| name | institute-website |
| description | Work on the Active Inference Institute public website (institute_website): a static, zero-dependency Node build that renders JSON content into clean-URL HTML for GitHub Pages, with a strict CSP + link + public-safety contract. USE WHEN adding or editing a page, adding a project, registering an external link, changing navigation, adding a language/locale, editing the design system, changing legacy-URL redirects, doing the domain cutover, or running the build and check gates. Covers the page schema, the slug→URL taxonomy, the live-sources external-anchor gate, and the verification workflow. |
Institute Website skill
The single entry point for agents working in this repository (institute_website).
Read this,
then the relevant Workflows/*.md for the task and the AGENTS.md in the folder
you are editing.
What this repo is
A static website for the Active Inference Institute, served by GitHub Pages
from the repository root of main. The build is plain Node (no framework, no
runtime deps): node src/build.mjs reads JSON content under src/content/ and
writes HTML into the repo root (about/index.html, projects/<slug>/index.html,
…) plus one flat index.html and 404.html. Built output is committed because
Pages serves it directly.
Canonical domain after cutover: https://activeinference.institute/ (set in
src/content/site.json baseUrl; a CNAME file claims it). Before cutover the
base was https://activeinferenceinstitute.github.io/institute_website/.
Golden rules (violating these fails npm run check)
- Edit source, never built output. Change
src/content/*.json,
src/render/*.mjs, assets/, then rebuild. Never hand-edit */index.html.
- External links must be registered. Every external
<a href> on a content
page must exist in src/content/live-sources.json and be referenced by
sourceId — NOT a raw href. Even vetted hosts (youtube.com,
*.activeinference.institute) fail as raw content hrefs. See
Workflows/RegisterLiveSource.md.
- Each content page needs:
primaryActions (non-empty), externalSourceIds,
and at least one primaryAction backed by a registered sourceId (a "verified
external action").
- Strict CSP. No inline
<script>/<style>, no iframe/object/embed/
form, no client-side fetch (connect-src 'none'). Page-specific JS is an
external assets/js/*.js file conditionally referenced from layout.mjs.
- Public-safety. No private emails, rosters,
coda.io data payloads, or
internal material. Structured InstituteOS data enters ONLY via
scripts/sync_instituteos_public_data.py.
- Run the gates before declaring done:
npm run check (see
Workflows/RunGates.md).
Slug → URL taxonomy (src/url-taxonomy.mjs)
Pages auto-register: drop a <name>.json anywhere under src/content/pages/
(folder nesting is for humans; it does NOT affect the URL). The slug field is
the identity:
slug: "x" → /x/ (top-level, e.g. structure → /structure/)slug: "project-x" → /projects/x/- program slugs in the taxonomy's programs set →
/programs/x/
Workflows
Folder map (each has its own AGENTS.md)
| Folder | What lives there |
|---|
src/ | Build entry (build.mjs), data load (data.mjs), URL taxonomy |
src/render/ | HTML rendering modules (layout, pages, links, seo, …) |
src/lib/ | Pure helpers (output, paths, text, instituteos) |
src/content/ | JSON registries (site, navigation, resources, live-sources, …) |
src/content/pages/ | Page content + the page schema and type templates |
src/pages/ | Registry-driven page renderers (calendar, directory, projects, search, sitemap, …) |
scripts/ | Check gates, InstituteOS sync, i18n translate |
assets/ | CSS design system, JS, images, redirects.js |
data/ | Build provenance (export-manifest.json) + projects feed |
simulations/ | Vendored interactive demos (excluded from first-party gates) |
Cross-references
