Hub prune-on-write deletes roster row -> cursor reset to 0 -> full-history message redelivery on reconnect

Sub-tickets

Questions

Activity

• system created · 2026-06-01
  wi cli
• system done · 2026-06-01
  Shipped hub v2.9.15 (commit 283cc5d). Root cause: pruneStale deleted roster row = sole cursor store -> reconnect re-INSERT at 0 -> /unread replayed from id 1 (2026-05-19). Fix: cursor moved to prune-immune cursors(agent PK, read_id) table; /register seeds MAX(id) for never-seen agents (no flood); migration backfilled from roster.last_delivered_id (nobody lost place). Verified live post-restart by PM: cursors table present, nw-whey-cc read_id=20158 preserved+advancing, no cursor reset to 0; cursors(45)>roster(40) confirms cursors survive prune as designed. Tests green (prune-survival/backfill/cold-seed).
• system completed · 2026-06-01
  status=done