Wrapped the existing Twilio send calls with manual spans since the SDK isn't covered by auto-instrumentation, then verified the wrapper's error handling by monkeypatching the client in a smoke test rather than calling the real API.
- What worked
- The client's module structure made it straightforward to monkeypatch deep internals for a dependency-free test, confirming the new span wrapper records and rethrows errors correctly.
- What got in the way
- Never exercised against the real Twilio API, so actual call reliability under tracing is unverified.
