artifact-scythe 0.6.0

CLI tools for cleaning artifacts and builds

Artifact-Scythe

Reclaim disk space by harvesting the build artifacts you forgot about.

MP4 version

scythe is a Python CLI that walks your projects directory, identifies each project's ecosystem (Node, Python, Rust, Java, Go, Ruby, .NET) from its marker files, and locates the bulky build artifacts that ecosystem leaves behind — node_modules, .venv, __pycache__, target/, build/, and the rest. It reports how much space each one is wasting and, when you tell it to, deletes them.

It is safe by default: every clean previews what is about to happen and asks for confirmation, and a --dry-run mode lets you check the plan before running for real.

Why

A laptop used daily for a year typically holds 30–80 GB of stale node_modules, abandoned virtualenvs, cached compiler output, and CI scratch directories. None of it is load-bearing — but finding and removing it by hand across dozens of project folders is tedious and error-prone. scythe does it in two commands.

How does this compare to npkill / kondo / cleanpy?

Existing tools cover one ecosystem each, well. scythe covers a mixed-stack ~/projects folder in one workflow.

Tool	Ecosystems covered	UX	Filters	Reports	Distribution
npkill	Node only (node_modules)	Interactive TUI	Sort by size / age	None	npm
kondo	Rust + a handful (Node, JS, Java, Haskell, …)	Interactive prompt	Older-than	None	Cargo, Homebrew
cleanpy	Python only	CLI	Cache types	None	pip
scythe	Node, Python, Rust, Java (Maven + Gradle), Go, Ruby, .NET (Swift planned)	CLI today, TUI mode scythe ui planned	--only, --older-than, --min-size (--ignore planned)	JSON / CSV export, dry-run report, recoverable trash + scythe restore	pipx, pip, Docker (multi-arch), standalones planned

When to reach for which:

• Only have a node_modules problem? npkill is purpose-built.
• Mostly Rust? kondo is great.
• Mixed-stack folder, want filters / reports / scripting / CI integration, or you also need to clean Python virtualenvs and Java build dirs in the same pass? That's where scythe is meant to live.

Quick start

pipx install artifact-scythe          # one-line global install

scythe scan ~/projects                # see what's eating your disk
scythe clean ~/projects --dry-run     # preview the deletions
scythe clean ~/projects               # do it (with confirmation)

Install

With pipx (recommended)

pipx installs scythe into its own isolated virtual environment and exposes the scythe command globally on your PATH. It's the right tool for Python CLIs: no clash with your project deps, no sudo, one command to upgrade.

pipx install artifact-scythe          # install
pipx upgrade artifact-scythe          # update later
pipx uninstall artifact-scythe        # remove cleanly

Don't have pipx yet? python -m pip install --user pipx && python -m pipx ensurepath (restart your shell once).

With pip

If you don't want the isolated install, plain pip works too:

pip install --user artifact-scythe    # user-level install

Naming note

The PyPI distribution is artifact-scythe (the scythe slot on PyPI was taken), but the installed command and the Python module are both scythe. Scripts written against scythe ... keep working unchanged.

With Docker

If you'd rather not install anything locally (CI agents, throwaway VMs, or just trying it out), the official image is published on GHCR for both linux/amd64 and linux/arm64:

# scan the current directory
docker run --rm -v "$PWD":/work ghcr.io/elielmengue/scythe:latest scan /work

# clean with a dry-run
docker run --rm -v "$PWD":/work ghcr.io/elielmengue/scythe:latest \
    clean /work --dry-run

Tags follow the PyPI release: :latest, :0.5.3, :0.5, :0. The rolling :edge tag tracks main.

From source

git clone https://github.com/elielMengue/scythe.git
cd scythe
pip install -e ".[dev]"

Usage

scythe scan — discover projects and measure artifacts

scythe scan .                                  # current directory
scythe scan ~/dev --depth 2                    # bound recursion depth
scythe scan ~/dev --only node,python           # filter by ecosystem
scythe scan ~/dev --older-than 30              # only artifacts older than 30 days
scythe scan ~/dev --min-size 500MB             # only artifacts at or above 500 MB
scythe scan ~/dev --format tree                # table | tree | compact | json
scythe scan ~/dev --format json -o report.json # also csv via .csv suffix

scan is read-only. It produces a report; nothing is deleted.

scythe clean — delete detected artifacts

scythe clean ~/dev --dry-run                   # simulate (always do this first)
scythe clean ~/dev --trash                     # recoverable cleanup (undo with `scythe restore`)
scythe clean ~/dev --interactive               # pick projects manually
scythe clean ~/dev --only rust                 # only Rust target/ directories
scythe clean ~/dev --older-than 30 --dry-run   # only target stale artifacts
scythe clean ~/dev --min-size 1GB --dry-run    # only large artifacts worth deleting
scythe clean ~/dev --force                     # skip the confirmation prompt
scythe clean ~/dev -o run-report.json          # export a JSON report

clean runs the same scan first, prints a summary, then either prompts before deleting or executes immediately depending on flags.

