hxi.ucsd.edu

Source for the website of the Human-centered eXtended Intelligence (HXI) Research Lab at UC San Diego, directed by Nadir Weibel.

The site is Hugo with the Wowchemy Research Group theme, hosted on Netlify. Everything a visitor sees comes from Markdown files under content/. There is no database and no admin interface, so editing the site means editing text files in this repository.

The site, page by page

Home What the lab is, a photo slider, the calls to action, partners and funders. A stack of widget files in content/home/, ordered by weight: welcome.md, slider.md, intro.md, cta.md, partners.md, support.md.
Research Every project, split into current and earlier work, with Domain and Technology filters. Two portfolio widgets over content/project/: content/research/research.md and foundations.md. A project moves between the sections by gaining or losing the Earlier tag.
People Current members, grouped by role. content/people/people.md over the profile folders in content/authors/. The grouping comes from each person's own user_groups.
Alumni Former members, same layout and different groups. content/alumni/alumni.md. Their profile folders sit in content/authors/_alumni/.
Publications The full list, newest first, with search and filters. One folder per paper in content/publication/. layouts/section/publication.html sorts preprints after peer-reviewed papers within a year.
Teaching The courses the lab teaches. One folder per course in content/course/.
FAQ Joining the lab, getting in touch, letters and committee requests. One folder per question in content/faq/.

And the three page types that make up most of the site:

A project Description, images, funders, and a publication list that is generated rather than written: it is every paper whose projects: field names this folder. content/project/<slug>/index.md plus its images.
A publication Authors, abstract, venue, and the Cite and DOI buttons. Author names matching a profile become links and render in bold. content/publication/<slug>/index.md plus cite.bib.
A person Bio, links, and every paper naming them, collected automatically. content/authors/<name>/_index.md plus avatar.jpg.

How the site is put together

A page is a folder, not a file. Publications, people and projects each get their own folder holding the text and its images:

content/publication/2025-chidambaram-uist-drivesimquest/
    index.md          metadata and abstract
    cite.bib          the BibTeX entry
    featured.jpg      optional, found by filename

Hugo picks up featured.* and avatar.* by name, so you never write an image path. The folder name becomes the URL, which is why renaming one breaks every link to it.

Section pages are assembled, not written. The home page, /research and /people do not exist as documents. Each stacks widget files in order of a weight value, as listed in the table above.

Three fields do the cross-linking, and each fails silently when wrong:

• authors: on a publication. A name that slugifies to a folder in content/authors/ links to that person, puts the paper on their page, and bolds them as a lab member.
• projects: on a publication. Names a project folder, which then lists the paper automatically. The project page itself needs no edit.
• categories: mirrors projects: and exists only to drive the "Related" suggestions, because Hugo cannot index the projects field directly. It is hidden from view.

Everything else: config/_default/ holds the Hugo settings, theme options and navigation; assets/ the custom SCSS and the research-page filter JavaScript; static/images/ the funder logos; netlify.toml the build, redirects and deploy gate. The theme itself is a Hugo module pinned in go.mod and is not in this repository.

I just want to add a paper or update my profile

You do not need to know Hugo or Markdown. Ask Claude or ChatGPT to do it, using the lab's packaged helper. Ask Nadir for hxi-website-claude.zip (Claude) or hxi-website-chatgpt.zip (ChatGPT, which also works for Gemini and Copilot).

You will need write access to this repository. Send Nadir your GitHub username.

The same Claude package installs three ways, and it is one skill, not three:

Where you work	Install	Reaches the repo via
claude.ai or the Claude desktop app	Settings, Capabilities, Skills, Upload skill	a GitHub connector you authorize
Claude Code	unzip hxi-website-claude.zip -d ~/.claude/skills/	the git already on your machine
ChatGPT, Gemini, Copilot	attach hxi-website-reference.md from the ChatGPT zip	nothing; it writes the files and you paste them into github.com

Claude Code is the one worth having if you already use it. It is the only route that can run the site and show you the page before it publishes, because it has a shell. The browser versions cannot, and should say so rather than implying a change was checked visually.

If you would rather do it by hand, copy the shape of an existing neighbour rather than writing a file from scratch. The paths are in the tables above.

Committing does not publish

The live site is published in batches, not on every commit. Netlify's free tier gives 300 build credits a month and a full rebuild costs about 15, so roughly 20 publishes a month for the whole lab.

scripts/netlify-ignore.sh enforces this. It reads the DEPLOY_POLICY environment variable set in the Netlify UI:

DEPLOY_POLICY	Behaviour
unset or manual	Nothing publishes unless the commit message contains [deploy]
auto	Everything publishes except commits containing [skip build]

Commit as often as you like; it costs nothing. Your change goes live at the next publish. Do not put [deploy] in a commit message unless you are the one releasing the queue, because it publishes everything anyone else has queued too.

