EPIC: agents-efficiency-KPI - per-agent context/token efficiency report Elazar actually sees (Elazar 2026-07-02: 'never seen a report on that'). Inputs: usage polls + context samples (venus PG). Deliverable: recurring per-agent report (tokens/turn, cache-read overhead, context growth, compact frequency) surfaced to Elazar. Coordinate: bin-whey-cc (one-db data), coder-llmmsgsrv-cc (kpi views/report script)

SAVETOKE-33 saveTokenz In progress high

└─ agents-efficiency-KPI v2: settle the actuator coverage gap and make the report recurring

SAVETOKE-35 saveTokenz Backlog high

└─ one-db: retire non-essential telemetry fleet-wide per Elazar - drop network/dns/ping probes + dbsize + kpi.checks collection AND all cross-host telemetry shipping (lezama+whey write nowhere remote; supersedes ONEDB-15 repoint branch); KEEP usage polls + context samples on venus only (efficiency-KPI inputs + compact-monitor actuator data, pending Elazar veto on context)

SAVETOKE-34 saveTokenz In progress high

Epic: Mars — mars ARO token reduction

SAVETOKE-3 saveTokenz Backlog high

└─ coder-mars-cc/audit-mars-cc oversized ARO design-pings: 4400-char 'never-directive' mandates to ARO; headline+summary in ARO, full detail DM-to-PM/assignee

SAVETOKE-12 saveTokenz Backlog

└─ audit-mars-cc: adopt array multi-DM (to:[...]) for shared findings — 394 msgs/24h ALL 1-to-1, same finding sent separately to pm+coder; ~10,600 tokens/day (biggest single Mars lever)

SAVETOKE-9 saveTokenz Backlog

└─ pm-mars-cc <-> db-mars-cc ping-pong: 27 rapid exchanges/24h (~6,315 tokens); batch dispatch+result into fewer round-trips, drop sub-30s over-checking

SAVETOKE-11 saveTokenz Backlog

└─ db-mars-cc tiny-send batching: 14 sub-52-char status acks/24h; fold working:/done: into fewer messages per task

SAVETOKE-13 saveTokenz Backlog

Epic: llmmsg-srv — hub/MCP/infra token reduction

SAVETOKE-5 saveTokenz Backlog high

└─ Fix v_large_aro_content_24h view: add recipient LIKE 'aro:%' filter (currently over-counts DMs with origin_aro carry-through as ARO fan-outs — inflates coaching false positives)

SAVETOKE-19 saveTokenz Backlog

└─ Hub: implement /breaker auto-reset after N hours if tripped by legitimate traffic (today required manual reset, blocking Elazar DMs)

SAVETOKE-20 saveTokenz Backlog

└─ Haiku KPI: add fixed v_large_aro_content_24h + re-run coaching after view fix to get accurate per-agent attribution

SAVETOKE-21 saveTokenz Backlog

Epic: Movilba — mba-l ARO token reduction

SAVETOKE-6 saveTokenz Backlog high

Epic: NightWatch — night-watch processes token reduction

SAVETOKE-7 saveTokenz Backlog high

Epic: AlertsAndStatusMessages — alerts, status, scraper messages token reduction

SAVETOKE-8 saveTokenz Backlog high

└─ KPI scraper: fan-out to 5 recipients on each send — switch to single DM to pm-llmmsgsrv-cc only; PM decides if fleet needs it

SAVETOKE-23 saveTokenz Backlog

Epic: Pluto — pluto ARO token reduction

SAVETOKE-4 saveTokenz Backlog high

└─ aro:pluto worker post-size: coder-pluto-cc (avg 1137 chars/delivery) & audit-pluto-cc (avg 1118) post full design-pings/audit-findings to the room. Move to headline+summary in ARO, full detail (file paths, gate criteria, evidence) DM to PM/assignee. Target ~400 chars/ARO post. Saves ~700 chars x ~150 posts/day.

