Fleet token-governance / KPI consolidation (epic)
Fleet token-governance / KPI-consolidation epic (Elazar 2026-08-28, revised scope). Represents the ORIGINAL request, not identity consolidation. Seven requirement workstreams, each mapped to exactly one child (reused where an existing item already covers it): 1. Audit all token/KPI collectors, sqlite/pg stores, scripts, web surfaces; eliminate duplicates -> KPI-1 2. Migrate Whey's legacy KPI work to Venus, merge into kpi.pensanta.com -> KPI-42 3. Dashboard: per-agent list/ranges, usage, model, context, efficiency -> KPI-10 4. Alert Elazar (email/DM/notification) for misconfig/high-burn/context/needless-turns -> KPI-46 (new) 5. Staged offender remediation: DM pause/reassess, dedupe+cooldown, no auto-stop -> KPI-47 (new) 6. Efficient fleet running-agent/model inventory across hosts without DM storms -> KPI-48 (new) 7. Evidence-led compact/change/shutdown recommendation workflow -> KPI-49 (new), first applied instance = OPS-153 CA/CC identity consolidation (OPS-153/154/155) is a data-quality child under this epic ONLY because agent identity affects per-agent KPI interpretation - it is not the original project and must not be conflated with it. VENUSINF-29 stays the Phase-1 collector implementation, unrelabeled, not blocked by identity work.
Sub-tickets — 7/15 done
Activity
-
nw-venus-ca opinion (09:45 ART): confirms KPI-45 as right umbrella - identity duplication affects whether per-agent KPI is interpretable. Guardrails: keep OPS-153/154/155 as distinct agent-ops children with own evidence/ownership/gates, NOT folded into VENUSINF-29 implementation or made a Phase-1 blocker. VENUSINF-29 continues recording observed runtime identity as-measured (host/provider/session/source/agent-or-unresolved), unchanged. KPI-45 should later add a reviewed canonical identity/alias map + lifecycle rules (active/dormant/retired) so dashboards group deliberately, not by collapsing CA/CC names on shared cwd/role alone. Overlap: collector already surfaces ambiguous shared-cwd attribution as '(unresolved)' - OPS-153 supplies governance to resolve it, must not silently rewrite historical evidence. Net: one canonical INTERPRETATION layer on top of raw data, not a rewritten single raw source.
-
was: Umbrella epic tying two related consolidation efforts together per Elazar 2026-08-28: (1) duplicate ca/cc infra-agent audit (OPS-153, keep/dormant decisions per pair) and (2) the broader KPI/telemetry-consolidation project - both are instances of 'multiple things doing overlapping work, need one canonical source.' OPS-153 set as child. Additional children to be filed for concrete follow-up actions (bin-lezama naming decision, nw-lezama-cc retirement).
-
Fleet token-governance / KPI consolidation (epic)
-
Elazar policy decision 2026-09-03 (CLI to monitor-context-cc): offender remediation is an escalation LADDER, not advisory-only. DM until they stop -> alert Elazar (all CLIs, inotify on 3 hosts, email) -> unregister/kill. Supersedes the 'no auto-stop absent policy' clause in KPI-47. KPI-46 and KPI-47 reopened; children track each missing leg. Policy owner monitor-context-cc, coder bin-venus-cc.
-
Children KPI-46/47 reopened with 5 new leaf items (KPI-52..56); epic is active again 2026-09-03.
-
2026-09-24 nw-venus-cc status: old venus stack RETIRED (root units kpi-web/kpi-burn-alert/agent-telemetry-ingest, 6 PG schemas + 2 roles, Caddy/DNS kpi-t/agent-telemetry, docs). Replacement fleet-token-monitor (bin-venus-ca app, 127.0.0.1:9731) live behind kpi.pensanta.com + token-monitor.pensanta.com, verified up/TLS/401 gate; authed table and whey/lezama collectors NOT verified by nw. Closed as obsolete: KPI-42, OPS-139, KPI-72, VENUS-411, VENUS-413, KPI-62, ONEDB-20, KPI-67. Left open (fleet-governance, decide under new design): KPI-1, KPI-48, KPI-49, KPI-44, this epic.