Architecture: How should VM execution, adaptive trust, and proactive intel...

Documentation for Sessions

🏗️ Architecture: How should VM execution, adaptive trust, and proactive intel...

Date: 2026-03-20
Time: 17:39:55 UTC
Session: gspace1-240.2
Agent: backend
Confidence: 93%
Status: ✅ Documented

📋 Decision Context

Question:

How should VM execution, adaptive trust, and proactive intelligence be unified into Vitto's architecture?

Task: S240: Unified Autonomy Engine for Vitto VM functionality

🎯 Decision Made

Chosen:

Unified Autonomy Engine — three independent services (TrustScoringService, TierResolverService, ActionDecider) feeding a single read-only AutonomyGate. Not a god object; each concern computes independently, gate assembles a read-only snapshot per task.

Options Considered:

  1. Incremental tier wiring (VM first, intelligence last, 6-8 sessions)
  2. Intelligence-first top-down (trust/triggers before VM, high risk)
  3. Unified Autonomy Engine with parallel tracks (VM + trust + proactivity in 3 paired sessions)

💭 Reasoning

Three approaches evaluated: (A) Incremental bottom-up (VM first, intelligence last) — low risk but 6-8 sessions to feel human. (B) Intelligence-first (trust/triggers first, VM later) — users see smart behavior but can't actually act. (C) Unified Autonomy Engine — parallel tracks where execution, trust, and proactivity share an AutonomyContext. Chose C because the core problem is that execution/trust/proactivity are disconnected. A human colleague naturally combines capability + judgment + anticipation. Single AutonomyContext assembled at gate avoids god-object coupling while enabling compound behavior. S237 decisions on context-aware tool filtering (0.94 confidence) and Jarvis gap analysis (0.96) directly inform this: tool selection + trust + tier must be one decision point.

📁 Files Involved

🔗 Related

Auto-generated by MCP Smart Consolidation