diff --git a/docs/CLAUDE_INTEGRATION.md b/docs/CLAUDE_INTEGRATION.md
index 8f6d69c..7abef40 100644
--- a/docs/CLAUDE_INTEGRATION.md
+++ b/docs/CLAUDE_INTEGRATION.md
@@ -3,6 +3,7 @@
 ## The Problem
 
 Copilot instructions get **ignored in longer conversations** (15+ messages) and Claude falls back to:
+
 - Reading files directly
 - Using grep patterns
 - Manual code analysis
@@ -17,6 +18,7 @@ Copilot instructions get **ignored in longer conversations** (15+ messages) and
 ## The Solution: System Prompt Engineering
 
 Make the system prompt **enforce MCP at protocol level**:
+
 - File reads become impossible (system block)
 - Grep becomes forbidden (protocol-level)
 - MCP becomes mandatory (only option)
@@ -36,7 +38,7 @@ Edit `~/.claude_desktop_config.json`:
   "mcpServers": {
     "lxdig": {
       "command": "node",
-      "args": ["/home/alex_rod/code-graph-server/dist/server.js"],
+      "args": ["/home/alex_rod/projects/lxDIG-MCP/dist/server.js"],
       "env": {
         "MCP_TRANSPORT": "stdio",
         "MEMGRAPH_HOST": "localhost",
@@ -53,19 +55,21 @@ Edit `~/.claude_desktop_config.json`:
 ### Step 2: Update VS Code Settings
 
 Create `.vscode/mcp.json`:
+
 ```json
 {
   "servers": {
     "lxdig": {
       "type": "stdio",
       "command": "node",
-      "args": ["/home/alex_rod/code-graph-server/dist/server.js"]
+      "args": ["/home/alex_rod/projects/lxDIG-MCP/dist/server.js"]
     }
   }
 }
 ```
 
