#15173: fix(session): reset totalTokens after compaction when estimate unavailablefix(session): reset totalTokens after compaction when estimate unavai…

## Summary When estimateCompactionTokensAfter() returns undefined, the pre-compaction totalTokens was left unchanged. This caused shouldRunMemoryFlush() to read stale high values and trigger unnecessary memory flushes immediately after compaction, potentially creating a compaction-flush bounce cycle. ## Changes - In incrementCompactionCount(), reset totalTokens to 0 when tokensAfter is unavailable - - The correct value is populated on the next turn via persistSessionUsageUpdate - - ## Why this works - Setting totalTokens to 0 is safe because persistSessionUsageUpdate runs on the very next turn and overwrites it with the actual value from the model response. The brief window where totalTokens=0 prevents the false flush trigger. - - Fixes #15101…lable When estimateCompactionTokensAfter() returns undefined, the pre-compaction totalTokens was left unchanged. This caused shouldRunMemoryFlush() to read stale high values and trigger unnecessary memory flushes immediately after compaction, potentially creating a compaction-flush bounce cycle. Reset totalTokens to 0 when tokensAfter is unavailable. The correct value is populated on the next turn via persistSessionUsageUpdate. Fixes #15101 <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates `incrementCompactionCount` in `src/auto-reply/reply/session-updates.ts` to avoid stale `totalTokens` values immediately after a compaction when no post-compaction token estimate is available. In the `tokensAfter`-missing path, it now sets `updates.totalTokens = 0` (and clears `inputTokens`/`outputTokens`) so `shouldRunMemoryFlush` won’t see a high pre-compaction token count and immediately trigger a flush. This fits with the existing `shouldRunMemoryFlush` logic (`src/auto-reply/reply/memory-flush.ts:83-86`), which returns `false` when `totalTokens` is falsy/<=0, so the post-compaction window won’t cause an immediate flush-trigger bounce. <h3>Confidence Score: 5/5</h3> - This PR is safe to merge with minimal risk. - Change is small, localized to session token bookkeeping after compaction, and aligns with existing `shouldRunMemoryFlush` behavior (treating `totalTokens<=0` as no-flush). No other callsites depend on `totalTokens` being preserved in the specific case where no `tokensAfter` estimate is available. - src/auto-reply/reply/session-updates.ts _{Last reviewed commit: 657170a} <!-- greptile_other_comments_section --> _{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!} <!-- /greptile_comment -->