One Move: Let Claude Check Its Own Setup

TL;DR Skills and MCP connectors pile up over time, and most people never clean them out. /checkup — an alias for the older /doctor — looks at your Claude Code setup, finds what's actually unused, and proposes a cleanup plan. The move is running it occasionally as maintenance, instead of only reaching for it when something's visibly broken.

↺ From previous weeks

Turn a repeating prompt into a skill — package it once, let Claude reach for it via a trigger description instead of you retyping it.

Give it a job that runs without you — a scheduled task saves the prompt and fires on its own cadence, cloud-only, with no memory between runs.

The move

Notice the drift: skills you tried once, MCP servers you connected for a single project, a CLAUDE.md file that's grown past what's still true.
Run /checkup (or claude doctor from a terminal for a read-only version): it checks install health, finds unused skills, MCP servers, and plugins against what you've actually used, and can split or trim CLAUDE.md files that have grown stale.
Read the plan before approving anything: it always shows proposed changes first — the value is in actually reading that list, not just accepting it.

▶ Try it now — run a health check on your own setup ~cheap

/checkup

Where it breaks — the three edges worth knowing

1. Claude Code specific. There's no equivalent inside claude.ai chat or the API — this is a feature of the CLI tool, not of Claude generally.

2. The CLAUDE.md trim step needs a fairly recent version (2.1.206 or later). An older install still runs the rest of the checkup, it just skips that one part.

3. The safety model depends on you actually reading the plan. Skimming past it and clicking approve defeats the entire point of "shows you first" — the plan is the safeguard, not a formality.

Under the hood

What actually runs when you call /checkup — for when you want to read the machinery, not just use it.

Open the hood: what it inspects, how it's built, and the primary sources

The lever: /doctor, aliased as /checkup — itself a bundled skill, not a separate program bolted onto Claude Code.
What it actually calls: Inspects local install state (PATH, config files, duplicate installs), diffs configured skills, MCP servers, and plugins against actual usage, and can rewrite or split CLAUDE.md files — every change gated behind a shown plan before anything is touched.
The standard underneath: The same SKILL.md mechanism as any custom skill (see the first page in this run) — Anthropic just ships this particular one by default.
Primary sources: code.claude.com/docs/en/commands — the full command reference, including the /checkup//doctor alias and the exact list of what each check covers.

Watch & listen

Read · ~6 min Commands reference (code.claude.com) — the complete diagnostic list and what triggers each check.

⚡ What did this spark?

A question this page raised, a thing that didn't work, or a topic you want next. It lands in Learn Sparks — and the weekly run turns it into a future page or a quick answer in the digest. You choose which on the form.

⚡ Send a spark

Opens the Learn Sparks form in a new tab — about ten seconds, one required field.