Actuator tiers must be per-agent-window: scale absolute gates (HARD_NUDGE_K/HARD_COMPACT_K, quarantine) by each agent's own CLAUDE_CODE_AUTO_COMPACT_WINDOW instead of assuming one fleet window

OPS-128 agent-ops Done high

AUDIT dropbox-orphan-sweep.sh (registry row 116): what it deletes, how it fails

OPS-127 agent-ops Done high

cc-context-monitor last_turn() v2.14 residual stale-read: agent force-compacted 'for no reason' every other session. Verified 2026-06-09 nw-whey-cc reported 184k while live /context=63.6k (3x phantom). last_turn (~line 312) reads transcript's last usage block = summarizer's pre-compact cache_read until next real turn; compact_boundary reverse-walk skip has a hole (boundary outside TAIL_BYTES, or pre-compact reading sampled in the gap before boundary written). Fix: gate actuator on LIVE window occupancy, not stale transcript estimate. NOT a cap problem - do NOT bump nw ROLE_CAP (Elazar explicit). Recurring user-facing churn.

OPS-11 agent-ops Done high

tick.py execute() is sequential: one slow agent blinds the whole roster and delays post-nudge escalation (2.5min measured)

OPS-106 agent-ops Done high

Post-compact state is structurally unmeasurable: 0k filter hides landed compacts, so compact_unverified fires on successes

OPS-112 agent-ops Done high

No regression guard on the idle classifier: a ghost-vs-typed inversion caused a silent fleet-wide coverage outage

OPS-123 agent-ops Done high

Stuck-busy agents leave coverage silently: unsubmitted prompt text pins idle=false, 14h measured uncovered, no alert

OPS-116 agent-ops Done high

abr:ss attribution false-negative: fast responders fail the text_in_pane visibility test

OPS-121 agent-ops Done high

nudge_pending_since consumed on decide, not on confirmed exec: an aborted escalation can shut all four triggers

OPS-108 agent-ops Done high

hub_send.py marker path was __file__-relative: correct only because the unit names the symlink, and it reported landed registrations as failures

OPS-114 agent-ops Done high

Idle prefetch used the 50% growth floor and compacted agents with 35pt of headroom (coder-pluto-cc at 51.2%)

OPS-113 agent-ops Done high

Idle-path compact runaway: NULL growth baseline made every above-floor agent perpetually eligible (pm-terra-cc 8x in a morning)

OPS-111 agent-ops Done high

db-size-monitor.service failing exit-127 daily (orphaned script)

OPS-104 agent-ops Done high

Fix rkhunter duplicate timers on venus + drop cadence weekly->monthly

OPS-98 agent-ops Done high ↳ OPS-97

nw-agent launcher template: haiku + --effort low + autoCompact + loop self-monitor rule

OPS-37 agent-ops Done high

whey gdrive-bisync broken (aborting every ~30s, 'cannot find prior Path1/Path2 listings') since ~2026-05-16: evolutiva/venus/odaia backups local-only in /gdrive, not reaching Google Drive; needs rclone --resync recovery (path2/local-wins) on Elazar GO

OPS-77 agent-ops Done high

bwi Tier 1 enhancements: add guard, mine banner, rot, status notify, evolutiva doc

OPS-38 agent-ops Done high

Pin hub registrations for cc-context-monitor-{whey,venus,lezama}

OPS-44 agent-ops Done high

memory-budget Layer 1: gdrive-canonical 'Memory Budget' directive + footprint-check threshold (thread 3); incl. maintainer-agents.md memory-lint ownership section (thread 4)

OPS-48 agent-ops Done high

v2.7 over-cap remediation: parked over-cap agent still in its AROs gets woken into a full-context read-turn by every ARO fan-out (~30-60s) — park does NOT stop the burn. Fix: over-cap sequence becomes ss -> DM handoff to PM -> aro_leave ALL AROs -> idle. Edits: v2.6 over-cap DM text in cc-context-monitor.sh (bin-whey-cc) + wind-down line in global directive (nw-whey-cc). Fresh bin-whey-cc takes alongside thread 2

OPS-50 agent-ops Done high

memory-lint §12/§9 vs global CLAUDE.md: per-agent identity + dates in memory/ — Elazar decision

OPS-52 agent-ops Done high

chat-duo: render origin_aro-tagged messages in their ARO room, stop spawning a room per DM

OPS-56 agent-ops Done high ↳ OPS-54

evolutiva-backup down for mars+pluto: Supabase removed legacy db.<ref>.supabase.co direct hostnames (NXDOMAIN) - DSNs need pooler connection strings

OPS-67 agent-ops Done high

settings.json PostToolUse push-to-gdrive-canonical silently no-ops: basquetwi.md local edits did NOT propagate up to /gdrive/config/claude-shared/ (canonical stayed mtime May-18), so SessionStart cp re-pulled STALE over the edit. Hook's 'cp ... 2>/dev/null' swallows failures = silent doc drift fleet-wide. Audit why push no-ops + surface failures.