By default deletion is permanent — files are unlinked, not moved to the OS bin. Pass --trash to route them through scythe's recoverable trash instead, then use scythe restore to bring them back.

scythe restore — undo a clean --trash run

scythe restore --list                # show recoverable runs (id, date, items, size)
scythe restore                       # undo the most recent --trash run
scythe restore 20260502-153000-123456  # undo a specific run by id

Trashed runs live under the per-user data dir (%LOCALAPPDATA%\scythe on Windows, ~/Library/Application Support/scythe on macOS, $XDG_DATA_HOME/scythe or ~/.local/share/scythe on Linux). A run that's already been restored, has a missing trash payload, or whose destination has been re-created since the clean, is reported as skipped rather than failing.

scythe info

Prints the installed version and the list of supported ecosystems and patterns.

Supported ecosystems

Ecosystem	Marker files	Artifact patterns
Node.js	package.json, yarn.lock, pnpm-lock.yaml	node_modules, dist, build, .next, .nuxt, out, .cache, .parcel-cache, .turbo, coverage
Python	requirements.txt, setup.py, pyproject.toml, Pipfile, poetry.lock	.venv, venv, env, __pycache__, .pytest_cache, .mypy_cache, .ruff_cache, .tox, *.egg-info, dist, build, .eggs, htmlcov
Rust	Cargo.toml, Cargo.lock	target
Java (Maven)	pom.xml	target, .m2/repository
Java (Gradle)	build.gradle, build.gradle.kts, settings.gradle	build, .gradle, out
Go	go.mod, go.sum	bin, pkg, vendor
Ruby	Gemfile, Gemfile.lock, .ruby-version	vendor/bundle, .bundle, tmp
.NET	*.csproj, *.fsproj, *.vbproj, *.sln	bin, obj, packages, .vs

The --only flag accepts both canonical names (node, python, java_maven, java_gradle, dotnet, ...) and short aliases (py, js, rs, golang, .net, cs).

Safety

• clean deletes via shutil.rmtree / Path.unlink — files are gone, not in the trash bin. Run --dry-run first when in doubt.
• Source-control dirs (.git, .svn, .hg, .bzr), editor metadata (.idea, .vscode), and OS metadata (.DS_Store, Thumbs.db) are skipped during traversal.
• Symlinks are not followed unless --follow-symlinks is passed.
• Use --depth N on very large filesystems to avoid runaway scans.

Development

Python 3.10+ is required.

git clone https://github.com/elielMengue/scythe.git
cd scythe
pip install -e ".[dev]"

pytest -v                       # full test suite
pytest --cov=scythe             # with coverage
pytest tests/test_scanner.py    # a single file

CI runs the test suite on Linux, macOS, and Windows for Python 3.10, 3.11, and 3.12. See .github/workflows/ci.yml.

Contributing

Issues and pull requests are welcome.

To add support for a new ecosystem, extend two places:

• PROJECT_MARKERS in scythe/scanner/scanner.py — how to recognise the project.
• ARTIFACT_PATTERNS in scythe/detector/detector.py — what to clean.

Then add a fixture and an assertion in tests/test_detector.py that exercises the new pattern. Keeping these two maps in sync is the core invariant of the codebase.

Roadmap

Shipped

• Configuration & foundations
• Directory scanner
• Artifact detection — 8 ecosystems
• Rich-based output (table / tree / compact / JSON)
• Cleaning engine (--dry-run, --interactive, --force, JSON report)
• Filters: --only, --older-than, --min-size
• Recoverable trash + scythe restore — scythe clean --trash moves artifacts under a per-user data dir and writes a per-run manifest; scythe restore undoes the most recent run (or a specific one by id). (v0.6.0)
• Distribution: PyPI (pipx), Docker (multi-arch GHCR)

Safety & UX

• scythe ui — interactive TUI mode (Textual) — full-screen browse-and-clean experience: filterable project list, expandable artifact tree per project, live total-size readout, item-level toggles, and an undo stack that pairs naturally with trash-mode. The CLI stays for scripts and CI; the TUI is for exploration.

Filters & customization

• --ignore PATTERN — extra ignore patterns on top of the built-in defaults (e.g. --ignore "*.archive,~/projects/keepme").
• Config file — pyproject.toml [tool.scythe] and/or ~/.scytherc to persist --only / --ignore / depth defaults per machine.

Distribution & polish

• Standalone binaries on GitHub Releases for users without Python:
  • macOS — Apple Silicon (arm64) and Intel (x86_64)
  • Linux — x86_64 and arm64 (glibc); musl/static build for Alpine
  • Windows — x86_64 (.exe)
  • Optional: FreeBSD x86_64
• Package-manager distribution — Homebrew tap, Scoop bucket, winget manifest, and an AUR package.
• Shell completions — scythe completion {bash,zsh,fish,powershell}.

Telemetry & quality

• Lifetime stats — track total space reclaimed across runs and surface it in scythe info.
• Comprehensive integration tests — broader scanner/cleaner coverage (large directory trees, permission edge cases, symlink loops).

See CHANGELOG.md for the release history.

License

MIT © Eliel MENGUE