[finding] DISK standing at 85% (Data 361Gi used) — cumulative, reclamation deferred

[4444J99]

Auto-derived from sysdiag doctor / sysdiag disk, 2026-05-26. Standing across ≥3 sessions.

Signal

△ DISK 85% used — /System/Volumes/Data = 361 GiB used, 66 GiB free. 0 APFS local
snapshots (nothing reclaimable there). $HOME = 214 GiB.

Trust / stratification

%used is the trustworthy headline (no snapshot noise to discount). The finding is real,
not a measurement artifact.

Localized vs cumulative

Cumulative — no single directory is "the bug." Space is accreted across caches, dev
artifacts, media, and old tool versions:

Size	Path	Reclaim posture
12 GiB	~/Library/Application Support/Claude	⚠️ cowork-vm + session data — do not blind-delete
7.7 GiB	~/.cache	regenerable
4.0 GiB	~/Library/Caches	regenerable
16 GiB	/Library/Developer/CoreSimulator/Volumes/iOS_23C54 (98%)	unused sim runtime — xcrun simctl delete unavailable
2.0 GiB	Homebrew cache	brew cleanup
42 GiB	~/.local	audit ~/.local/share/claude/versions (prune to current), pnpm store
5.4 GiB	~/Downloads	user review
1.1 GiB	_diagnostics/captures/raw	gitignored/regenerable (this repo)

Reconciliation

84% → 85% → 86% → 85% across recent snapshots — flat-to-slowly-worsening. The delta is
small; the level is the problem.

Action / close

Safe, regenerable-first reclamation (no user data deleted without review):

1. brew cleanup (~2 GiB) · clear ~/.cache + ~/Library/Caches (~12 GiB)
2. xcrun simctl delete unavailable (frees the 16 GiB sim volume)
3. Prune ~/.local/share/claude/versions to the active version
4. Drop captures/raw/ blobs (regenerable; 1.1 GiB)
5. Review ~/Downloads and the 12 GiB Claude support dir manually (contains real data)

Estimated safe reclaim ≈ 20–30 GiB → moves the headline from 85% toward ~78%. Re-run
sysdiag disk after, and sysdiag reconcile against a pre-reclaim snapshot — the delta
is the proof the loop closed.

[4444J99]

Reclamation attempt 1 — 2026-05-26 (safe regenerable batch)

Ran the regenerable-first batch. Measured result: disk 85% → 85% (no movement).
sysdiag reconcile (snap 17-47-38Z → 18-25-11Z): disk_used_pct 85 → 85 stable.

Freed (cleanly)

• npm cache clean (~1.4 GiB), go clean -cache, pre-commit clean (~361 MiB),
  pnpm store prune (438 pkgs / 18,628 files), captures/raw/ (1.1 GiB).
• ~3 GiB deleted, but df moved only 361 → 360 GiB used — the rest offset by the
  concurrent agent session writing to cache and/or pending APFS purge.

Blocked — these were the big line items, and none landed

• uv cache (3.6 GiB) — uv cache clean timed out on the cache lock; another uv
  process is live (the concurrent opencode/agent session). Did not --force —
  forcing a clean while a process holds the lock risks corrupting its operation. Retry
  when no uv process is running.
• iOS simulator volume (16 GiB, 98%) — xcrun simctl is not on PATH (Xcode CLT
  only, no full Xcode). Needs xcode-select pointed at a full Xcode, then
  xcrun simctl delete unavailable.
• Homebrew cache (2 GiB) — brew cleanup -s freed only 12.7 MiB; the rest is
  current bottles brew won't remove.

Still standing → keep open

The finding is unchanged at 85%. The real reclaim is gated on:

1. retry uv cache clean once no uv process holds the lock (3.6 GiB),
2. fix simctl availability to reclaim the 16 GiB sim volume,
3. manual review (real data, not automatable): ~/Library/Application Support/Claude
   (12 GiB — cowork/session data) and ~/Downloads (5.4 GiB).

Lesson: the projected 20–30 GiB assumed the big targets were freely reclaimable; in
practice the two largest were lock-held / tool-unavailable. The safe automated batch
alone cannot close this finding.