intelligence-engine
🤖 Generated by Smart Project Inventory MCP Server 📅 Last Updated: 2026-02-22 18:19:02 🎯 Analysis Mode: UPDATED 📊 Version: 0.14.1
MANDATORY: SPINE Initialization — READ THIS FIRST
This project is SPINE-orchestrated. Before doing ANY work, read and follow:
Full Protocol: _protocols\SPINE-INIT.md
WSL: _protocols/SPINE-INIT.md
Key Rules
- SPINE is READ-ONLY — never create/modify/delete files in ``
- Classify every task by tier (1/2/3) before starting
- State tier out loud before working: “SPINE Classification: Tier [1/2/3]”
- Use subagents for Tier 2/3 tasks (Explore, Plan, code-architect, etc.)
- Use MCP servers proactively — the right question is “would this tool add value?” not “do I need it?”
Tool Usage Expectations
Do not default to minimal tool usage. Proactively use available MCP servers and Claude Code native capabilities throughout every session. At session start, identify all available MCP servers (via .mcp.json and ToolSearch) and native capabilities (Task agents, Plan mode, parallel execution).
MANDATORY: SPINE Tiered Usage Protocol
This project uses SPINE for orchestration. You MUST follow the tiered enforcement protocol.
Quick Reference
| Tier | When | Requirements |
|---|---|---|
| Tier 1 | Single-file, simple tasks | Direct implementation OK |
| Tier 2 | Multi-file, new features | SHOULD use subagents, document if skipped |
| Tier 3 | Architecture, research, UI-heavy | MUST use subagents and MCP |
Full Protocol Location
_protocols\tiered-spine-usage.md
Or WSL: _protocols/tiered-spine-usage.md
Available Capabilities
Native Subagents (Task tool):
| Subagent | Model | Purpose |
|---|---|---|
Explore |
Fast | Codebase exploration, file discovery |
Plan |
Opus | Architecture planning, implementation design |
code-architect |
Opus | System design, architectural decisions |
research-coordinator |
Opus | Multi-source research |
context-engineer |
Opus | Context stack design |
visual-tester |
Haiku | UI verification via browser |
Model Strategy: Claude Opus 4.6 for all substantive work.
MCP Servers (23 available — full list, self-contained):
| Server | Use For |
|---|---|
| Research & Analysis | |
research-agent-mcp |
Multi-stage research workflow |
research-notes-mcp |
Parse, cluster, structure research notes |
research-log-mcp |
Research logging with citations |
evaluation-mcp |
Rubric-based scoring, ranking, verdicts |
| Project Management | |
session-handover-mcp |
Session continuity and handover docs |
next-conductor |
NEXT.md task tracking |
paaf |
Project audits, doc health, technical debt |
mcp-builder-mcp |
Pattern analysis, MCP generation (testing) |
| Documentation & Content | |
manual-mcp |
Multi-part documentation manuals |
showcase-mcp |
GitHub Pages showcase generation |
content-mcp |
Semantic content orchestration (video/image/audio/text) |
content-analyzer-mcp |
Content analysis, extraction, classification |
smart-inventory |
CLAUDE.md generation |
| Infrastructure & Testing | |
browser-mcp |
Browser automation, screenshots, click memory |
content-extractor-mcp |
Screenshot analysis, OCR, UI region detection |
mcp-server-checker |
MCP server validation and quality |
8do-mcp |
Ralph loop task orchestration with verdict gates |
8me-mcp |
Tier 3 workflow management |
gen-loop-mcp |
Async task scheduling, loop execution |
backup-mcp |
Project backup and restore |
| Context & Memory | |
context-glue-mcp |
Context assembly, scenario composition, vocabulary |
minna-memory |
Persistent entity memory across sessions (recommended) |
| Agent Communication | |
agent-comm-mcp |
Inter-agent messaging, registry, relay |
Minna Memory (mem-system-lite-mcp):
- Per-project database at
.spine/minna.db - Features: Cross-session memory, error pattern storage, context enrichment
- Three MCP servers have built-in Minna adapters: session-handover-mcp, content-mcp, 8do-mcp
Before Completing Any Task
- Classify the task tier (1/2/3)
- If Tier 2: Use recommended capabilities or document why skipped
- If Tier 3: Use required capabilities, document usage
Documentation Template (Tier 2/3)
SPINE Usage (Tier [2/3]):
- Subagents used:
- [subagent_type]: [purpose]
- MCP servers used:
- [server]: [purpose]
- Screenshots: [location if applicable]
- Skipped (with reason): [capability]: [reason]
📋 Project Overview
Type: Code Intelligence Tool Primary Language: Python 3.12 + TypeScript Framework: FastAPI (backend), React 18 (frontend), FastMCP (MCP server) Build System: Hatchling (Python), Vite 6 (frontend) Package Manager: uv (Python), npm (frontend)
📊 Project Statistics
- Type: Code Intelligence Engine (AST-driven knowledge graphs + hybrid search + MCP server + web UI)
- Version: 0.14.1
- MCP Tools: 12
- REST Endpoints: 23
- Tests: 734
- Languages Parsed: Python, JavaScript, TypeScript, TSX, Java, Go
- Last Analysis: 2026-02-25
🏗️ Architecture Overview
Code Intelligence Engine — AST-driven knowledge graphs + hybrid search + MCP server + web UI.
