Each step is one pull request, merged when CI is green. Decisions that are the owner’s are asked before the step that needs them and recorded in memory/.

Decisions so far (2026-10-23)

Decision Choice Record
Library consumption Build the JVM jars in CI from a pinned smart-trust-tools commit memory/dir-library-consumption.md
Harness Full cat-harness downstream of smart-trust-tools memory/dir-downstream.md
Home country (J2–J4) A fictional GDHCN DEV participant; J1’s issuer stays XXR memory/dir-journeys-demo.md
OHS base for the app Option B, staged; Rust/wasm for the web meanwhile (proposals/app-platform.md) memory/dir-app-platform.md
IG identity smart.who.int.trust-demo, http://smart.who.int/trust-demo memory/dir-ig-identity.md

Steps

1. Plan and skeleton (this PR)

README (including the smart-trust-phw upstream note), AGENTS.md, this plan, the app-platform proposal, owner directives in memory/, the dependency register, and a CI job that checks every GitHub Action is pinned by commit SHA and every JSON file parses.

2. Journeys as cat-harness content

2c. L2 DAK and L3 FHIR IG

The journeys as a WHO SMART Guidelines DAK and FHIR IG, shaped as litlfred/smart-immunizations (sushi-config.yaml, ig.ini, the DAK in index.config.json (contentTypes.dak), input/). The L2 content is generated from the harness sources by tools/ig-gen/ig_gen.py: roles become ActorDefinitions, the BPMN is copied to input/business-processes/, stories and scenarios become pages. CI checks the generated files are current and that SUSHI builds with no errors. Then: publish the IG on gh-pages and add example resources per journey (bean IG-01).

3. The HCERT library at a pinned commit

lib/hcert/: a script and Gradle wiring that checks out smart-trust-tools at a pinned SHA, builds hcert-api and hcert-core (JVM) with that repository’s own Gradle and verification-metadata.xml, and copies the jars to lib/hcert/build/libs/, failing if a jar’s sha256 differs from the one recorded in lib/hcert/hcert.lock.json (the build is reproducible). The app and the Android tests depend on those files. Bumping the library is the commit in the lock plus build.py --update.

4. Android emulator test of the library (STT-GEN-004)

(In progress: android-test/ and .github/workflows/android.yml. The Gradle verification metadata is written on GitHub’s runner, since this session cannot reach Google Maven.)

android-test/: an Android library module whose instrumented tests run on an emulator in CI (Android API levels to be chosen: one current, one old enough to matter for clinics, e.g. 26 and 35):

Results are reported back to smart-trust-tools as evidence for STT-GEN-004 (and as beans there for anything Android refuses, e.g. algorithm names). Google Maven artifacts (AGP, AndroidX test) are pinned with sha256 like every other. The emulator job uses a pinned action by SHA and caches no unverified artifacts.

5. The app

(Started: app/, J2’s offline read, and J3’s verification and import of the guide’s DVCs on Android, workflow app.yml. A verified DVC becomes a FHIR R4 Immunization with its Provenance (dev.ohs.fhir:fhir-model-r4); an unverified one is not imported. Next: J3’s schedule, then J1 issuance.)

On the base the owner chooses from the proposal (recommended: a lean Compose Multiplatform app on the OHS KMP libraries, Android first):

An emulator UI test drives the four journeys end to end.

6. Web and publication

The web target (once hcert-core runs in a browser, or via the Rust/wasm library meanwhile), and the gh-pages site (agent-published): the journeys, BPMN viewers, personas and the test evidence.

Open items

Edit this page on GitHub