refactor: migrate to observability v1

[vreshch]

Moves landing onto @agentage/observability v1 - the module-singleton log, the /react reporter component, and the fetch-native health() factory.

Changes

• @agentage/observability ^0.20.0 -> ^1.0.0 (packages/landing), lockfile regenerated with npm 10. The v1 tree drops @opentelemetry/sdk-node for an explicit tracer provider, so the lock shrinks by ~700 lines.
• src/lib/logger.ts deleted. createLogger({ service }) is gone in v1 - log is a module singleton that takes its service from OTEL_SERVICE_NAME, which docker-compose.yml already sets to agentage-landing. instrumentation.ts was its only consumer.
• instrumentation.ts now imports the singleton and keeps register re-exported from /next.
• Browser reporter is <ErrorReporter /> from @agentage/observability/react instead of a hand-rolled useEffect + installErrorReporter. Same endpoint/service options, so the beacon target is unchanged.
• /health uses health() (fetch-native handler factory) instead of healthResponse(). Identical envelope; verified locally.

The cross-origin collector is unchanged

Landing posts client errors cross-origin to the web backend at ${API_URL}/errors/report (prod: https://api.agentage.io/api/errors/report), built from the same NEXT_PUBLIC_SITE_FQDN build arg as the waitlist POST. <ErrorReporter /> is passed that exact same ERROR_REPORT_URL - this PR does not touch the endpoint, and web keeps that route alive for exactly this caller.

Worth knowing about the v1 browser lane: observeBrowser also mints a W3C trace id per user action and sends it as a traceparent header on same-origin fetch calls only. Cross-origin requests - including this beacon - are deliberately untouched, so the trace id rides the error payload rather than the collector request. Landing makes few same-origin fetches, so this is a no-op here.

Deviation from the v1 notes

onRequestError from /next is documented as the non-deprecated replacement, but in the published 1.0.0 it is still the factory onRequestError(log), not a ready value - dist/internal/patch/error-emitters.js exports function onRequestError(log). Re-exporting it directly would have handed Next a function that treats its err argument as a logger and silently swallows every server error. Kept the factory form, now fed the root singleton:

import { log } from '@agentage/observability';
import { onRequestError as kitOnRequestError } from '@agentage/observability/next';
export { register } from '@agentage/observability/next';
export const onRequestError = kitOnRequestError(log);

Worth a follow-up in the kit if a value form was the intent.

Verification

• npm run verify green: type-check, lint, format:check, 70 tests in 8 files, production build (42 pages).
• next dev booted from the worktree and driven with Playwright/Chromium:
  • <html data-obs-trace> = e9ebcabf0ab8139800b8d6dbd9fc3eda on load (valid 32-hex), and a fresh id after a history.pushState navigation - one id per user action, as designed.
  • A thrown browser error beaconed POST http://localhost:3001/api/errors/report (local API_URL), confirming the reporter is installed and the endpoint is byte-identical to before. ERR_CONNECTION_REFUSED on that POST is expected - no backend runs locally.
• GET /health answers the standard envelope: {"success":true,"data":{"status":"ok","service":"unknown",...,"durationMs":0.1}}. service is unknown only because OTEL_SERVICE_NAME is unset in local dev; compose sets it in the deployed container.
• Swept for removed exports (captureError, createLogger, withSpan, tracedFetch, installErrorReporter, healthResponse) - none remain. @agentage/shared does not touch the kit; it is the only other package in this repo.

⚠️ DEPLOY CAUTION

This repo deploys BOTH environments on push, and a green workflow run can still leave prod stale. After merge, do not trust the green check - commit-assert the deployed surface:

curl -s https://agentage.io/health   | jq -r .data.commit
curl -s https://docs.agentage.io/health | jq -r .data.commit

Both must equal the merge commit's short SHA. A frozen checkedAt or an old commit means prod did not roll; redeploy before calling it shipped.

[github-actions]

✅ CI - Verify PASSED

Commit: 5534ed4f925f50cbab713e954c4015d11f654fd0 · Branch: feature/obs-v1

• ✅ Type check
• ✅ Lint
• ✅ Format check
• ✅ Unit tests + coverage
• ✅ Build
• ✅ Landing e2e

🔗 View workflow run