feat(agent): inject top 3 HIGH/CRITICAL research notes into system prompt (400 char hard cap, zero context drain)

research: 1h cadence (3600s), expand domain pool to 10 incl. FREE_CHOICE, human bio/psych, evolutionary bio

hotfix: Dynamically inject persona into research and planning prompts to un-hardcode JARVIS

hotfix: Route report and propose tasks to RESEARCH_ENGINE key to prevent draining Chat/Cloud keys

hotfix: Increase continuous research sleep to 1 hour and expand topics to 125+

hotfix: Rewrite inject_key to round-robin tier assignment to perfectly distribute partitioned 7-key systems

hotfix: Automatically partition NVIDIA keys into PC (1-7) and Cloud (8-14) to prevent rate limit collisions

hotfix: Implement robust JSON extraction to parse fallback LLMs conversational wrapper text

hotfix: Dynamically generate failover chain based on model tier

hotfix: Implement robust 2D auto-failover (models + keys) in call_nvidia_model

hotfix: Fix TokenManager probe to check each exhausted key independently instead of a hardcoded general key

hotfix: Enforce 30s timeout on NVIDIA API calls so the model fails over instantly instead of hanging for 5 minutes

hotfix: Harden ResearchEngine verification gate to stop theoretical physics from queuing as software features

hotfix: Enable graceful fallback to pure Kokoro when rvc_python is missing on cloud deployments

hotfix: Restore TTS kwargs architecture to omega_event_bus backend/voice/engines/kokoro_engine.py

hotfix: Restore TTS kwargs architecture to omega_event_bus backend/events/omega_event_bus.py

hotfix: Fix omega_events DB schema and TTS kwarg mismatch backend/events/omega_event_bus.py

hotfix: Fix omega_events DB schema and TTS kwarg mismatch backend/db/migrations.py

hotfix: Emit UI live_key_status during normal chat stream routing

hotfix: Sync cognitive router 429 logic with EXHAUSTED_GOOGLE_KEYS memory leak backend/services/connectors/__init__.py

hotfix: Sync cognitive router 429 logic with EXHAUSTED_GOOGLE_KEYS memory leak backend/services/token_manager.py

hotfix: Add missing gradio_client for SF3D fallback to backend/requirements.txt

hotfix: Add missing gradio_client for SF3D fallback to requirements.txt

Force hard rebuild via Dockerfile push

hotfix: Fix ws_manager import from agent_ws in backend/family_device/remote_actions.py

hotfix: Remove missing config import and use os.environ in backend/family_device/enrollment.py

hotfix: Fix verify_token import and add qrcode dependency for backend/routers/family_device_routes.py

hotfix: Fix verify_token import and add qrcode dependency for backend/requirements.txt

hotfix: Fix verify_token import and add qrcode dependency for requirements.txt

sync: Updated new feature checklist with MDM features

sync: MDM feature additions for backend/main.py

sync: MDM feature additions for backend/routers/family_device_routes.py

sync: MDM feature additions for backend/family_device/remote_actions.py

sync: MDM feature additions for backend/family_device/enrollment.py

sync: MDM feature additions for backend/family_device/__init__.py

fix: clean connector chain — Gemini primary, NVIDIA final fallback, Ollama skipped on cloud, others auto-skipped if no key

feat: add NVIDIA Vault (15 keys) as final fallback tier in stream_with_fallback connector chain

fix: add completion log statement so user can visually verify successful research processing

fix: force variety in autonomous research topics by randomly selecting scientific domains

tune: reduce deep research sleep interval from 2 hours to 10 minutes for much faster learning

fix: specify diverse NVIDIA models in secondary verification gate prompt

feat: add strict secondary LLM validation gate before queuing feature proposals to ensure they are actually codable in Python

fix: increase char limit for dynamic topics since fallback NVIDIA model is sometimes chatty

fix: restrict LLM proposal prompt so JARVIS only proposes implementable code features, not abstract knowledge

fix: remove stray topic_index reference

fix: remove invalid episodic_memory import

fix: add omega_events table to central db migrations

feat: upgrade background research loop to dynamically generate vast unbounded internet topics via LLM instead of fixed list

fix: update hardcoded 2025 year in research queries to use dynamic current year

fix: add omega_events table creation to SQLite migrations