jrittelmeyer tidy
Reclaim local disk and keep the working machine healthy. Most stacks have at least one unbounded grower — a build or engine cache with no native TTL or size cap that quietly accumulates gigabytes. This skill's core job is pruning that cache; the rest it surfaces rather than runs, because those need judgment.
Adapter: .claude/ai-dev-kit.config.json (cache, hygiene); a missing
field → ask or skip that section. Per-stack cache locations, growth patterns,
and platform commands: references/hygiene-recipes.md.
1. Prune the build cache (safe, automatic)
- Report first: run the adapter's
cache.sizecommand. - Prune to the cap: run
cache.prune(typically deletes oldest entries until under a ceiling while keeping recent builds for cache hits; prefer a--dry-runpreview when the command offers one). - Report reclaimed space and current disk free (platform commands in the recipes file).
This same prune often runs automatically at two points (so it rarely has much to do here): the checkpoint cadence during active build sessions, and any pre-push backstop the repo wires. This skill is the on-demand / deeper pass.
2. Surface the judgment-required cleanups (do NOT auto-run)
Report each with its reclaim potential and the exact command, then let the user pick:
- Orphaned dev processes on the adapter's
hygiene.devPorts(or the stack's equivalent: dev servers, watchers, emulators, editor daemons) — a stopped background task can leave a process holding a port or lock. Tree-kill by PID; keep containers and the browser. - Stale test data matching the adapter's
hygiene.e2eUserPattern(or the project's seeded-data convention) — accumulated throwaway rows eventually break pagination- and count-dependent tests. A mass DELETE always needs explicit user OK; keep everything that doesn't match the pattern. - Dangling container images / build layers — CI image work leaves large
layers; prune commands are in the recipes file. Keep the running
containers named in
hygiene.keepContainers. - Stack-specific growers the recipes file lists for this project type (engine import caches, dependency caches, old toolchain versions) — surface only what the repo actually uses.
3. Report
One short summary: disk free before → after, cache size before → after, and which surfaced items (if any) the user chose to run. Don't silently perform step-2 actions.