OPS-15 agent-ops Done high

cc-context-monitor: send ss+Enter, wait 60s, then send /compact (not back-to-back)

OPS-20 agent-ops Done high

Full inventory + sense-check sweep of all current venus systemd timers/cron.d

OPS-103 agent-ops Done ↳ OPS-97

mars-uxpain-relay.service failing exit-1 weekly (origin_aro_not_member)

OPS-105 agent-ops Done

Build heartbeat/dead-man's-switch watchdog for registered scheduled jobs

OPS-102 agent-ops Done ↳ OPS-97

Design + build fleet-wide scheduled-job registry (systemd timers + cron.d) on venus

OPS-101 agent-ops Done ↳ OPS-97

cc-context-monitor: fix silent dedup + add process-not-in-roster alert

OPS-68 agent-ops Done

cc-context-monitor v2.4: M2 absent-branch hardening

OPS-45 agent-ops Done

cc-context-metrics: SQLite token-metrics time series (cc-context-monitor v2.3)

OPS-43 agent-ops Done

lezama host-global plugin baseline alignment - disable 8 plugins + uninstall telegram

OPS-64 agent-ops Done

memory-lint C: cron + report-file. Schedule nw-whey-cc (whey projects) + nw-venus-cc (venus) daily 03:00 ART. On hard-fail, write ~/.claude/projects/<slug>/memory/_lint-report.md (NOT DM — survives /clear). Owning agent reads at SessionStart, does LLM cleanup.

OPS-41 agent-ops Done

Inventory systemd units + standalone scripts across hosts (whey/lezama/venus): name, host, owner ccs, function, alert path. Seed: backup units, ccs-context email alerts, defunct llmmsg-srv-kpis.

OPS-27 agent-ops Done ↳ OPS-26

cc-context-monitor v2.5: cum_turns column (turns/hour + tokens-per-turn)

OPS-46 agent-ops Done

agent-health-watch.{service,timer}: 5min health check (process liveness, model, ARO membership, settings.json drift) + email/push on drift; v1 whey only

OPS-36 agent-ops Done

gitpush.sh strict-staging mode: .gitpush-strict repo marker + --only <pathspec> + abort (not prompt) on stray files outside the set; non-strict repos keep add -A. Bump version same commit. Drop+commit marker in sh.git + llmmsg-srv repos.

OPS-78 agent-ops Done

/rtshared low-latency WAN mount: whey:/var/rtshared served to venus+lezama as /rtshared/ over ZeroTier (sshfs or NFS, patient timeouts) for live cross-host editing; keep /gdrive for bulk. GATED on lezama coming online; whey /etc side is ready.

OPS-87 agent-ops Done

notify-nw helper: report scripts (rkhunter-scan, clamav-scan, gdrive-usage-watchdog) DM the host's nw-<host>-cc via hub POST /send (sender self-registers), so nw agents get security/health reports directly instead of Elazar relaying. Optional --report-file drops full HTML to /var/log/nw-reports/<host>/ with a systemd-tmpfiles.d 3-day GC line. Event-driven: NO haiku poller, NO inbox agent, zero idle cost. Helper lives in ~/.local/bin (bin-whey-cc PATH lane, co-owned); wire into the 3 scripts + bump each version.

OPS-88 agent-ops Done

VPN work-hours auto-connect: connect-only reconciler (Mon-Fri 09:00-16:00 ART), never disconnects, exponential backoff on failed connect

OPS-89 agent-ops Done

whey-nw-cc: scheduled open ARO-ask digest for Elazar (hub query → notify-elazar.sh)

OPS-35 agent-ops Done

whey-nw-cc: scheduled wi pending --to elazar digest → notify-elazar.sh + chat-duo

OPS-34 agent-ops Done

memory-lint KPI email spam + report-never-clears: batch kpi-enforce emails into one digest; exclude _lint-report.md from its own lint sweep; fix infer_agent misattribution

OPS-66 agent-ops Done

Restore simple llmmsg KPI timer/report: daily compare worst 3 coached agents vs yesterday/week

OPS-33 agent-ops Done

CLAUDE.md instruction drift: stale references to .agent-name-cc in agent startup checklists (fleet-wide sweep)

OPS-75 agent-ops Done ↳ OPS-73

Zombie ca-session detector: scan llmmsg-srv bridge registrations for lastActivityAt > N hours + token totals; canonical example sh-ca-w (~448k tokens, ~21h idle as of 2026-05-09T07:00). Output list + kill recommendation.

OPS-29 agent-ops Done ↳ OPS-26

Implement scrp-<script_name>-<host> sender naming: notify-elazar.sh + notify-nw.sh derive scrp-$(basename caller, strip ext, - to _)-<host> and stamp it as hub sender + email From-name/subject prefix; rename direct-hub-poster scripts (evolutiva-errscan errscan-whey-cc to scrp-evolutiva_errscan-whey, llmmsg-kpi-report llmmsg-srv-cc fallback to scrp-kpi_report-whey, sweep rest). -cc/-ca reserved for real agents. Convention: agent-conventions.md Script/Automated Sender Naming (Elazar 2026-06-10)

OPS-13 agent-ops Done

bwi-desweep-roundtrip-test

OPS-96 agent-ops Done

Scrub stale eq / Elazar Pending text from bwi-tool PROJECT canonical docs (MSG-36 followup): /home/rob/Documents/work/pensanta/projects/basquetWi/basquetWiInstructions.md + that project's CLAUDE.md still carry old eq/Pending wording. Not the global-rule SSOT (gdrive is, already fixed) but scrub for consistency. Owner: bwi-tool maintainer.

OPS-16 agent-ops Done

cc-context-monitor FIX#1+#2: idle-sense (last-ASSISTANT-ts not getmtime) + safe-compact REPL-ready marker

OPS-19 agent-ops Done

monitor peer-watchdog: roster-presence→probe-age (post-wave-1 prune-churn fix)

OPS-18 agent-ops Done

applog-listen: catchUp re-pages inflight rows when postAlert fails on reconnect

OPS-92 agent-ops Done

Strip stale telegram:false plugin entries from mars/pluto agent configs

OPS-62 agent-ops Done low

Strip stale telegram:false plugin entries from movilba-web agent configs

OPS-63 agent-ops Done low