Troubleshooting

Google Calendar Connected, But No Events?

A fast, proof-first workflow to fix calendar visibility loops without endless reconnect retries.

Common pattern: Calendar shows as connected, but the agent says there are no events, misses meetings, or keeps asking to reconnect.

Reality check: “Connected” is not proof. You need one successful canary read from the exact calendar you care about, in the same surface/session.

Fast path (4 steps)

1) Confirm the exact Google account + calendar

Most failures are account mismatch (personal vs workspace) or wrong calendar selection. Verify the exact calendar name before testing.

2) Complete OAuth once in one clean browser profile

Finish every consent/security screen in one pass. Partial OAuth completion causes reconnect loops with stale access state.

3) Run a tiny canary read

Ask for events in a short known window (for example, next 24 hours) and require event evidence:

Return event title + start time + timezone + calendar name. If none found, return explicit "0 events" and the exact window queried.
4) Only then trust reminder or planning workflows

Do not jump into full daily planning until canary read proof is consistent in the same chat surface/thread.

Copy/paste recovery prompt

My Google Calendar appears connected, but event reads are unreliable. Do this exactly: 1) Show which Google account is currently authorized. 2) Query ONLY the next 24 hours. 3) Return: calendar name, queried time window, timezone, and each event title + start time. 4) If zero events, return explicit "0 events" plus exact raw error text if any. 5) Do not reconnect unless you show a specific auth/scope error first.

High-friction failure patterns

Advanced edge cases (when basics pass but reads still fail)

Loop-break rules

Stop reconnect spirals: no more than one reconnect per diagnostic cycle. If canary read still fails, escalate once with exact evidence instead of repeating setup.

What to include if you escalate

Bottom line: prove one canary read first. If it works, scale. If it fails, escalate with exact evidence once—don’t burn cycles in reconnect loops.