Mac Disk Top Consumers: Safe Cleanup Order

When a Mac crosses 80% disk usage, the fastest safe fix is not deleting random Library folders. First identify the biggest storage buckets, then classify each one as safe, review-first, or do-not-touch.

Start With Known High-Impact Buckets

A full filesystem walk can be slow on a nearly-full Mac. Scan these common storage roots first, then fall back to a broader pass only if the largest culprit is still unclear.

• ~/Downloads, ~/Desktop, and ~/.Trash
• ~/Library/Caches and developer caches under ~/.cache
• ~/Library/Developer/Xcode/DerivedData, CoreSimulator, and iOS DeviceSupport
• ~/.ollama/models for local model weights
• Claude, Cursor, Zed, and other AI coding tool state under ~/Library/Application Support
• Docker Desktop storage, including images, build cache, volumes, and VM disk files

Safe, Review-First, Do-Not-Touch

The same size can mean different risk. A 25 GB cache may be rebuildable; a 25 GB database volume may be the only copy of local app state.

• Usually safer: Xcode DerivedData after Xcode closes, package-manager caches, old installers, browser cache with the browser closed.
• Review first: Ollama models, Docker volumes, simulator devices, Claude or Cursor session state, large app support folders.
• Do not automate blindly: synced Dropbox project state, release archives, local databases, active VM images, current checkpoints, and unknown system folders.

Find The Growth Vector

If disk incidents repeat, store a top-N snapshot each time the Mac crosses 80%, 90%, or 95%. The next alert should name what changed, not just say the disk is full.

Example recommendation:
Ollama models grew by 42 GB since the last incident.
Xcode DerivedData is 18 GB and safe to rebuild after Xcode closes.
Docker volumes are 31 GB and require review before pruning.

Cleanup Order

1. Show the single largest exact directory first.
2. Delete one low-risk, regenerable item only after the owning app is closed.
3. Move review-first items to Trash only when the user understands what generated them.
4. Keep model files, Docker volumes, archives, and VM bundles review-first.
5. Record reclaimed bytes so the next disk incident can tell whether the cleanup worked.