My 5-hour quota on the 5x plan just went from 100% at 22:32 to 0% at 23:04 while running only two sessions with subagents.
I dug into my local logs to see how that was even possible, and prompt caching completely broke mid-session:
- Same session, same model, 12s apart: cache reads dropped from 242,032 tokens straight to 0.
- Another main session sent 7 requests in a row with 0 cache reads, burning through 2.84M raw input tokens.
Logs are attached. Is there a known caching bug right now or something broken in the client? Paying for the 5x tier makes zero sense if broken caching can wipe out an entire 5-hour window in half an hour.
Can someone from the team look into these logs, share a per-request usage breakdown, and restore the burned quota + credit my account with extra usage for getting locked out?