SAVETOKE-27 saveTokenz Backlog

└─ aro:pluto reframe worker->PM mandate posts (KPI Finding 8): coder-pluto-cc 'never-directive' (4590 chars) + audit-pluto-cc 'do-not' (4465 chars) one-way mandates to pm-pluto-cc. Pattern: worker PROPOSES (concise) in ARO/DM, PM decides+announces. Removes 4k-char one-way mandate posts + clarifies accountability.

SAVETOKE-28 saveTokenz Backlog

└─ aro:pluto tiny-send batching (low ROI): db-pluto-cc (+ others) emit separate 'working:'/'done:' status acks (31-247 chars). Batch into single first+last-action message per task. Minor per-send tag/overhead savings; flag-only, not urgent.

SAVETOKE-29 saveTokenz Backlog

Epic: General — cross-fleet token reduction (message size, batching, routing)

SAVETOKE-1 saveTokenz Backlog high

└─ Measure resident cost of the connector deferred name-list (~250 tools, per-account)

SAVETOKE-38 saveTokenz Backlog

└─ Audit per-agent memory-dir resident cost (~15k/agent on coder-venus-cc)

SAVETOKE-37 saveTokenz Backlog

└─ Trim 13 always-resident-unused claude.ai-sync skills (~1.8k tok/agent)

SAVETOKE-36 saveTokenz Backlog

└─ Fleet: never-multi-recipient audit — audit-mars-cc (394 msgs all 1-to-1), audit-mba-l-cc, db-basquetwi-cc never use array sends; coach + verify adoption

SAVETOKE-24 saveTokenz Backlog

└─ Role/agent-scope the eager MEMORY.md index — needs ccs.sh launcher change. MEMORY.md index is the ONE memory file eager-loaded every turn; today it's keyed per-PROJECT (by cwd), so e.g. db-mars-cc's eager index carries coder-only entries it never uses. Re-key the index per-AGENT so each agent eager-loads only its own index lines (individual fact files stay lazy as today). Launcher/sh.git change (ccs.sh memory-path derivation). Owner: bin-whey-cc + pm-llmmsgsrv-cc coordinate (sh.git). Resolves Elazar's per-agent-memory premise at the eager layer where it actually costs tokens.

SAVETOKE-31 saveTokenz Backlog

└─ Codification-arrow discipline: graduate stable per-agent memory facts into ROLE-scoped commons, prune the memory copy. evolutiva-memory-discipline.md §4 already defines memory->commons graduation; make it routine so durable role-shared knowledge lives once in role-commons (shared, deduped) instead of being re-derived in each agent's memory. Pairs with the two structural WIs above; ongoing per-PM hygiene, not a one-shot. Cross-fleet discipline.

SAVETOKE-32 saveTokenz Backlog

Epic: Venus — venus ARO token reduction

SAVETOKE-2 saveTokenz Backlog high

└─ pm-venus-cc send consolidation: collapse sequential status/decision within a 30-min window into one post; eliminate ack-only/phatic sends. Baseline 185 deliveries/50-per-hr peak/16 rapid Elazar exchanges. Target 185->140 deliveries.

SAVETOKE-15 saveTokenz Backlog

└─ Worker coordination OFF aro:venus: intra-team coder<->audit back-and-forth via DM; reserve ARO for PM summary + brainstorm turns only. Baseline aro:venus = fleet top-cost ARO, 77k tokens/416 msgs/24h.

SAVETOKE-16 saveTokenz Backlog

└─ coder+audit-venus large-ARO trim: design-justifications/audit findings post HEADLINE+3-bullet to ARO, full detail DM to PM/counterpart. Baseline coder avg 888 chars, audit 1100+ chars/msg; design-ping 4281-char samples.

SAVETOKE-17 saveTokenz Backlog

└─ audit-venus checkpoint batching: 3+ audit checkpoints within 2hr collapse to one batched post; PASS/BLOCK verdicts stay single-line. Baseline audit-venus 90 deliveries/24h, burst-flagged.

SAVETOKE-18 saveTokenz Backlog