Repository navigation
Verify queued-message completion and ordering after the restart snapshot #45
Description
PR #43 records the first coordinated update's restart observation. At the retained 2026-11-12 07:47:57 UTC snapshot, the 59 original queued messages were 21 completed, two running and 36 queued. No duplicate continuations were observed, but later completion and ordering of the outstanding queued messages were not verified by that snapshot.
Track this coverage limit separately from completion of the bounded restart observation. If later evidence is collected, reconcile the original message/run IDs and their per-thread dispatch order without starting replacement turns or rerunning accepted messages. Do not treat missing later evidence as a demonstrated duplication defect.
The decision record in PR #43 will explicitly mark later queue completion and ordering unverified. This P2 follow-up does not hold that documentation merge.
Source finding: #43 (comment)
Activity
chrisohearn commentedon Nov 13, 2026
Seen again in the forced update to 0.1.10-preview.20261113.1 (2026-11-13 01:02Z), in thread mcp:82704295 "Coordinate Shared Resource Recovery":
behavior-report-…-delivery, requested 00:57:29Z).Queued sends behind an interrupted run should run after the continuation rather than be cancelled.
🤖 Generated with Claude Code
chrisohearn commentedon Nov 13, 2026
Status: still valid, open evidence. A 2026-11-13 01:02Z update again cancelled a queued delivery (run 74) when restart continuations started. Tracked as item 7 of the round-four plan (queued messages cancelled before starting are never re-run). Ledger:
jarvis-t3/round-4-upgrades; no worker row for item 7 yet.Triage by the round-four GitHub hygiene pass (ledger key
r4/w7-github), 2026-11-13. Left open.