Claude Code version 2.1.149 shipped on May 22 with a per-category breakdown of the /usage command. Per Claude Code’s official changelog, the breakdown itemizes skills, subagents, plugins, and per-MCP-server cost — the first time the tool exposes which sub-component of a session is driving consumption against your limit.

What the breakdown shows

Until 2.1.149, /usage reported a single percentage against a single limit — useful for noticing that something was burning your budget, less useful for finding which thing. The 2.1.149 release attributes consumption to four sub-components Claude Code dispatches itself: a skill firing inside a session contributes to the skills line; a subagent spawned with the Agent tool contributes to subagents; each registered MCP server reports its own cost separately. The category list is the exact wording from the changelog and is not extensible at the user layer.

Three smaller items in the same release

The same release adds keyboard scrolling to the /diff detail view — arrow keys, j/k, PgUp/PgDn, Space, and Home/End — and renders Markdown output with GitHub-flavored task-list checkboxes (- [ ] todo / - [x] done) in place of plain bullets. For enterprise installations, an allowAllClaudeAiMcps managed setting loads claude.ai cloud MCP connectors alongside managed-mcp.json.

Why this version is the inflection

Claude Code has shipped observability changes for months, but every prior one tracked usage across sessions. Version 2.1.149 is the first that tracks usage within a session at the sub-component level. The shift matters because Claude Code sessions now routinely fan out across skills, subagents, plugins, and several MCP servers — a single number cannot tell you which of those is the line item to throttle.

The reframe — that an itemized bill is itself an admission about how the original product was sold — is in the bill finally itemized.