Architecture: Is the Training Center's 3-tier agent model (Cursor Agents /...

Documentation for Sessions

🏗️ Architecture: Is the Training Center's 3-tier agent model (Cursor Agents /...

Date: 2026-03-03
Time: 15:14:51 UTC
Session: gspace1-167.2
Agent: backend
Confidence: 95%
Status: ✅ Documented


📋 Decision Context

Question:

Is the Training Center's 3-tier agent model (Cursor Agents / Platform Intelligence / Space Agents) still correct after Session 166?

Task: Verify agent categorization correctness after S163-166 changes


🎯 Decision Made

Chosen:

NO — the tier model is stale. Frontend hardcodes 'cursor-architect', 'cursor-implementer', 'cursor-reviewer' IDs and 'Cursor Agents' tier name, but Session 166 commit 8bd3ce1c6 removed all cursor- prefixes. Backend has NO tier/category field. Correct model: Development Agents (8 MCP functional types, platform-agnostic), Vitto Sub-Agents (4 system agents in backend-vitto-api), Space Agents (10 user-facing types in backend-ai-api).

Options Considered:


💭 Reasoning

Three independent code analysis agents traced: (1) training-center.component.ts hardcodes cursor-architect/cursor-implementer/cursor-reviewer at lines 107-139 plus tier names at lines 335-355, (2) backend-ai-api platform-agent.service.ts seeds agents as module-generator/widget-generator/architecture-expert/code-quality-expert with NO cursor prefix, (3) MCP agent-router.ts has LEGACY_AGENT_ID_MAP converting old cursor-* to new {type}-agent format. Backend schemas (AgentRegistry, SubAgentDefinition, decisions) have NO tier/category field — platform is tracked separately. The 'Cursor Agents' category name is especially wrong because these agents now work across Cursor, Claude Code, Claude Desktop, etc.


📁 Files Involved


🔗 Related


Auto-generated by MCP Smart Consolidation