🏗️ Architecture: How should training data flow from ai-api to sub-agent syste...

Date: 2026-03-12
Time: 19:24:36 UTC
Session: gspace1-209.4
Agent: backend
Confidence: 92%
Status: ✅ Documented

📋 Decision Context

Question:

How should training data flow from ai-api to sub-agent system prompts?

Task: Not specified

🎯 Decision Made

Chosen:

Add getTrainingExamplesForAgent() to BackendAiClientService, update buildSystemPrompt() to accept remote training examples with embedded fallback, fetch both learning context and training examples in parallel via Promise.all in spawnFromDefinition() and executeTask()

Options Considered:

• No alternatives documented

💭 Reasoning

ai-api is the source of truth for training data (TrainingDataService on ai-agents DB). Definition's embedded trainingExamples[] on vitto-mind DB are a duplicate. By fetching from ai-api first (2s timeout, graceful fallback to empty), we ensure single source of truth while maintaining offline resilience. The controller got a validatedOnly fast-path to avoid the heavy getAllExamples decision-mapping for this use case.

📁 Files Involved

• APIs/backend-ai-api/src/app/platform-ai/platform-ai.controller.ts
• APIs/backend-vitto-api/src/app/connectivity/backend-ai-client.service.ts
• APIs/backend-vitto-api/src/app/engine/sub-agent.service.ts

🔗 Related

• Decision ID: dev_1773343476622_yn8sniv
• Session: 209
• Workspace: gspace1

Auto-generated by MCP Smart Consolidation