Architecture: How should Vitto's VSCode extension achieve Claude Code-leve...

Documentation for Sessions

🏗️ Architecture: How should Vitto's VSCode extension achieve Claude Code-leve...

Date: 2026-03-22
Time: 20:59:41 UTC
Session: gspace1-240.2
Agent: angular
Confidence: 92%
Status: ✅ Documented

📋 Decision Context

Question:

How should Vitto's VSCode extension achieve Claude Code-level IDE integration — native agent mode architecture?

Task: S240 handoff: Vitto Native Agent Mode spec

🎯 Decision Made

Chosen:

Full native Agent Mode with SSE streaming chat, IDE tool executors (file/terminal/editor/git), and Vitto backend as the AI brain. Extension host executes tool calls locally via VSCode API. Chat UI in a new Lit view within the MCP tab. Leverages existing WebSocket service, executor registry, and message bridge.

Options Considered:

  1. Minimal — reuse Claude Code executor only
  2. Lightweight — file read/write + terminal in message protocol
  3. Full native — SSE streaming chat + IDE tool executors (chosen)

💭 Reasoning

Three options evaluated. (A) Minimal (reuse Claude Code executor) — works today but requires Claude Code installed separately and provides no differentiation. (B) Lightweight (file read/write only) — easy but can't run commands or show diffs, which are the core value. (C) Full native — significant effort (~5-6 sessions) but achieves feature parity with Claude Code while leveraging Vitto's unique capabilities (proactive intelligence, trust scoring, memory, sub-agent orchestration). The extension already has 95% of the infrastructure: WebSocket (780 lines), executor registry with plugin architecture, 50+ message types, CodeLens/CodeActions providers, 15 commands with keybindings. The missing 5% is IDE tool handlers + streaming chat UI. Chose C because the foundation is so strong that the incremental cost is low relative to the differentiation it creates.

📁 Files Involved

🔗 Related

Auto-generated by MCP Smart Consolidation