CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What this is

A Jekyll static site (Stefano Ribes’ academic homepage, ribesstefano.github.io) built on the Academic Pages template (a fork of the Minimal Mistakes theme). There is no application code — content is authored as Markdown/YAML front matter, and the theme is Liquid templates + SCSS. The site is served straight from GitHub Pages (repo name matches <user>.github.io, no CNAME, no custom GitHub Actions deploy workflow) — pushing to the default branch is what publishes it.

Commands

bundle install                        # install Ruby deps (Jekyll pinned via the github-pages gem)
bundle exec jekyll serve -l -H localhost   # local dev server at localhost:4000, live-reloads on change
bundle exec jekyll build              # one-off build into _site/

• Always run Jekyll through bundle exec so the github-pages gem’s pinned Jekyll version is used (matches what GitHub Pages actually builds with).
• If bundle install fails after a Gemfile change, delete Gemfile.lock and retry (per README).
• There is no test suite, linter, or JS build step to run for content changes. The package.json uglify/build:js scripts only re-bundle the theme’s vendored jQuery plugins into assets/js/main.min.js — irrelevant unless you’re touching assets/js/_main.js or assets/js/plugins/.
• _config.yml is not hot-reloaded by jekyll serve; restart the server after editing it. _config.dev.yml holds local-only overrides (disables analytics, expands compressed CSS) — merge it in with --config _config.yml,_config.dev.yml if you need those overrides.
• _site/ is the build output (gitignored) — never edit it directly.

Content architecture

Content lives in Jekyll collections, each a directory of Markdown files with YAML front matter, configured in _config.yml under collections:. Defaults for layout/author_profile/ etc. per collection are also set there under defaults: — don’t repeat them in individual files unless overriding.

Standalone pages (About, CV, Publications/Talks/Portfolio index, Supervision, etc.) live in _pages/. The homepage is _pages/about.md (permalink: /).

Permalinks are set explicitly in each file’s front matter, not derived from the filename — check a sibling file before adding a new one:

• Publications use singular /publication/<slug> (e.g. /publication/2026-02-20-PROTAC-Splitter), even though the collection is publications. One legacy 2020 entry still uses the old plural /publications/... form — don’t copy that one.
• Talks use the plural /talks/<slug>, matching the collection name.

_posts/default/ contains the theme’s original demo posts (2012–2015 + a year-2199 future post). It’s excluded from the build via _posts/default/* in _config.yml’s exclude: — it’s template reference material, not live content.

Shared front matter: resource-link pills

_includes/resource-links.html renders a row of pill links (Paper, DOI, arXiv, ChemRxiv, Preprint, PDF, Code, Demo, Colab, Dataset, Zenodo, Website, Slides, Poster, Video) from whichever of these front-matter keys are present on a publication/talk/portfolio entry: paperurl, doi, arxiv, chemrxiv, preprint, pdf, github, colab, spaces, dataset, zenodo, website, slides, poster, video. It’s included from both _includes/archive-single.html (list/archive views) and _layouts/single.html (single publication pages) — add the relevant key to an entry’s front matter rather than hand-writing link markup in the body.

The DOI pill only renders when doi looks like a real DOI (10. prefix) — some older entries store a Zenodo id in that field instead, which is intentionally excluded to avoid a dead link.

News feed

_data/news.yml is a simple reverse-chronological list (date is free text, body is inline HTML) rendered on the homepage via `

• June 2026
  Presented a TACK poster at the Machine Learning for Materials and Molecular Discoveries (ML2MD) symposium in Uppsala. Tingting Mo and Jia Xin Zhu also defended their MSc thesis on PROTAC synthesizability.
• May 2026
  TACK was accepted at KDD 2026 (AI for Science Track) — a 3,514-PROTAC dataset and a statistical benchmark of degradation activity models. Data on Hugging Face.
• March 2026
  Gave a tutorial on significance testing for molecular property prediction at the AIME lab workshop.
• February 2026
  PROTAC-Splitter is out in the Journal of Cheminformatics, together with a synthetic dataset of ~1.3M annotated PROTACs and a live demo.

`. Add new entries to the top.

Theming / styling

assets/css/main.scss is the single entry point that @imports the theme’s SCSS partials from _sass/ in cascade order, ending with _sass/_custom.scss imported last so it overrides theme defaults. Site-specific design (card grid, .resource-links pills, .funding-note callout, .news-list, .intro-hero, dark-mode .theme-toggle, etc.) belongs in _custom.scss, not in the upstream theme partials (_page.scss, _archive.scss, _sidebar.scss, …) — keep the theme partials close to stock so template updates stay mergeable.

Other tooling in the repo

• markdown_generator/ — Jupyter notebooks (publications.ipynb, talks.ipynb, PubsFromBib.ipynb) and equivalent plain-.py scripts that bulk-convert a TSV (publications.tsv, talks.tsv) or a .bib file into individual front-matter-formatted Markdown files for _publications//_talks/. Useful for importing many entries at once instead of hand-writing each file.
• talkmap/ + talkmap.py/talkmap.ipynb — scrapes the location field out of every file in _talks/, geocodes it (geopy/Nominatim), and regenerates the Leaflet cluster map assets in talkmap/ via the getorg library. Only relevant if talk locations changed and the map (linked from the Talks page when talkmap_link: true in _config.yml) needs regenerating; requires the getorg/geopy Python packages, which aren’t part of the Jekyll toolchain.