on the command line.
Telling you why an organization policy failed to load
Enterprise setups got attention too. The output of /status and claude doctor now includes an "Organization policy" line explaining why your organization's policy could not be loaded. One example given is a proxy that does not pass the endpoint through.
Failures like this usually leave you second-guessing your configuration with no clue as to the cause. A single line naming the reason cuts the time spent narrowing it down considerably.
Behavior changes that lean toward caution
Several smaller behaviors changed. The dangerous-rm safety prompt now also catches rm -rf applied to positional parameters and rm -rf hidden inside double-quoted sh -c scripts.
The scope of auto-approval was revisited as well. A link that packs content into a public diagram renderer's URL is now treated as an upload to that site, so it is no longer auto-approved unless you asked for it. The direction is clear: content leaving your machine should require an explicit instruction.
For input handling, word-editing keys now match Bash. Ctrl+W deletes back to whitespace, Alt+F and Alt+D stop at word ends, and punctuation counts as a word separator. As a result, the keybindingFlavor setting no longer has any effect. The goal appears to be a consistent feel with the terminal itself.
In VS Code, sessions can now be archived
The VS Code extension changed too. The session list's right-click menu now offers Archive session, and Unarchive was given its own icon. Alongside deleting, there is now an option to simply set a session aside.
The MCP servers dialog gained a form for adding a server and an action for removing one, so you can change your configuration without leaving the editor. The model picker is now one flat list of every model, with rows for older model spellings placed last. Permission and question prompts have a fold button, letting you read the conversation behind them without dismissing the card.
Summary
The centerpiece of Claude Code 2.1.261 is /skill-doctor. It reports whether your loaded skills are actually used and how much context they consume, so an overgrown setup can be tidied up on evidence. Alongside it come the new output-limit settings, a stated reason when an organization policy fails to load, a stricter dangerous-command check and narrower auto-approval, and archiving in VS Code. Less a flashy feature release than a round of maintenance that pays off most in heavily used setups.
