Trace One Failed Request Through the Logs
A page can fail after three services have each logged something that looks successful. Start with one request identifier, not the loudest error message.
Give Claude Code a redacted log sample and the time of one failed attempt. Ask it to connect the request from the browser response through the server handler to the upstream call. The useful output is a short timeline with a source for each step and a clear gap wherever the identifier disappears.
Do not paste access tokens or complete customer records to make the logs easier to follow. If correlation is missing, add a harmless request ID at the boundary and reproduce with synthetic data. That may reveal a timeout or wrong response shape before anyone spends time replacing working credentials.
