diskreap: reclaim disk from build artifacts & dep caches, across every language

My disk filled up last week. Not from anything I was actively working on — from the graveyard of projects in ~/code. Forty-odd repos, each carrying a node_modules I haven't touched in months, a target/ from that Rust experiment, a .venv here, a .next there. Tens of gigabytes of stuff that a single command would regenerate, just… sitting there.

The annoying part isn't deleting it. It's finding it. du -sh */node_modules only catches Node. Then you do it again for .venv, again for target, again for __pycache__. Different command, different ecosystem, every time.

So I built diskreap — one zero-dependency command that finds reclaimable build artifacts and dependency caches across every language at once, shows you what's eating your disk, and deletes only what's safely regenerable.

npx diskreap            # scan the current directory

# or

pip install diskreap

Why not just npkill / kondo?

They're good — I used both. But each covers one corner:

• 
  npkill is node_modules-only.


• 
  kondo is great and multi-language, but it's a Rust binary you have to install first.

diskreap runs instantly with npx (or pipx run diskreap), needs nothing installed, and reaches every ecosystem in a single scan: node_modules, .venv/venv, __pycache__, .pytest_cache, .mypy_cache, .ruff_cache, .next, .nuxt, .svelte-kit, .turbo, .parcel-cache, target, .gradle, vendor, dist/build, .terraform, Pods.

Using it

Scan first — it never deletes without being asked:

$ diskreap scan ~/code



  /Users/you/code



     1.2 GB  Rust/Maven build scanner/target         · 5d ago

     412 MB  Node deps        webapp/node_modules     · 3w ago

      88 MB  Python venv      ml-notes/.venv          · 2mo ago

     ────────────────────────────────────────────────

     total reclaimable: 1.7 GB across 12 dir(s)

Then reclaim, with filters to scope the sweep:

diskreap clean ~/code --dry-run        # preview exactly what would go

diskreap clean ~/code --older 60d      # only long-idle projects

diskreap clean ~/code --min 100MB      # skip the small fry

diskreap clean ~/code --yes            # no prompt (CI / scripts)

--json is there too, if you want to pipe the scan somewhere.

The one design decision that mattered: safety

The scary names are the generic ones. node_modules is unambiguous — but target, dist, build, vendor could just as easily be your folders.

So diskreap only treats a generic name as reclaimable when a sibling project file proves its origin: target/ counts next to a Cargo.toml or pom.xml; dist/ and build/ count next to a package.json/pyproject.toml/setup.py; vendor/ next to a composer.json/go.mod. A bare build/ of your own files is never touched. Symlinks are never followed, and clean always confirms unless you pass --yes.

It's also genuinely zero-dependency — pure Node stdlib and pure Python stdlib, same logic in both, byte-for-byte aligned output. No supply chain, nothing to audit.

Try it

npx diskreap          # Node

pipx run diskreap     # Python

• npm:
  


• GitHub: https://github.com/jjdoor/diskreap
  

What's the most disk you've ever clawed back from a single ~/code sweep? And which artifact dir surprised you the most — for me it's always .gradle. 👇