mirror of
https://github.com/MBombeck/HealthLog.git
synced 2026-11-07 04:12:04 +00:00
Marc-André Bombeck
c2ab9cda11
The live evaluation now captures the tool calls a generation makes, with their arguments, and hands them to the judge. runScenarioLive puts a scenario to the model in the chat turn's own context (system prompt, tool-mode and dialog addenda, DATA INVENTORY, EARLIER TABLES, transcript) over the real tool catalogue, answers each call with a stand-in result, and reads the reply through the turn's clarification and chip parsers. runScenarioJudge grades each scenario a few times with the same evaluateScenario the deterministic suite uses and reports a pass rate per scenario; the server-only scenarios and checks are left out. Like the existing judge it runs only with COACH_EVAL_API_KEY, from the nightly workflow's non-blocking step, and never throws.
scripts/
One-shot operator and build scripts. These are not part of the running application — they are run by hand during a deploy, a backfill, or a release, or wired into CI. Recurring work belongs on pg-boss (src/lib/jobs/), not here.
How to run them
From a source checkout:
pnpm dlx tsx scripts/<file>.ts
The production image intentionally removes npm, pnpm, and Corepack. Only
container-safe scripts named by an operator runbook use its pinned
healthlog-tsx launcher; scripts that import the full application dependency
graph must run from a source checkout. scripts/ is excluded from the project
typecheck.
What lives here
generate-openapi.ts— emitsdocs/api/openapi.yamlfrom the registered Zod schemas (pnpm openapi:generate).check-openapi.ts— compares a freshly generated spec against the committed YAML; CI fails on drift (pnpm openapi:check).check-env.ts— pre-deploy env-var sanity check againstenv-manifest.json(pnpm check-env); wired into CI.assert-deploy.ts— post-deploy/api/versionassertion.rotate-encryption-key.ts— AES-256-GCM key rotation; seedocs/ops/encryption-key-rotation.md.restore-backup.ts— restore an off-host AES-GCM backup from an S3-compatible bucket.repair-apple-health-distance.ts— repair walking-distance rows theexport.zipimport stored 1000x too small; dry-run by default, seedocs/ops/apple-health-distance-repair.md.repair-intake-anomalies.ts— repair historic medication-ledger defects (duplicate dose-slot rows, implausibletaken_at); dry-run by default, seedocs/ops/intake-repair.md.backfill-rollups.ts,drain-per-sample-cumulative.ts,backfill-mood-note-column.ts— historical data backfills (note: the boot-timerollup-full-backfillqueue handles new accounts automatically; the CLI is the manual fallback).seed-demo.ts+seed-demo-v15.sql— populate a demo tenant.test-notifications.ts— exercise the notification dispatcher + reminder check.fetch-geolite2.sh— download the MaxMind GeoLite2 MMDBs (needs a licence key) beforedocker build.install-hooks.sh— install the local git hooks.generate-sw-version.mjs,print-bundle-report.mjs— service-worker version stamp and bundle-size report.i18n/— locale tooling;__tests__/— tests for the scripts above.
Conventions
See ../CLAUDE.md "Tests and commands" + "Self-hosting gotchas". Operator runbooks that drive these scripts live in ../docs/ops/.