🏗️ Architecture: Session 128: MCP tools audit — what was the finding and what...

Date: 2026-02-20
Time: 17:38:23 UTC
Session: gspace1-128.10
Agent: architecture
Confidence: 100%
Status: ✅ Documented

📋 Decision Context

Question:

Session 128: MCP tools audit — what was the finding and what needs to happen?

Task: Not specified

🎯 Decision Made

Chosen:

Only 5.8% of 104 sub-actions are used, reinforcement learning loop is broken

Options Considered:

1. Most tools are used effectively
2. Only 5.8% of 104 sub-actions are used, reinforcement learning loop is broken

💭 Reasoning

Audit of 17 MCP tools (104 sub-actions) found only 6 routinely used: preflight, log_decision, find_similar, convention.get, session.set, stats.basic. The intelligence tool (25 sub-actions: recall, simulate, reflect, diagnose, heal, calibrate, code DNA, time machine, collective) has 0% usage. The outcome tool (report, track, module) is never called, breaking the reinforcement learning loop — decisions are logged but never evaluated. The predecessor plan (supercharge-ai-mcp.plan.md, Session 80) correctly diagnosed this 3 weeks ago but was never executed. New plan: enhance-and-wire-mcp.plan.md focuses on wiring existing tools, not building new ones. Phase 1 (close learning loop) is the critical foundation.

📁 Files Involved

• No files specified

🔗 Related

• Decision ID: dev_1771609103213_ib61xet
• Session: 128
• Workspace: gspace1

Auto-generated by MCP Smart Consolidation