Image: Anthropic / github.com Claude Code 2.1.260 and 2.1.261 Add /diff Panel, /skill-doctor, and Prompt Cache Diagnostics
Two updates in two days: Claude Code now shows uncommitted diffs beside your conversation, can tell you which skills you never use, and explains exactly why your prompt cache keeps missing.
Anthropic shipped two Claude Code updates back to back this week: version 2.1.260 on September 3, and version 2.1.261 on September 4. Together they add three features developers will notice immediately, along with a long list of fixes.
/diff shows your changes beside the conversation
Version 2.1.260 adds a /diff command that opens a diff panel in fullscreen mode, showing all uncommitted changes beside your conversation. It’s a simple idea that removes a constant context switch: you no longer have to alt-tab to a terminal or source control sidebar to see what Claude has actually changed.
The panel stays live as Claude works, so you can watch changes accumulate during a task without breaking your focus.
Prompt cache miss explanations
The same update added cache miss diagnostics to /cost and the status line. Instead of just showing you a low cache hit rate, Claude Code now explains why the cache keeps missing: tool definitions changed, the model was idle past the TTL, the context exceeded cache limits, and so on.
For anyone running Claude Code on large projects with long sessions, this is useful. A bad cache hit rate on Fable 5.1 meaningfully affects cost per token. Knowing the cause makes it fixable.
Also in 2.1.260: a text-form /advisor command (previously UI-only), improved support for the 1 million token context on Fable 5.1’s auto-compact, and the /ultrareview timeout extended from 30 to 45 minutes.
/skill-doctor shows unused skills
Version 2.1.261 adds /skill-doctor, which shows you which loaded skills a session never used and what each one costs in context. If you have a set of skills loaded globally, some of them may be contributing nothing except token overhead. /skill-doctor makes that visible so you can prune accordingly.
The update also added an “Organization policy” line to /status and claude doctor that explains why managed policies couldn’t be loaded. Previously, if a proxy was blocking the endpoint or credentials were misconfigured, you’d just see the policy wasn’t applied. Now you get a reason.
Two other developer-facing additions: bashOutputMaxChars and taskOutputMaxChars settings let you raise how much command and background-task output Claude receives inline, up to 128K characters (previously there was a much lower default that would cut off long build outputs). And --append-subagent-system-prompt-file lets you pass large subagent system prompts from a file instead of the command line.
Bug fixes
The fix list for both versions is long, but some worth noting:
In 2.1.260: file permission rules with parentheses were being silently dropped, which could cause Bash sandbox to behave unexpectedly. A separate issue had Bash permission checks auto-approving zsh commands that contained command substitution. Both are fixed.
In 2.1.261: typed or pasted characters could land out of order or drop during fast input or key repeat. Also fixed: VS Code sessions now properly show the model picker with managed org model lists without requiring a double reload.
The VS Code extension in 2.1.261 also adds an “Archive session” option to the session list right-click menu, and a fold button on permission prompts so you can read the conversation without dismissing the prompt.
Sources:
- Claude Code Changelog — Anthropic
- Claude Code v2.1.261 Release — AI/TLDR