Architecture: Session 178: How should users communicate with Vitto during ...

Documentation for Sessions

🏗️ Architecture: Session 178: How should users communicate with Vitto during ...

Date: 2026-03-05
Time: 18:28:51 UTC
Session: gspace1-178.17
Agent: angular
Confidence: 93%
Status: ✅ Documented


📋 Decision Context

Question:

Session 178: How should users communicate with Vitto during Computer Use sessions that need human input (2FA, CAPTCHAs)?

Task: Not specified


🎯 Decision Made

Chosen:

Route Computer Use human-in-the-loop interactions through the existing Vitto chat panel instead of a separate reply input in the Computer Use panel. When Vitto encounters a blocker (2FA, CAPTCHA, ambiguity), it should post a message in the active chat session asking the user. User replies in chat, Vitto receives it and continues the Computer Use session. This creates a natural conversational flow: 'I need your 2FA code' → user types '83' → Vitto types it into Chrome.

Options Considered:


💭 Reasoning

Current UX has a reply input buried in the Computer Use panel that only appears when the session result contains a question mark or supervision is escalated. This is discoverable and unnatural. Users already have a mental model of chatting with Vitto. The chat panel is always visible. The architectural connection: ComputerUseLoopService emits a 'needs_input' SSE event → VittoChatService picks it up and posts an assistant message → user replies in chat → reply is routed back to the Computer Use session via the existing replyToSession API. No new infrastructure needed, just wiring existing systems together.


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation