freeze-deps

Skill: Freeze Deps

Pin the dependencies of the notebook at $ARGUMENTS so the analysis can be reproduced. Adapts to the project's package manager: emits requirements.txt for pip projects and a poetry add command for Poetry projects.

Step 1 — Detect the project's package manager

Look at the current working directory for these signals:

• pyproject.toml exists AND contains [tool.poetry] → Poetry project
• pyproject.toml exists with [project] table (PEP 621) but no [tool.poetry] → PEP 621 project (treat as pip-compatible — emit requirements.txt)
• No pyproject.toml → pip project → emit requirements.txt

This decision drives Step 4. Capture the choice for the report.

Step 2 — Extract imports from the notebook

Run a Python one-shot via Bash that:

1. Loads the .ipynb as JSON.
2. Iterates over cell_type == "code" cells.
3. Joins the source lines per cell and parses each with ast.parse (catching SyntaxError and skipping the cell with a warning — IPython magics like %load_ext are not valid Python and will fail to parse).
4. Walks the AST collecting ast.Import and ast.ImportFrom nodes.
5. Extracts the top-level package name for each import:
   • import sklearn.cluster → sklearn
   • from scipy.stats import ttest_ind → scipy
   • import pandas as pd → pandas
6. Filters out modules from the standard library using sys.stdlib_module_names (Python 3.10+).
7. Maps import names to distribution names where they differ (use importlib.metadata.packages_distributions()):
   • sklearn → scikit-learn
   • cv2 → opencv-python
   • PIL → Pillow

Returns a sorted list of unique distribution names.

Step 3 — Resolve installed versions

For each distribution name, call importlib.metadata.version(name). If a package is imported by the notebook but not installed in the current environment, include it in the output with a clear marker:

# pandas==2.1.0
# WARNING: <pkg> imported but not installed in current environment