Uv

uv

Use uv for Python in projects that have adopted it. uv replaces pip, python, python -m venv, and python -m pip with a single fast tool.

When NOT to use this skill

If the repo uses Poetry (poetry.lock, [tool.poetry] in pyproject.toml), pip-tools (requirements.in + requirements.txt), or Pipenv (Pipfile), follow that toolchain instead. Don't introduce uv into a project that hasn't adopted it.

Signals that a repo uses uv:

• uv.lock present
• pyproject.toml has [tool.uv] or requires = ["uv_build"] in [build-system]
• README references uv run / uv add

Quick reference

uv run script.py                       # Run a script
uv run --with requests script.py       # Run with ad-hoc dependency
uv run python -m ast foo.py >/dev/null # Verify syntax without writing __pycache__
uv add requests                        # Add dependency to project
uv init --script foo.py                # Create script with inline metadata
uv lock --script foo.py                # Lock script's deps to foo.py.lock

Inline script metadata

Standalone scripts can declare their own dependencies (PEP 723):

# /// script
# requires-python = ">=3.12"
# dependencies = [
#   "requests<3",
#   "rich",
# ]
# ///

import requests
from rich import print

Then just: uv run script.py. No virtualenv to manage; uv resolves and caches the deps automatically.

For an executable shebang script:

#!/usr/bin/env -S uv run --script
# /// script
# dependencies = ["httpx"]
# ///
import httpx
print(httpx.get("https://example.com").text)

chmod +x myscript
./myscript

See scripts.md for full script-running details (locking, reproducibility, alternative indices).

Build backend

For pure Python packages, use uv_build:

[project]
name = "my-package"
version = "0.1.0"
requires-python = ">=3.12"

[build-system]
requires = ["uv_build>=0.9.28,<0.10.0"]
build-backend = "uv_build"

For C-extension modules, use hatchling instead — uv_build is pure-Python only.

See build.md for project structure, namespace packages, and file inclusion rules.

Common tasks

Syntax check without writing __pycache__

python -m py_compile writes pycache files into the source tree. To check syntax without that side effect:

uv run python -m ast script.py >/dev/null

Exit code is non-zero on syntax error; stderr has the parse error.

Reproducibility

Pin the dependency resolution date in inline metadata so a script run today resolves the same way next year:

# /// script
# dependencies = ["requests"]
# [tool.uv]
# exclude-newer = "2026-05-01T00:00:00Z"
# ///

Adapted from

• mitsuhiko/agent-stuff