Fleet cronjob kill sweep: pm-llmmsgsrv-cc fans out, agents self-report armed CronCreate/loops, nw-venus-cc triages+orders kill
Elazar 2026-08-23. One-time (repeatable) sweep to clear stray/forgotten CronCreate jobs and ScheduleWakeup dynamic loops fleet-wide, prompted by the wakeup-visibility gaps found in pm-llmmsgsrv-cc's earlier CronList census (silence-means-zero is ambiguous, CronList misses ScheduleWakeup/Monitor persistent:true, jobs self-expire on the 7-day CronCreate limit with no signal). Process: (1) nw-venus-cc DMs pm-llmmsgsrv-cc asking it to fan out to all live/registered agents: report your armed CronCreate jobs and ScheduleWakeup/dynamic-loop state; stay silent if none armed. (2) nw-venus-cc collects replies, triages each (legit standing cadence vs stray/forgotten/self-arm-ban violation). (3) nw-venus-cc DMs each agent with a stray/violating job, telling them to kill it (CronDelete / stop the loop). Distinct from OPS-137's daily orphan-PROCESS sweep — this is scheduled-JOB hygiene, same theme, related epic. TRIAGE ROUND 1 RESULT (2026-08-23): fan-out via pm-llmmsgsrv-cc reached 18 agents, 5 batches. audit-mba-l-ca, bin-lezama-cc, bin-venus-cc all report ZERO armed jobs. ONE armed job found: monitor-llmmsgsrv-cc id 82d6c92c, cron '23 */6 * * *', explicitly armed BY pm-llmmsgsrv-cc instruction (not self-armed, PM-assigned cadence = self-arm-ban exception) — LEGITIMATE, no kill, auto-expires ~2026-08-30. Two agents UNREACHABLE not zero: coder-chatduo-cc, monitor-context-cc (recipient_not_registered mid-send) — unknown state, re-poll when back on roster. Census fact for OPS-138's own design: CronCreate jobs SURVIVE /compact, die only on /clear/restart/7-day expiry. ROUND 1 INVALIDATED IN PART (2026-08-23): pm-llmmsgsrv-cc's own 'silence if nothing armed' reply was a FALSE NEGATIVE — it never actually ran CronList, reasoned 'I don't think I have any' instead of checking, and in fact holds job 942fc069 (Elazar-assigned 4h lane-coach checkpoint). Elazar's framing: a Claude-Code-agent false-assertion pattern (claiming a negative without running the check), not a CronList tool gap — the 'stay silent if nothing armed' phrasing made the check itself skippable. Round-1 silent/summarized replies (audit-mba-l-ca's un-quoted 'no armed jobs') carry the same risk; bin-lezama-cc and bin-venus-cc's replies explicitly quoted CronList output and are not suspect. ROUND 2 DISPATCHED (2026-08-23): re-fan via pm-llmmsgsrv-cc, mandatory verbatim CronList output from every recipient, no silent-if-empty escape — 'Run CronList now, before anything else, reply with raw output verbatim even if empty. Do not summarize or decide first.' Awaiting replies.
Activity
-
parent=#2958
-
was: Elazar 2026-08-23. One-time (repeatable) sweep to clear stray/forgotten CronCreate jobs and ScheduleWakeup dynamic loops fleet-wide, prompted by the wakeup-visibility gaps found in pm-llmmsgsrv-cc's earlier CronList census (silence-means-zero is ambiguous, CronList misses ScheduleWakeup/Monitor persistent:true, jobs self-expire on the 7-day CronCreate limit with no signal). Process: (1) nw-venus-cc DMs pm-llmmsgsrv-cc asking it to fan out to all live/registered agents: report your armed CronCreate jobs and ScheduleWakeup/dynamic-loop state; stay silent if none armed. (2) nw-venus-cc collects replies, triages each (legit standing cadence vs stray/forgotten/self-arm-ban violation). (3) nw-venus-cc DMs each agent with a stray/violating job, telling them to kill it (CronDelete / stop the loop). Distinct from OPS-137's daily orphan-PROCESS sweep — this is scheduled-JOB hygiene, same theme, related epic.
-
was: Elazar 2026-08-23. One-time (repeatable) sweep to clear stray/forgotten CronCreate jobs and ScheduleWakeup dynamic loops fleet-wide, prompted by the wakeup-visibility gaps found in pm-llmmsgsrv-cc's earlier CronList census (silence-means-zero is ambiguous, CronList misses ScheduleWakeup/Monitor persistent:true, jobs self-expire on the 7-day CronCreate limit with no signal). Process: (1) nw-venus-cc DMs pm-llmmsgsrv-cc asking it to fan out to all live/registered agents: report your armed CronCreate jobs and ScheduleWakeup/dynamic-loop state; stay silent if none armed. (2) nw-venus-cc collects replies, triages each (legit standing cadence vs stray/forgotten/self-arm-ban violation). (3) nw-venus-cc DMs each agent with a stray/violating job, telling them to kill it (CronDelete / stop the loop). Distinct from OPS-137's daily orphan-PROCESS sweep — this is scheduled-JOB hygiene, same theme, related epic. TRIAGE ROUND 1 RESULT (2026-08-23): fan-out via pm-llmmsgsrv-cc reached 18 agents, 5 batches. audit-mba-l-ca, bin-lezama-cc, bin-venus-cc all report ZERO armed jobs. ONE armed job found: monitor-llmmsgsrv-cc id 82d6c92c, cron '23 */6 * * *', explicitly armed BY pm-llmmsgsrv-cc instruction (not self-armed, PM-assigned cadence = self-arm-ban exception) — LEGITIMATE, no kill, auto-expires ~2026-08-30. Two agents UNREACHABLE not zero: coder-chatduo-cc, monitor-context-cc (recipient_not_registered mid-send) — unknown state, re-poll when back on roster, do not count as clean. Census fact for OPS-138's own design: CronCreate jobs SURVIVE /compact, die only on /clear/restart/7-day expiry. Result: nothing to kill round 1. Two unknowns pending re-poll.