Auditing Claude Code setups to cut context bloat

My notes

Summary

A short-form video describes a purported Claude Code /checkup command that audits and cleans up a setup: pruning unused skills/MCPs/plugins, deduping and splitting oversized CLAUDE.md files into nested per-directory files, checking hook performance, auto-updating Claude Code, defaulting to auto mode, and auto-approving frequently-used read-only commands. The claimed goal is lower baseline context consumption via progressive disclosure.

Key Insight

• Core idea: most context bloat in an agentic coding tool comes from stale or unused skills, MCPs and plugins plus oversized, overlapping CLAUDE.md files loaded in full at every session start.
• The fix pattern described: mine session history to see what is actually invoked, then (1) prune what is unused, (2) dedupe CLAUDE.md content across user, project and local scopes, (3) split a single large CLAUDE.md into nested, directory-scoped files so information loads progressively rather than all at once.
• Secondary claims: hook execution time gets profiled and slow hooks get flagged or optimized; auto-update and auto-mode-by-default get configured automatically; frequently-approved read-only commands get promoted into an auto-approved list from historical approval patterns.
• Caveat: the video is promotional (it links to a paid course) and its own numbered list is internally inconsistent (item “sixth” is listed twice, one step duplicates “tests hooks”), suggesting a rushed or partially auto-generated script. Treat the specific /checkup command name and exact behavior as unverified until confirmed against an actual Claude Code release or changelog. It may not exist as a literal built-in command, or may differ from this description.