🏗️ Architecture: How should Vitto handle missing context when generation is r...

Date: 2026-03-11
Time: 18:40:40 UTC
Session: gspace1-204.12
Agent: angular
Confidence: 95%
Status: ✅ Documented

📋 Decision Context

Question:

How should Vitto handle missing context when generation is requested?

Task: Not specified

🎯 Decision Made

Chosen:

Return hard error with explicit tool blocklist telling the LLM to ask the user. Not a soft instruction, not needs_context status — a direct error that says DO NOT call other tools, just ask.

Options Considered:

1. Soft instruction (failed — LLM ignored)
2. Hard error without blocklist (risky)
3. Hard error with explicit tool blocklist (chosen, works)

💭 Reasoning

Three approaches tested in this session: (1) Soft instruction with needs_context status — LLM ignored it and went on a 5-tool investigation spree (query_platform, recall_memory, etc.). (2) Hard error without blocklist — LLM might still try to resolve. (3) Hard error with explicit 'do NOT call query_platform, recall_memory, search_decisions, http_request' — leaves zero ambiguity. The LLM's natural tendency is to be helpful by trying to resolve context itself. Only an explicit blocklist overrides this tendency. This is a general pattern for any tool that needs user input — the error must forbid alternative resolution paths.

📁 Files Involved

• APIs/backend-vitto-api/src/app/tools/tool-registry.service.ts

🔗 Related

• Decision ID: dev_1773254439985_66otcdf
• Session: 204
• Workspace: gspace1

Auto-generated by MCP Smart Consolidation