Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

uvtask

image image image Actions status PyPIDownloads

A task runner for pyproject.toml scripts, optimized for the uv workflow.

Define commands once in TOML, run them with uvx uvtask from any machine — zero runtime dependencies in your project.

Editor sidebar for VS Code and Cursor: aiopy/vscode-uvtask.

Why uvtask

  • Run anywhere with uvx — no install required; zero runtime dependencies
  • Scripts live in pyproject.toml — under [tool.run-script] or [tool.uvtask.run-script]
  • Compose pipelines without shell glue — chain commands by referencing other script names
  • Pre/post hooks — Composer-style (pre-test / post-test) or NPM-style (pretest / posttest)
  • uv-like CLI — colored output, structured help, and typo suggestions for unknown commands
  • Forward arguments safely — extra CLI args pass through to the underlying command, including JSON and values with spaces

Pick uvtask when you want npm/composer-style project scripts in Python, with a CLI that feels at home next to uv, ruff, and ty.

Quick Start

1. Add scripts to pyproject.toml:

[tool.run-script]
lint = "uv run ruff check ."
test = "uv run pytest"
check = ["lint", "test"]

2. Run them:

uvx uvtask check
uvx uvtask test -k integration   # args forwarded to pytest

For daily use, install once with uv tool install uvtask, then run uvtask directly.

Editor extension (VS Code & Cursor)

List and run scripts from the sidebar with the vscode-uvtask extension (aiopy.uvtask):

# Cursor
cursor --install-extension aiopy.uvtask

# VS Code
code --install-extension aiopy.uvtask

Or install a .vsix from Releases via Extensions: Install from VSIX…. Full steps: install guide.

Features

Configuration formats

Scripts support several TOML shapes:

[tool.run-script]
# Simple string
format = "uv run ruff format ."

# With description (shown in help)
lint = { command = "uv run ruff check .", description = "Check code quality" }

# Multiple commands run in sequence
check = ["lint", "test"]

# Multiline commands
deploy = """
echo 'Building...'
uv build
echo 'Done!'
"""

You can also nest scripts under [tool.uvtask.run-script] if you prefer a namespaced layout.

See this repository's pyproject.toml for a full real-world script catalog.

Command composition

Reference other script names to build pipelines without repeating shell commands:

[tool.run-script]
lint = "uv run ruff check ."
test = "uv run pytest"
static = ["lint", "test"]
all = ["static"]

Running uvx uvtask all executes lint then test.

Hooks

Define hooks that run automatically before and after a command:

[tool.run-script]
pre-test = "echo 'Setting up...'"
test = "uv run pytest"
post-test = "echo 'Cleaning up...'"

Both naming styles are supported:

Style Pre-hook Post-hook
Composer pre-test post-test
NPM pretest posttest

Skip hooks when needed:

uvx uvtask --no-hooks test

Arguments

Extra arguments after the command name are forwarded to the underlying script:

uvx uvtask test -k integration -x
uvx uvtask celery-call example -k '{"kwarg": "value"}'

Values with spaces and JSON are quoted correctly for the shell on both Unix and Windows.

Namespaced commands

Use colons to group related commands:

[tool.run-script]
static-analysis = { command = ["static-analysis:linter", "static-analysis:types"], description = "Run all static analysis checks" }
"static-analysis:linter" = "uv run ruff check ."
"static-analysis:types" = "uv run ty check ."

CLI reference

Flag Purpose
-q / --quiet Suppress stdout (stackable)
-v / --verbose Show command and exit codes (stackable)
--no-hooks / --ignore-scripts Skip pre/post hooks
--color Control color output: auto, always, or never
help [command] Show per-command documentation from description
-V / --version Print version
-h / --help Print general help

Security model

uvtask runs the shell strings written in pyproject.toml, so a project's manifest is trusted exactly like a Makefile or an npm script: opening a repository is safe, running one of its commands is not. Treat uvx uvtask <command> in an unfamiliar repository the same way you would treat npm run.

Everything else is treated as untrusted:

  • Forwarded arguments are quoted before reaching the shell — shlex.join on Unix, and list2cmdline plus caret-escaping of cmd.exe metacharacters on Windows — so a value like foo&whoami is passed through as data rather than run as a second command.
  • Script names and descriptions are stripped of terminal control characters before display, so a manifest cannot rewrite what you see in --help or uvtask help <command>.
  • Malformed scripts are rejected rather than coerced. A table without a command key, a non-string command, a circular reference, or a reference graph that expands past 512 commands fails with a config error instead of being handed to the shell.

Hooks are skipped only by --no-hooks / --ignore-scripts placed before the command name. The same flag after the command name is forwarded to the underlying script, so arguments meant for a child process cannot silently bypass a guard hook.

Comparison

Tool Best for uvtask difference
uv run / uvx One-off tool invocations Named, documented project commands with hooks and composition
Poe the Poet Rich task runner with templating Zero deps, uvx-native, uv-styled CLI
Hatch scripts Hatch-managed projects Tool-agnostic pyproject.toml config, works with any uv project
npm / Composer scripts JS / PHP ecosystems Same mental model, Python-native

Development

Run the development version from a local checkout:

uvx -q --no-cache --from $PWD uvtask

Common project tasks:

uvx uvtask static-analysis
uvx uvtask test

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT © uvtask contributors


Note: uvtask is an independent, third-party project — not an official Astral tool. It is inspired by and designed to work seamlessly with Astral's excellent tools (uv, ruff, ty). We're grateful for the work the Astral team does for the Python ecosystem.

About

An extremely fast Python task runner.

Topics

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages