Publishes the org's shared FHIR Implementation Guide packages — @max-health-inc/fhir-* — to GitHub Packages, generated with babelfhir-ts.
Always publishes with the latest babelfhir-ts that passed parity. There is no pin. Each run resolves the version once, in the workflow's setup job, and hands that exact version to both the matrix builder and the generator — so a run can never key a package to one generator and build it with another.
"Latest that passed parity", not simply latest on npm. One generator release becomes ~30 published packages here, while the gate that decides whether a release actually works — 30 IGs against two external validators — runs in the generator's repo. scripts/resolve-babelfhir.js therefore picks the newest release with a recorded stable parity result:
https://max-health-inc.github.io/BabelFHIR-TS/history-stable.json
A hard parity failure leaves no parity data to publish, so no entry is written — absence is the "did not pass" signal. It fails closed: if nothing qualifies, the run stops rather than falling back to the newest npm release, because a silent fallback would defeat the point. config.json → minParityValidation adds an optional score floor for soft regressions; it is off by default (see the note in that file).
Consequences, by design:
- A new passing release changes every package's build key, so the next run republishes every IG once with a patch bump. That is the point — packages track generator improvements automatically. Nothing republishes between releases, because the key is unchanged.
- A commit of this repo no longer determines its output: the generator version is resolved at run time rather than recorded here.
- A release that never gets a parity result never gets published with, and this repo will refuse to run until one does.
The IG list is derived from babelfhir-ts's validated parity matrix (src/test/parity/parityConstants.ts → AVAILABLE_PACKAGES), read from the parity-matrix.json artifact babelfhir-ts commits and ships, fetched at the git tag of the release this run resolved. scripts/list-igs.js does this at CI time. Reading the matrix at that same tag is what keeps the published set equal to what that exact generator validated. Each entry → @max-health-inc/fhir-<name> at the IG's upstream version, generated for the FHIR version babelfhir validated it against.
parity-matrix.json is a contract, not scraped source. This script used to regex parityConstants.ts directly, which made a formatting change in another repository a silent break in publishing here. babelfhir-ts guards the artifact with a CI drift check and a unit test, and declares a schemaVersion — this publisher refuses to run on a version it does not understand rather than guess an IG set. The artifact ships from babelfhir-ts 1.5.18 onward, so tracking latest always finds it.
This deliberately has no hand-maintained IG list — a duplicated list drifts (e.g. an IG marked R5 in parity but defaulted to R4 here). babelfhir decides what's validated and what's published, and matrix changes are picked up as soon as they are released.
config.json holds only this repo's publishing config:
displayLanguages— the--display-languageflag valueexclude— IG names in the parity matrix to not publish (normally empty)
Generation flags (shared): --tx-server https://tx.fhir.org/<fhirVersion> and --display-language <displayLanguages>.
The Publish IG Packages workflow runs on:
- push to
maintouchingconfig.json/package.json/scripts/**/ the workflow — publishes changed IGs - weekly
schedule— picks up upstream terminology (tx.fhir.org) refreshes workflow_dispatch—packagesinput: a comma-separated subset orall
Publishing is input-driven via a per-package build key "<spec>|babelfhir@<resolved>|<flags>" — the concrete resolved version, never the string latest, since a key containing latest would never change and no IG would ever pick up a generator improvement. It is recorded as a bk-<sha1> dist-tag on the published version (GitHub Packages does not preserve custom package.json fields, but dist-tags survive). A package whose build-key dist-tag already exists is skipped (no regenerate). Version tracks the upstream IG version, with an auto patch-bump when the key changes.
npm i @max-health-inc/fhir-us-coreThe packages are public, so any GitHub token with read:packages resolves them — in GitHub Actions, the built-in GITHUB_TOKEN (with permissions: packages: read) is enough; no PAT needed. Add an .npmrc:
@max-health-inc:registry=https://npm.pkg.github.com
//npm.pkg.github.com/:_authToken=${GITHUB_TOKEN}
Add it to babelfhir-ts's parity matrix (parityConstants.ts → AVAILABLE_PACKAGES) so it's validated, release babelfhir-ts, then bump the babelfhir-ts pin in package.json here — the new IG is picked up automatically. To bump the generator for all IGs, bump the pin. To publish everything with the current pin (e.g. after a parity matrix change at the same version), run the workflow via workflow_dispatch.