Summary

This PR converts nmaci from a collection of standalone scripts (downloaded as a tarball by course repo CI) into a proper pip-installable Python package with a unified CLI entry point.

What changed

Package structure (src/ layout with hatchling build backend)

• Added pyproject.toml — declares dependencies, entry point, and build config
• Source moved to src/nmaci/ — one module per command, importable as a package
• src/nmaci/cli.py — single nmaci entry point dispatching to subcommands via lazy importlib imports
• src/nmaci/__main__.py — enables python -m nmaci <command>

Scripts refactored

• generate_book.py, generate_book_dl.py, parse_html_for_errors.py — removed module-level sys.argv access that crashed on import
• select_notebooks.py, find_unreferenced_content.py — wrapped bare __main__ blocks into main(arglist=None)
• All modules now follow the main(arglist=None) convention (argparse reads sys.argv[1:] when arglist is None)

Backward-compatible shims (scripts/)

• All scripts in scripts/ are now 3-line shims that delegate to src/nmaci/
• Course repo CI (python ci/process_notebooks.py) continues to work unchanged as long as nmaci is installed

Tests & CI

• Tests import from nmaci.* directly; subprocess tests use python -m nmaci <command>
• CI switched from pip install -r requirements.txt to uv sync --extra dev
• from __future__ import annotations added to fix Path | str union syntax on Python 3.9

Backward compatibility

Course repos that currently do python ci/process_notebooks.py <args> are unaffected — the shims preserve that interface. The only new requirement is that nmaci be installed as a package, which the updated setup-ci-tools action handles.

Testing

• All 16 unit + integration tests pass locally (uv run pytest tests/ src/nmaci/)
• End-to-end: PR opened against a NeuroAI Course fork with setup-ci-tools patched to pip install git+https://github.com/OleBialas/nmaci@main — verifies shims work correctly in a real CI environment. See test: verify nmaci pip package shims work end-to-end OleBialas/NeuroAI_Course#2