Claude Code 2.1.149: /usage Now Shows What's Actually Eating Your Quota

Anthropic shipped Claude Code v2.1.149 on May 22, followed by v2.1.150 on May 23 with internal infrastructure work only.

The headline feature in 2.1.149 is a much more useful /usage command.

/usage per-category breakdown

Before 2.1.149, /usage told you how much of your rate limit you’d consumed. That was it. If you were running heavy agentic workflows with skills, subagents, MCP servers, and plugins, you had no way to tell which component was responsible for most of your usage.

Version 2.1.149 breaks it down by category: skills, subagents, plugins, and per-MCP-server cost are each listed separately. If a specific MCP server is burning through your quota, you’ll see it. If most of your usage comes from subagents launched during long runs, that’s visible too.

This is practical for anyone on a Pro or Max plan managing their 5-hour windows, and for teams running automated pipelines trying to understand where their credits go.

The /diff detail view now supports keyboard scrolling. Arrow keys, j/k (Vim-style), PgUp/PgDn, Space, Home, and End all work to move through a diff without touching the mouse. For reviewing large diffs during an agentic session, this removes the friction of switching between keyboard and mouse.

GFM task list checkboxes

Markdown output now renders - [ ] todo and - [x] done as proper GFM task list checkboxes instead of plain bullet points. If you use Claude Code to generate task lists or planning documents with checkbox syntax, those render correctly in the output view now.

Enterprise: load all cloud MCP connectors

A new managed setting — allowAllClaudeAiMcps — lets enterprise admins load all claude.ai cloud MCP connectors alongside whatever’s in managed-mcp.json. Previously you had to enumerate each cloud connector explicitly. This is a convenience setting for organizations that want to make all available connectors accessible without listing them out.

Security fixes

2.1.149 includes several fixes for PowerShell and shell permissions on Windows:

PowerShell permission bypass where built-in cd functions (cd.., cd\, cd~, X:) changed the working directory without detection
PowerShell prefix/wildcard allow rules not pre-approving native executables and scripts
Permission-analysis gap with stale variable-tracking values for PWD, OLDPWD, and DIRSTACK
Git worktree sandbox write allowlist now covers only the shared .git directory, with hooks/ and config explicitly denied

There’s also a fix for find in the Bash tool exhausting the macOS system file/vnode table on large directory trees. This could cause Claude Code to burn through macOS’s file descriptor limit during broad searches across large repos.

Other fixes

A lot of smaller issues got addressed in this release:

Managed-settings approval dialog was freezing the terminal after startup acceptance
/ultraplan and remote session creation were failing with “Could not capture uncommitted changes”
Thinking spinner was staying amber across tool calls
Status bar was showing baseline /effort instead of the applied effort level
Ctrl+O transcript view was freezing
Prompt history editing was losing changes on arrow key navigation
Race condition where submitted prompts appeared twice in history
/config was reporting phantom changes on exit
/insights was crashing with missing session metadata
“Jump to bottom” pill wasn’t dismissing in fullscreen mode

Updating

Run claude update or install from claude.ai/code to get 2.1.149. Version 2.1.150 is infrastructure-only with no user-facing changes, but it’ll apply automatically on the next update.

Sources: v2.1.149 release, v2.1.150 release