feat: ship llm-only turn understanding and memory governance upgrades

Add LLM memory quality eval gate and promote OpenAI-agent readiness updates

feat: add fault chaos permutations and operator UX contracts

feat: complete wave 113 no-hardware reliability expansion

refactor: close waves 108-112 runtime and reliability hardening

Wave 38: proactive dedupe hardening and runtime decomposition

Wave 37: add multimodal grounding, context-aware nudges, and repeatable soak cycles

Wave 36: add proactive triage, preference learning, and operator guidance

Extract services runtime bind and expansion-state bootstrap

Wave 5: add Jarvis persona posture, tone controls, and sim validation suite

Implement provider-backed integration, automation pipeline, and autonomy loop

Add release-readiness tooling and persistent expansion state

Complete remaining Jarvis roadmap: proactive, trust, planner, integrations, and deployment tracks

Add dead-letter queue and replay controls for outbound delivery failures

Add scheduled CI fault profile matrix and local profile runner

Add control presets, runtime profiles, signed operator actions, and observability analytics

Implement VX04 per-user voice preference profiles

Implement VX05 STT confidence diagnostics

Add scoped memory classes with retrieval policy (MX01)

Add unified Jarvis scorecard to status and tools (Q20 EX06)

Add readable audit explainability fields (Q18 IX06)

Implement MX02 memory confidence and provenance trail (Q17)

Implement recovery journal lifecycle tracking and restart reconciliation

Add per-integration circuit breakers with status reporting

Add adaptive end-of-turn timing diagnostics and status fields

Add conversational turn choreography cues across phases

Add intent-level success metrics to observability and status

Implement global safe-mode toggle with runtime persistence

Add first-class memory correction commands for voice flows

Add nudge policy and quiet-window deferral for due reminders

Add turn-phase timeout budgets and expose them in system status

Implement risky-action ambiguity detection and preview-token execution gate

Require inbound webhook token and sanitize operator/inbound payload logs

Redact inbound webhook auth headers and harden SSE writes

Improve operator SSE stream and skills tool gating behavior

Harden audit retention, skills state persistence, and operator/config safeguards

Add operator console, observability, skills, and resilience/security hardening

Add voice attention controller and inbound webhook event tooling

Add identity trust controls with per-user policy and approval flow

Add memory PII guardrails and email policy tightening

Add strict-policy email send and summary integration

Add retention windows and startup credential safety warnings

Add Slack and Discord notification hooks

Add weather/webhook/media integrations with health probes

Add HA todo/timer/area tools plus reminders and calendar helpers

Add Home Assistant capability discovery tool

Add system_status contract tool and schema versioning

Harden HA conversation policy/taxonomy and persist timers

Add HA conversation and timer tools; refresh 80+ item roadmap

Avoid leaking exception details in external tool errors