Architecture: How to close the Vitto learning loop and bridge the 'Built B...

Documentation for Sessions

🏗️ Architecture: How to close the Vitto learning loop and bridge the 'Built B...

Date: 2026-02-26
Time: 15:39:03 UTC
Session: gspace1-144.2
Agent: angular
Confidence: 90%
Status: ✅ Documented


📋 Decision Context

Question:

How to close the Vitto learning loop and bridge the 'Built But Not Wired' gap?

Task: Implement Vitto Super-Agent Gaps 1-8


🎯 Decision Made

Chosen:

8-gap implementation: (1) Quality gate + lessons for unified agent loop, (2) simulateAction before EXECUTE, (3) createReflection in resumeVerifyPhase, (4) Intelligence/outcome MCP tools in CLAUDE.md/AGENTS.md, (5) OpenAI+Gemini streaming, (6) Workflow builder with canvas+step editor, (7) 10 pre-built connector templates, (8) Voice feature flag + toolbar mic button

Options Considered:

  1. Wire only the learning loop (Gaps 1-3)
  2. Wire everything including UI gaps (Gaps 1-8)
  3. Full rewrite of the agent pipeline

💭 Reasoning

The Vitto super-agent had 90% of the infrastructure built but key wiring was missing. The outcome chain existed (MindService.recordExecution → OutcomeFeedbackService.sendOutcome → POST /ai-api/learning/execution-outcomes) but: (a) unified agent loop sent empty lessonsLearned, (b) resumeVerifyPhase skipped reflection, (c) simulateAction was never called before EXECUTE, (d) intelligence MCP tools had 0% usage. Fixed wiring first (highest ROI), then filled content gaps (connectors, streaming, workflow UI, voice).


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation