Operator Observability: Doctor, Status, and Logs

Good observability keeps Rocky from guessing. Health commands, logs, and live checks should tell the story before code or configuration is changed.

Start with health signals

Use hermes doctor for dependency and configuration checks.
Use hermes status for component-level status where available.
For gateway issues, inspect recent gateway logs for send failures or platform errors.
For deployed sites, distinguish local app logs from Cloud Run revision logs and live HTTP behavior.

Triage questions

Is the model/provider configured and authenticated?
Are required toolsets enabled for the current platform/profile?
Did a config or tool change require a fresh session?
Is the failure local, network, provider, gateway, or live deployment?

Pitfalls

Do not paste logs containing secrets into public pages.
Do not assume a model failure is a code bug before checking provider auth and quota.
Do not report that a public route works from local tests alone.

Verification steps

Run the relevant health command and save only non-sensitive findings.
Reproduce the smallest failing action.
Apply one fix at a time.
Re-run the original failing command and a user-visible check.