Architecture: How should the scaffold fallback generator produce code when...

Documentation for Sessions

🏗️ Architecture: How should the scaffold fallback generator produce code when...

Date: 2026-03-10
Time: 18:57:22 UTC
Session: gspace1-199.2
Agent: angular
Confidence: 95%
Status: ✅ Documented

📋 Decision Context

Question:

How should the scaffold fallback generator produce code when the LLM times out or returns malformed output?

Task: S199: Fix module code generation quality

🎯 Decision Made

Chosen:

Align scaffold templates with implement.prompt.md golden rules: models extend BaseEntity, services extend BaseCrudApiService, components use useCrudList/useCrudDetail/useCrudForm composables, all routes include authGuard, domain events generated per entity, .vitto/instructions.json generated per module.

Options Considered:

  1. Fix only the LLM prompt (already correct, doesn't help when LLM times out)
  2. Add more retries to LLM calls (wastes tokens, doesn't fix quality)
  3. Remove scaffold entirely and fail loudly (bad UX, users see nothing)
  4. Align scaffold with prompt rules (chosen — guaranteed quality floor)

💭 Reasoning

ROOT CAUSE: The scaffold generator (generateScaffoldFiles, ~430 lines) was written in Session 108 before the composable infrastructure existed. It produced code from a different era: manual signals, raw UniverseDataService, no BaseEntity, no authGuard, no events. When the LLM timed out (which happens ~20% of the time on multi-module generation), users got broken modules that violated every rule in implement.prompt.md. The fix is not to make the LLM more reliable (that's a separate problem) but to make the fallback indistinguishable from correct LLM output. The scaffold IS the safety net — it must be production-quality. Also added detail + form component generation (scaffold only generated list components, but routes referenced detail/form → guaranteed 404). Also added .vitto/instructions.json with editable areas and available actions — this is the foundation for contextual chat editing.

📁 Files Involved

🔗 Related

Auto-generated by MCP Smart Consolidation