Skip to content

Observability

Effectively none.

What exists

  • console.log / console.error calls in AuthContext.tsx (login flow, debug logs)
  • console.error(\API ${method} request failed:`, error)inapi.ts`
  • Toasts to the user via sonner and shadcn toast

What does not exist

Capability Status
Error tracker (Sentry, Bugsnag, Rollbar) ❌
Real User Monitoring (Sentry RUM, Datadog RUM, SpeedCurve) ❌
Analytics (GA, Mixpanel, Amplitude, Segment) ❌
Console-log scrubbing for production ❌ — console.log("login - extracted user:", user) ships to prod with sensitive PII
Source-map upload to an error tracker ❌
Session replay (FullStory, LogRocket) ❌
Web Vitals reporting ❌
Feature flag system ❌

Specific concerns

  1. console.log is left in production. AuthContext.tsx has lines like console.log("login - extracted user:", user) that fire on every login. PII in console (visible to anyone with devtools, but also potentially captured by any browser extension the user has installed) is a moderate exposure.

  2. Fix: a tiny logger wrapper that no-ops in production, plus a Vite build-time replacer (drop_console in esbuild config) to strip residual ones.

  3. No source-map upload. If the team eventually adds Sentry, the production minified stack traces will be unreadable without source-map upload at build time.

  4. No correlation with backend logs. The FE doesn't propagate a request id; the BE doesn't issue one. Reconstructing a user's session across FE + BE logs is impossible without both ends emitting a shared id.

  5. No usage telemetry. The team has no data on which admin features are actually used. If /prompts is unused (which is consistent with the route being hidden from the sidebar), no one knows for sure.

Minimal improvements (in cost order)

  1. Strip console.log in production — vite.config.ts: 
    esbuild: { drop: ['console', 'debugger'] },
  2. Add web-vitals — 5-line integration to report LCP/INP/CLS to any backend endpoint.
  3. Add a feature-flag library (e.g., Statsig SDK, ~10 lines) so the team can ship dark — relevant if/when the admin tool grows.
  4. Add Sentry — @sentry/react SDK with BrowserTracing + source-map upload step in CI. ~30 minutes of setup.
  5. Add an analytics SDK — only if the team will actually use the data. Otherwise it's data theatre.

Recommendations

For a 10-50 user admin tool, the cost/benefit of full observability is low. Priority should be:

  1. Strip console.log (this week)
  2. Add Sentry (next sprint) — catches the next regression before users complain
  3. Add Web Vitals (next sprint) — cheap, no downside
  4. Defer analytics until there's a concrete decision riding on usage data

The platform-wide observability gap is larger; addressing it here first is reasonable (low traffic, internal users tolerate breakage during instrumentation).