TLDR

These are my defaults for a new project.

• Toolchain: Use uv for dependencies, interpreters, tools, scripts, and builds.
• Formatting & Linting: Use ruff for both linting and formatting.
• Data Modeling: Use Pydantic at I/O boundaries; native dataclasses or attrs internally.
• Configuration: Use pydantic-settings for configuration, validated strictly at startup.
• Project Metadata: Everything configured in pyproject.toml; dev dependencies managed via standard [dependency-groups].
• Project Structure: Use src/ layout, commit your lockfile, and enforce uv lock --check in CI.
• Type Checking: Run a type checker in CI, strict on new code (pyright, or pyrefly on large codebases).
• Logging: Structured logging configured once at the application entry point—never inside reusable libraries.
• Concurrency: Use TaskGroup and asyncio.timeout for async I/O; processes for CPU-bound tasks.
• Testing: Use pytest with fixtures and parametrization; Hypothesis for property-based invariants; Testcontainers instead of database mocks.
• CLIs & Tooling: Use Typer or Cyclopts for production CLIs; # /// script inline metadata for single-file tools.
• Performance: Profile before optimizing using py-spy, scalene, or memray.
• Security & CI/CD: Pin GitHub Actions to commit SHAs; publish packages to PyPI via Trusted Publishers (OIDC).