Greptile Summary

Implements exponential backoff for heartbeat retries when requests are in-flight, replacing the previous behavior of immediately advancing to the next full interval. The implementation adds a consecutiveSkips counter to track retry attempts and schedules retries at 1m, 2m, 4m, 8m, and 16m intervals before falling back to the normal schedule.

Key changes:

• Added consecutiveSkips field to HeartbeatAgentState to track consecutive in-flight skips
• Modified advanceAgentSchedule to reset the skip counter on successful runs
• Implemented exponential backoff with Math.min(60_000 * Math.pow(2, skips - 1), agent.intervalMs) to cap retries at the agent's interval
• Falls back to normal scheduling after 5 retry attempts (~31 minutes total)

Issue found:

• consecutiveSkips is not preserved when config is updated (line 1066-1072), which will reset the retry counter mid-backoff if the config is reloaded

Confidence Score: 4/5

• Safe to merge after fixing the consecutiveSkips preservation issue
• The exponential backoff implementation is mathematically correct and improves the existing behavior. One logical bug was found where consecutiveSkips is not preserved across config updates, but this only affects the edge case when config is reloaded during active retries. The core retry logic is sound.
• Fix the consecutiveSkips preservation in updateConfig function (line 1066-1072)

_{Last reviewed commit: 134c656}

Thanks for the catch — this is actually already handled in the current code. Line 1072: consecutiveSkips: prevState?.consecutiveSkips preserves the counter across config reloads. The suggestion matches what's already implemented.