There is a second way to publish that costs no credits at all: building the site locally and uploading the result, so Netlify never runs a build. That is scripts/deploy-manual.sh, it needs a Netlify token, and the trade-offs and traps are in scripts/README-deploy.md. Worth knowing it exists; only whoever holds the token can run it.

Seeing your change before it publishes

Because publishing is batched, a commit sits in the repo for a while before it reaches hxi.ucsd.edu. There is no preview on the hosting side either: Netlify's pull-request previews run through the same gate as production, so opening a PR does not build you one. That leaves two routes.

Have the assistant read the file back. Quick, needs nothing installed, and it is what the packaged skill does after every commit: it re-reads what it wrote and checks it against the conventions that fail silently, such as the group string on a profile or the project slug on a paper. This catches most mistakes, but it is not a picture of the page. A chat assistant in a browser has no shell, so it genuinely cannot render the site; if one offers to, it is wrong.

Run the site yourself. The only way to actually look at the page:

git clone https://github.com/WeibelLab/hxi-website.git
cd hxi-website
hugo server

Then open http://localhost:1313. In Claude Code, which does have a shell, you can ask Claude to do this for you and it will start the preview and check the page.

Hugo must be the extended build, version 0.89.2, the version Netlify uses. Newer versions break this theme's pinned modules, so brew install hugo will not work. Get it from the v0.89.2 release page.

Two things that otherwise waste an afternoon:

• Restart the server after changing anything in config/, static/, assets/js/ or layouts/, and after renaming a folder. Live reload misses those and quietly serves stale content, which looks exactly like your change not working.
• A clean build is not proof. A wrong user_groups value or a nonexistent project slug builds without complaint and simply omits your content. Open the page and look for it.

Conventions that matter

These are the ones that fail silently, producing a clean build with your content missing:

• user_groups must match one of the strings the people widget lists. Ph.D has one period and Master is singular. A value not on that list removes the person from /people with no error. Copy it from an existing profile rather than retyping it.
• projects: values are project folder names. A slug that does not exist just does nothing.
• Spell out every author, in paper order, never "et al.". A misspelled name silently costs that person the credit.
• Nothing under review goes on the site unless it is already on arXiv.
• Future dates are fine. The site builds with buildFuture, so an accepted paper can carry its real conference date.

Old URLs

Renaming a project folder changes its URL and breaks every link anyone already has. Keep the old URL working by listing it as an alias on the page that should now receive the traffic. Rename willo to student-wellbeing, and in content/project/student-wellbeing/index.md:

---
aliases:
  - /project/willo/
title: 'Student Mental Health and Well-Being: Research with WILLO'

Four things to get right:

• Leading and trailing slashes, and the section prefix. /project/willo/, not willo.
• Lowercase. Hugo lowercases every URL, so an alias with capitals never matches.
• The alias goes on the destination page. The old folder no longer exists.
• Never alias a path a real page occupies. The redirect is forced, so it would shadow that page and make it unreachable.

Aliases work the same way on author profiles. /author/hridy/ sits in content/authors/hridyanshu/_index.md.

Check it with hugo server and visit the old URL. Hugo writes a small stub page at every alias path, which is what makes this testable locally. At build time layouts/index.redirects turns the same front matter into forced 301s in public/_redirects, and that is what production serves. The stub and the 301 land on the same page.

netlify.toml holds the redirects that have no page to live in: wildcards, off-site destinations such as /join, and rewrites that keep the URL bar unchanged such as /weibel.

Bibliography

The full publication list is exported as BibTeX at hxi.ucsd.edu/hxi.bib, generated at build time from each publication's own cite.bib. There is no hand-maintained master .bib file, deliberately: the previous one silently went stale for four years.

Layout overrides

Everything in layouts/ overrides the theme. Keep this list short.

File	Why
layouts/section/publication.html	Sorts preprints last within each year
layouts/authors/list.html	Groups people by user_groups on /people and /alumni
layouts/partials/widgets/portfolio.html	Adds exclude_tags and two-axis filtering on /research
layouts/index.bib	Generates the combined BibTeX export
layouts/_default/_markup/render-link.html	Sends external links to one shared tab instead of a new tab each
layouts/index.redirects	Compiles page aliases: into forced 301s in _redirects

External links open in a single shared tab named hxi-external, rather than a new tab per click. The render hook above covers links written in content; assets/js/external-links.js covers the ones theme templates generate, such as DOI and publication buttons.

Screenshots in docs/screenshots/ are used only by this README.

Code of conduct

See CODE_OF_CONDUCT.md. Report concerns to hxi@ucsd.edu.