OiMy Cloud Efficient Branch · May 26, 2026

Skill Engine Status & Phase 2B Handoff

Latest implementation snapshot for the cloud-native OiMy Skill Engine: router, onboarding, behavioral signals, QA results, remaining Phase 2B gaps, and developer verification steps.

✅ Internal engine mostly fixed ✅ Public /chat routed to patched engine ❌ Agenda + Honcho still open Branch: cloud-efficient

User inputTelegram / Web / WhatsApp

↓

Intent RouterGemma 4 31B via OpenRouter

↓

83 Skills6 workflow archetypes

↓

Behavior Layerbarriers + motivators

↓

Responsestructured envelope

Skills loaded

Router prompt covers the cloud skill library across action, coach, expert, planner, monitor, and profile patterns.

12/15

Internal QA checks passed

Onboarding, routing, name extraction, relationship coach, and major barrier detection passed internally.

Phase 2B+

Partially complete

Action OpenRouter migration is mostly done; Agenda Manager, Big Five, Honcho, and Hermes injection remain.

What is completed

OpenAI embeddings: text-embedding-3-small.
Full onboarding metadata envelope.
Name extraction adversarial fixes.
Cloud OpenRouter/Gemma intent routing.
Action extraction no longer depends first on local Ollama.
Coach tall-model path wired to OpenRouter.
Deep-profile text injected into coach prompts when present.
Barrier detection fixed for forgetfulness, guilt/shame, social pressure, overwhelm.

What is still open

Public route: api.oimyai.com now tunnels to the patched engine and returns the full envelope.
Agenda Manager: schema exists, not wired into api.py flow.
Big Five OCEAN: schema exists, signal→trait mapping not wired.
Honcho: enabled; QA shows insight_used after turns.
Hermes profile injection: not implemented.
Remaining polish: agenda completion matching is basic and needs a better classifier.
Skill gaps: mental-load schema, meal-planning missing-field UX.

Architecture completion estimate

Core cloud routing / skill execution:

Behavioral intelligence:

Deep profile + memory:

Phase 2B readiness:

These are implementation-readiness estimates from the latest QA, not user-volume production guarantees.

Developer verification

ssh root@116.203.107.13
cd /opt/oimy-engine
python3 -m py_compile api.py intent_router.py archetypes/profile.py archetypes/action.py archetypes/coach.py archetypes/signal_detector.py
curl -s http://localhost:8080/health
python3 /tmp/oimy_qa_rerun.py

Public /chat should return the structured envelope. If it returns only status/message/next_step/prompt, the public backend is stale or routed incorrectly.

Phase 2B next steps

Fix public API routing mismatch
Ensure Cloudflare/api.oimyai.com hits the patched engine behind nginx.

Fix remaining 500s
Resolve duration and iterable-join regressions found by QA.

Wire Agenda Manager
Previous commitments, completion detection, and new action extraction.

Add Big Five continuous inference
Map conversation signals to OCEAN EMA updates.

Enable Honcho + Hermes profile injection
Persist deep profiles into each user’s Hermes context.