CI/CD

Workflows

This repository ships with three GitHub Actions workflows:

• Repo Checks: runs on pull requests and pushes to main or codex/**
• Docs Pages: deploys the VitePress site to GitHub Pages on pushes to main
• Release Package: builds and publishes package artifacts when a v* tag is pushed

Repo Checks

Repo Checks is the main safety net for everyday development. It currently verifies:

• required documentation assets exist
• the docs site builds successfully
• maintained Python surfaces compile
• maintained runtime and package files pass ruff
• the Python package builds as wheel and sdist
• the editable package exposes the expected entrypoints
• matanyone2-runtime --help and python -m matanyone2 --help work from the packaged install

This job intentionally focuses on maintained runtime surfaces instead of the entire upstream model tree, so we can keep CI signal tight while still inheriting upstream code.

Docs Pages

Docs Pages builds the site from docs/ and deploys docs/.vitepress/dist to GitHub Pages.

This means documentation quality is checked in two stages:

• Repo Checks confirms the docs build is healthy on every pull request
• Docs Pages publishes the same output after main is updated

Release Package

When you push a tag like v0.1.0, the release workflow:

1. builds the source distribution and wheel
2. generates SHA256SUMS.txt
3. uploads the artifacts to the workflow run
4. creates or updates the GitHub Release for that tag
5. attaches the generated files to the release
6. publishes the same artifacts to PyPI through GitHub trusted publishing

Example:

git tag v0.1.0
git push origin v0.1.0

PyPI setup

PyPI publishing needs one-time configuration outside the repository:

1. Create the matanyone2-runtime project on PyPI, or claim the name you intend to publish.
2. In PyPI, add a trusted publisher for this GitHub repository and workflow.
3. Match it to the pypi environment used by .github/workflows/release-package.yml.
4. Keep tagged releases on main so the workflow and PyPI publisher configuration stay aligned.

The current workflow uses the official pypa/gh-action-pypi-publish action with OIDC, so no long-lived API token needs to be stored in GitHub secrets once trusted publishing is configured.

Published package:

• matanyone2-runtime on PyPI

Local verification before pushing

These are the same checks we use locally before relying on CI:

.\.venv\Scripts\python.exe -m compileall hugging_face matanyone2 scripts
.\.venv\Scripts\python.exe -m ruff check pyproject.toml hugging_face\app.py hugging_face\__init__.py scripts\run_pipeline_check.py
.\.venv\Scripts\python.exe -m ruff check matanyone2\__init__.py matanyone2\__main__.py matanyone2\cli.py matanyone2\demo_core.py matanyone2\gradio.py matanyone2\runtime.py
.\.venv\Scripts\python.exe -m build

If you already installed the project in editable mode, you can also smoke test the packaged entrypoints:

.\.venv\Scripts\matanyone2-runtime.exe --help
.\.venv\Scripts\python.exe -m matanyone2 --help