Stop hub traffic for missing PMs and missing agents: kill successionNotify's promotion DM, strip dead idle-alert rule 28
Elazar: minimize all traffic for missing PMs in AROs - 'nobody cares' - and for missing agents - 'it is ok, if i need them i launch them'. Shutting an agent down is NORMAL operation, not an incident. Census of sender=llmmsg-srv-hub over 14d found the surviving emitter: successionNotify's SUCCESSFUL-election DM 'You are now PM of aro:X. Previous PM Y is unavailable (...)'. 12 fired in the 5 days to 2026-09-20, every one carrying previous_pm_present=false - i.e. every one caused by a deliberate shutdown, none actionable. bin-whey-cc received one for aro:infra-coders, a room already OBE. The other two successionNotify branches (fully-dark 2026-07-12, parked-seat Q5 2026-08-22) had already been suppressed for the same reason; this is the third and last. FIX: delete the emit, keep flushPmQueue (it delivers real held mail, not a notification). Also: served guide rule 28 still advertised the IDLE-AGENT ALERT and LLMMSG_IDLE_ALERT_* tunables, whose implementation MSG-414 deleted along with the stuck-agent detector (last stuck-agent emit 2026-09-19 22:02, 265 lifetime) - rewritten to state the hub does not alert on idle/stuck/missing agents and will not start. Guide 4.25 -> 4.26. /opt/llmmsg-srv/CLAUDE.md line 38 carried the same dead reference. NOT IN SCOPE, different owner (bin-venus-cc): scrp-stray_agent_check-venus, scrp-kpi_burn_alert-venus and the notify-elazar.sh email path are scrp-side emitters, not hub code.
Activity
-
985a28c, hub 2.61.6, live on venus 2026-09-22T03:17:07Z. successionNotify's promotion DM removed (flushPmQueue kept); guide rule 28 rewritten; guide 4.25 -> 4.26 incl. header + changelog; /opt/llmmsg-srv/CLAUDE.md dead LLMMSG_IDLE_ALERT_* reference retired. Pre-apply audit by audit-general-cc: first cut BLOCKed on unbumped guide header + stale changelog, re-cut PASSed.
-
EXERCISED: GET /guide against the restarted hub returns guide_version 4.26 and the new rule 28 verbatim. node --check clean, tree clean, package.json 2.61.6. NOT EXERCISED: the successionNotify hunk itself - proving the DM is gone needs a real PM departure (an electable member succeeding a dead PM), which was not forced on prod. Static evidence only: the emit and its stmtRosterLive liveness lookup are deleted, flushPmQueue still called on the same path, and no other caller of that DM exists. Absence of exercise is not verification of that hunk.
-
Shipped.