See ai-memory/shared/architecture.md for full architecture diagram.
See KB/CLAUDE.md for main instruction document.
Core Pipeline: Tree-sitter → Knowledge Graph → Hybrid Search → MCP Server / Web UI
Components:
src/parser/— Tree-sitter AST parsing, 6 languages (Python, JS, TS/TSX, Java, Go)src/graph/— Knowledge graph (NetworkX + KuzuDB dual-backend)src/search/— BM25 + Semantic (LanceDB) + Graph search with RRF fusionsrc/mcp/— FastMCP server (12 tools for AI assistants)src/registry.py— Multi-project managementsrc/web/— FastAPI backend (23 REST endpoints) + React frontend (Sigma.js graph explorer)src/change_detector.py— Incremental indexing (git diff + hash fallback)src/indexer.py— Shared pipeline with mode param (auto/full/incremental)src/llm/— LLM provider abstraction (Claude, OpenAI, Gemini, Ollama) + entity summarizersrc/quality.py— Code quality metrics (complexity, doc coverage, coupling, scores)src/graph/clustering.py— Community detection (Louvain + label propagation)src/index_history.py— Append-only performance history per project
Status: All phases complete (1-12 + Batch Summaries + Code Quality Metrics + UI Polish), 12 MCP tools, 23 REST endpoints, 734 tests.
🚀 Development Setup
cd intelligence-engine
python3 -m venv .venv
source .venv/bin/activate
# Use uv for all package management
uv pip install tree-sitter tree-sitter-python kuzu lancedb sentence-transformers rank-bm25 fastmcp pyyaml
See KB/REF-technology-stack.md for complete dependency list.
📚 API Documentation
Add your API documentation here. This section is preserved during updates.
🧪 Testing
Add your testing information here. This section is preserved during updates.
🚢 Deployment
Add your deployment notes here. This section is preserved during updates.
📝 Team Notes
Project Identity:
- Name: Intelligence Engine
- Reference:
[PROJ-EBQKC-TNHEV] - Path (Win):
intelligence-engine - Path (WSL):
intelligence-engine
Key Constraints:
- Personal-use tooling — NOT enterprise
- Fail loud, never silent
- No backward compatibility
- MVP-first, each phase building on the last
- Reuse existing project code where possible
- Local only — no data leaves the machine
- Python 3.11+, use
uvfor packages
KB Reference Documents:
KB/CLAUDE.md— Main instructionKB/REF-gitnexus-architecture.md— Architecture deep-diveKB/REF-existing-projects-audit.md— 49+ project auditKB/REF-implementation-phases.md— Step-by-step phasesKB/REF-mcp-integration-guide.md— MCP integration patternsKB/REF-technology-stack.md— Technology choices
CRITICAL: Do NOT re-run analyze_project_smart without explicit instruction — it overwrites this file!
🔄 Recent Activity
- All 12 phases complete (AST Parsing through Graph Clustering + Batch Summaries + Code Quality Metrics + UI Polish)
- Project Type: Code Intelligence Engine
- Primary Technology: Python 3.12 + TypeScript (FastAPI + React 18 + FastMCP)
This documentation is automatically maintained by the Smart Project Inventory MCP Server.
Custom sections (between CUSTOM-START and CUSTOM-END markers) are preserved during updates.
Auto-updated sections are refreshed with each analysis.