For agents: start with the agent guide. Every page is Markdown by default; add Accept: text/html for HTML.

CAFE all stars

What it does

CAFE all stars lists people and channels whose public work builds with Cloudflare, Alchemy, Foldkit²⁷ and Effect³⁵. A channel is a YouTube channel, a GitHub org or a blog.¶

Each entry lives in .brain/data/cafe/people.json, beside the CAFE news file. The content build decodes it with the CafePerson schema³⁰ in packages/core/src/cafe.ts. A bad entry fails the build with its index and field, the same way a bad news row does.¶

Five read-only capabilities answer every question about the data:¶

• listPeople lists every entry, and listChannels lists only the channels.¶
• readPerson reads one profile by its stable id: a GitHub login, or the handle of an org or channel.¶
• searchPeople matches words against names, handles, topics, packages, repos, bios and videos.¶
• peopleFor returns the entries whose topics or repo packages cover every term, with the evidence for each term. Entries that list a term as a topic rank ahead of entries that only list its package.¶

The /all-stars index and each /all-stars/<id> page are built from listPeople and readPerson answers. They are Markdown by default and HTML for a browser. llms.txt lists the index and every profile, and /all-stars.xml is an Atom feed of the entries and their activity.¶

A news row joins a profile when its GitHub login or X handle matches one of the profile’s handles. A row with no author joins the profile that owns its repo. The profile lists those rows newest first, and the row on /news links back to the profile.¶

The standard

• A person or channel enters through a /news row, pinned CAFE stack contributions, vendored or credited CAFE code, or sourced public CAFE work.¶
• The typed entry field records the basis. A news entry names at least one row.¶
• A stack entry names a pinned package and its contributors or maintainers source. A vendored entry links the credited upstream, which must itself build on or teach the CAFE stack.¶
• A work entry links the project and dated stack evidence.¶
• Choose the first qualifying basis in that order. Stack contributors lead or rank among the repository’s top listed contributors. Public work builds on or teaches Cloudflare, Alchemy, Foldkit or Effect.¶
• There is no per-person preview. An entry ships with the next site update like any other content.¶
• The content build warns about every author with a /news row and no profile.¶
• Every fact carries its source URL and the verified date it was checked, so an agent can cite it.¶
• A bio holds two or three sentences about CAFE work only, each quoted or linked from its source. Use names or projects, not gendered pronouns; the content build rejects those pronouns.¶
• An avatar is a file committed under apps/site/content/avatars. Pages never load an image from another site.¶
• A profile never includes location, employer guesses, family, follower counts, private messages, or anything the person did not post publicly about their work.¶
• Each page ends with a removal line that opens an issue in joelhooks/rat-stack. A removal ships with the next site update.¶

To remove a profile, delete its entry from people.json and its avatar file, then rebuild. Nothing else stores the profile.¶

How to check

Run the core tests:¶

pnpm --filter @rat-stack/core exec vitest run test/cafe-people.test.ts

They check that a generated news row lands on exactly the profile that authored it, newest first. They check that topic holders rank first in peopleFor, however many repos list a package. They also answer three agent questions from the shipped data through the capabilities alone.¶

Then call a capability²⁸ on a running site:¶

curl --request POST 'https://ratstack.sh/api/peopleFor' \
  --header 'content-type: application/json' \
  --data '{"terms":["foldkit"]}'

Last changed 2026-11-04 in 83069f3. Source on GitHub. Change log.