reusable workflows

This repository contains reusable GitHub Actions workflows for C3S projects. There is currently one CI workflow for Jupyter Notebook QA automation.

notebook-qa

This workflow implements QA automation for Jupyter Notebooks. The checks below run automatically — all checks that are not skipped or disabled must pass for the workflow to succeed.

Checks

Code linting (linter) — Runs ruff check on all code cells. No Python lint violations allowed.

Code formatting (formatter) — Runs ruff format --check on code cells. Code cells must conform to ruff formatting rules.

Notebook linting (pynblint) — Runs pynblint on each notebook. Checks notebook-level quality issues such as non-linear execution order, empty cells, or untitled notebooks.

Link availability (links) — Runs lychee against all notebooks. Every URL in markdown and code cells must be reachable.

Notebook execution (execute) — Executes each notebook end-to-end with ploomber-engine. The notebook must run without errors. Memory usage and runtime are profiled per cell and uploaded as an artifact.

Version metadata (metadata) — Looks for **Last updated:** YYYY-MM-DD (e.g. **Last updated:** 2025-01-15) in the first markdown cell(s) before any code cell. Falls back to a README.md in the same directory if not found in the notebook.

Tests & coverage (tests) — If test files exist (test_*.py, *_test.py, tests/*.py), runs pytest with coverage. Coverage must meet the configured threshold (default 80%). When no test files exist the check passes by default, unless require_tests: true is set in the config.

Accessibility (accessibility) — Runs WCAG compliance checks on notebooks using jupyterlab-a11y-checker.

Figure attribution (figures) — Every figure output (PNG/JPEG) in code cells must have source attribution in a nearby markdown cell (within 2 cells). Recognized attribution patterns include source:, credit:, data from:, attribution:, reference:, dataset:, a DOI, or a URL.

License file (license) — A non-empty LICENSE file must exist in the repository root.

Changelog file (changelog) — A non-empty CHANGELOG.md file must exist in the repository root.

How to use notebook-qa.yml workflow

Configure the target repository which you want to run the QA check against using this format:

.github/workflows/qa.yml

------------------------

name: Notebook QA

on:
  push:
    branches:
      - develop
  pull_request:
    branches:
      - develop
  workflow_dispatch:
    inputs:
      notebooks:
        description: "Space-separated list of notebook paths to check (default: all *.ipynb)"
        required: false
        type: string
        default: ""

jobs:
  notebook-qa:
    uses: ecmwf-training/reusable-workflows/.github/workflows/notebook-qa.yml@main
    with:
      notebooks: ${{ inputs.notebooks || '' }}
    secrets: inherit

This sets up automated checks on new pull requests and merges/pushes into develop branch. It also allows manual Action runs in the GitHub Actions UI.

Configuration

You can customize check behavior by adding a .github/notebook-qa.yml config file in the consuming repository:

# Globally disable specific checks
disabled_checks:
  - linter
  - formatter

# Notebooks to skip entirely (all checks), supports glob patterns
skip_notebooks:
  - "./draft.ipynb" # To skip notebooks at the top level include "./" to ensure that the path is resolved
  - "notebooks/draft.ipynb"
  - "notebooks/experimental/**"

# Per-notebook check configuration
notebooks:
  "notebooks/example.ipynb":
    skip:
      - figures

# Test configuration
require_tests: false     # Set true to fail when no test files exist
coverage_threshold: 80   # Minimum coverage percentage for pytest-cov

# Pynblint rule configuration
pynblint:
  exclude:                    # Additional rules to suppress (extends baseline)
    - untitled-notebook
  exclude_mode: extend        # "extend" (default) or "override"

The baseline pynblint exclusion list suppresses missing-h1-MD-heading (MyST notebooks use YAML frontmatter for titles). In extend mode (default), your exclude list is merged with the baseline. In override mode, your list fully replaces the baseline.

Available pynblint rule slugs: non-linear-execution, notebook-too-long, untitled-notebook, non-portable-chars-in-nb-name, notebook-name-too-long, imports-beyond-first-cell, missing-h1-MD-heading, missing-opening-MD-text, missing-closing-MD-text, too-few-MD-cells, duplicate-notebook-not-renamed, invalid-python-syntax, non-executed-notebook, non-executed-cells, empty-cells, long-multiline-python-comment, cell-too-long

Valid check IDs: linter, formatter, pynblint, links, tests, figures, metadata, accessibility, license, changelog, execute

QA criteria reference

ID	Description	Tool / Check
1.2.3	Link availability	lychee
1.2.4	License file	LICENSE existence
1.2.6	Version metadata	metadata_checker.py
2.2.1	Code execution	ploomber-engine
2.2.3	Code style	ruff, pynblint
2.2.4	Execution profiling	ploomber-engine
2.2.6	Memory profiling	ploomber-engine
2.3.1	Test existence	pytest
2.3.2	Coverage threshold	pytest-cov
3.1.3	Accessibility	jupyterlab-a11y-checker
3.3.2	Figure attribution	figure_checker.py
4.2.3	Changelog	CHANGELOG.md existence

How to configure access to cdsapi for notebook execution check

The action responsible for notebook execution allows setting a cdsapi key via CDSAPI_KEY secret set either on repository or organisation level.

How to setup reusable-workflows repository in GitHub organisation

1. Fork this repository into your organisation
2. Leave the fork network in the newly forked repository settings