You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
## Why another fix?
Message-ID guard reduced duplicates but users still observed rare
double-sends.
Root cause in those cases: two distinct summary messages were created
for
one compaction-run (different IDs) due to duplicated completion signals
(delete + message + final bump, reconnect/replay, or abort/accept
flows).
## What changed
- Add requestId to compaction-result metadata (source compaction-request
user msg)
- WorkspaceStore: dedupe performCompaction by compaction-request id
- useAutoCompactContinue: guard on requestId when present, fallback to
message id
- Fix retry cleanup to remove the correct guard key
## Why this is obviously correct
- A compaction-run is uniquely identified by the compaction-request user
message ID
- We now:
- Perform compaction only once per requestId (store-level dedupe)
- Send continue only once per requestId (hook-level idempotency)
- If requestId is missing (unexpected), fallback to summary message id
(still safe)
## Testing notes
- Run repeatedly, interrupt/retry, and force reconnection
- Expect exactly one continue send per compaction-run
_Generated with _
0 commit comments