 Create `.vscode/settings.json`:
+
 ```json
 {
   "claude.alwaysUseMCP": true,
@@ -82,6 +86,7 @@ Close and reopen Claude completely.
 Ask Claude: "How does src/main.ts work?"
 
 **Expected**:
+
 - Claude calls `graph_set_workspace`
 - Claude calls `code_explain('main')`
 - NO file reads
@@ -90,12 +95,12 @@ Ask Claude: "How does src/main.ts work?"
 
 ## Why This Works
 
-| Before | After |
-|--------|-------|
-| Instructions fade in long chats | System prompt is protocol-level |
-| Claude reads files anyway | File reads are system-blocked |
-| Uses grep by default | Grep is forbidden |
-| Context gets out of sync | Health checks re-anchor every 5 messages |
+| Before                          | After                                    |
+| ------------------------------- | ---------------------------------------- |
+| Instructions fade in long chats | System prompt is protocol-level          |
+| Claude reads files anyway       | File reads are system-blocked            |
+| Uses grep by default            | Grep is forbidden                        |
+| Context gets out of sync        | Health checks re-anchor every 5 messages |
 
 ---
 
@@ -146,16 +151,19 @@ System Prompt:
 ## Troubleshooting
 
 ### Claude still reads files
+
 - Check system prompt in Claude Desktop config
 - Verify "NEVER read files" is present
 - Restart Claude completely
 
 ### Long conversations break
+
 - Ensure `graph_health()` re-anchoring is in system prompt
 - Test with 50+ message conversation
 - If fails at message N, check if `graph_health()` was called at N-5
 
 ### MCP tools not available
+
 - Verify MCP server running: `curl http://localhost:9000/health`
 - Check Docker: `docker-compose ps`
 - Restart Claude
diff --git a/docs/GRAPH_STATE_QUICK_REF.txt b/docs/GRAPH_STATE_QUICK_REF.txt
deleted file mode 100644
index 1f4e6aa..0000000
--- a/docs/GRAPH_STATE_QUICK_REF.txt
+++ /dev/null
@@ -1,231 +0,0 @@
-╔════════════════════════════════════════════════════════════════════════════╗
-║                   GRAPH STATE ANALYSIS - QUICK REFERENCE                   ║
-╚════════════════════════════════════════════════════════════════════════════╝
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-QUESTION 1: MULTIPLE PROJECTS SETUP
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Answer: ONE project per session, MULTIPLE isolated sessions
-
-Architecture:
-┌─ Server Process (shared)
-│  ├─ ToolContext.index (shared, never cleared)
-│  ├─ Memgraph connection (shared)
-│  └─ ToolHandlers (contains engines)
-│
-└─ Per Session:
-   ├─ ProjectContext (workspace + projectId)
-   ├─ FileWatcher (monitoring workspace)
-   └─ SessionId header identification
-
-Session Usage (REQUIRED for multi-project):
-  POST /initialize → Returns mcp-session-id: "sess-a"
-  POST /tools/graph_set_workspace
-    Header: mcp-session-id: sess-a
-    Body: {projectId: "project-a"}
-
-Without session ID: All requests use defaultActiveProjectContext (global)
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-QUESTION 2: CONTEXT SWITCHING (graph_set_workspace)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-What Changes:
-  ✅ ProjectContext (workspace, sourceDir, projectId)
-  ✅ FileWatcher (restarted for new directory)
-
-What DOESN'T Change (CRITICAL):
-  ❌ GraphIndexManager (NOT cleared - still has old project's data)
-  ❌ ProgressEngine (NOT reset - still has old project's tasks)
-  ❌ ArchitectureEngine (NOT reset - still references old index)
-  ❌ TestEngine (NOT reset - still references old index)
-  ❌ EmbeddingEngine (NOT reset - still references old index)
-  ❌ HybridRetriever (NOT reset - still references old index)
-
-Implementation (src/tools/tool-handlers.ts:1543-1615):
-  1. Resolve new ProjectContext
-  2. setActiveProjectContext(nextContext) - updates session/default context
-  3. startActiveWatcher(nextContext) - restarts file watcher
-  4. Return success
-
-PROBLEM: In-memory index accumulates data from multiple projects!
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-QUESTION 3: GRAPH REBUILD BEHAVIOR
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Clear in-memory index first?
-  ❌ NO - Shared index is NOT cleared
-
-Append to existing index?
-  ❌ NO - Orchestrator creates NEW internal index for each build
-
-Load index from Memgraph?
-  ❌ NO - Parses files from scratch
-
-Actual Behavior (src/graph/orchestrator.ts:181-423):
-  1. GraphOrchestrator creates NEW GraphIndexManager() [internal]
-  2. Parses all source files
-  3. For each file: addToIndex(parsed) → adds to orchestrator's internal index
-  4. Generates Cypher statements
-  5. Executes Cypher batch → Memgraph UPDATED ✓
-  6. Returns BuildResult
-  7. Orchestrator.index is DISCARDED (not synced)
-  8. ToolContext.index UNCHANGED (remains empty) ✗
-
-Result:
-  • Memgraph database: UPDATED ✓
-  • Shared in-memory index: UNCHANGED (empty) ❌
-  • Orchestrator's internal index: WASTED (discarded) ❌
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-QUESTION 4: INDEX INITIALIZATION
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Where does GraphIndexManager get populated from?
-
-At Startup:
-  src/mcp-server.ts:618 → new GraphIndexManager()
-  Result: EMPTY ◯
-
-After graph_rebuild:
-  Orchestrator.build() populates ITS internal index
-  But NEVER syncs to ToolContext.index
-  Result: Still EMPTY ◯
-
-When queries run:
-  Tools query Memgraph directly (not index) → WORKS ✓
-  Embedding engine queries index (EMPTY) → FAILS ❌
-  Progress tracking reads index (EMPTY) → FAILS ❌
-
-Sources of Potential Population:
-  ✓ Manual .addNode() calls (rarely used)
-  ✗ Memgraph database (never loaded at startup)
-  ✗ File system (never loaded directly)
-  ✗ Orchestrator rebuild (not synced)
-
-Summary: Started empty, usually stays empty.
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-CORE ARCHITECTURE ISSUE
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-TWO SEPARATE, UNSYNCED INDEX SYSTEMS:
-
-  ┌─────────────────────────────────────┐
-  │ ToolContext.index (Shared)          │
-  │ • Initialized: EMPTY                │
-  │ • Updated: NEVER                    │
-  │ • Used by: ALL engines              │
-  │ • Status: Empty or stale            │
-  └─────────────────────────────────────┘
-
-  ┌─────────────────────────────────────┐
-  │ GraphOrchestrator.index (Internal)  │
-  │ • Created: During build()           │
-  │ • Populated: YES (during parsing)   │
-  │ • Synced: NO (never goes back)      │
-  │ • Status: Temporary, discarded      │
-  └─────────────────────────────────────┘
-
-  ┌─────────────────────────────────────┐
-  │ Memgraph Database (Source of Truth) │
-  │ • Updated: By Orchestrator Cypher   │
-  │ • Queried: YES (by tools)           │
-  │ • Status: Current and accurate      │
-  └─────────────────────────────────────┘
-
-Consequence:
-  • Tools using Memgraph directly: ✓ WORK
-  • Embedding engine: ❌ FAILS
-  • Progress tracking: ❌ FAILS
-  • Architecture validation: ❌ FAILS
-  • Multi-project support: ❌ RISKY
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-QUICK FIXES (PRIORITY ORDER)
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-1. CLEAR INDEX ON CONTEXT SWITCH (30 min)
-   File: src/tools/tool-handlers.ts:1543-1615 (graph_set_workspace)
-   Add:  if (oldProjectId !== newProjectId) this.context.index.clear();
-   Why:  Prevents data accumulation from multiple projects
-
-2. SYNC ORCHESTRATOR INDEX (2 hours)
-   Files: src/graph/orchestrator.ts:70-176 (add getIndex method)
-          src/tools/tool-handlers.ts:1617-1776 (sync in graph_rebuild)
-   Why:   Makes embedding and progress tracking work
-
-3. ADD PROJECTID FILTERS (1 hour)
-   Files: All Cypher queries in src/tools/tool-handlers.ts
-   Add:   WHERE n.projectId = $projectId
-   Why:   Ensures query results respect project boundaries
-
-4. PROJECT-SCOPED INDICES (1-2 days)
-   Files: Refactor entire index management system
-   Why:   Future-proof, complete isolation
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-KEY FILE LOCATIONS
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-ToolContext Definition
-  src/tools/tool-handlers.ts:41-46
-
-ProjectContext Definition
-  src/tools/tool-handlers.ts:48-52
-
-Context Switching
-  src/tools/tool-handlers.ts:87-106
-
-Session Management
-  src/tools/tool-handlers.ts:69-71
-
-graph_set_workspace Implementation
-  src/tools/tool-handlers.ts:1543-1615
-
-graph_rebuild Implementation
-  src/tools/tool-handlers.ts:1617-1776
-
-GraphOrchestrator Constructor
-  src/graph/orchestrator.ts:86-176
-
-GraphOrchestrator.build()
-  src/graph/orchestrator.ts:181-423
-
-GraphIndexManager Implementation
-  src/graph/index.ts:35-178
-
-ProgressEngine Construction
-  src/engines/progress-engine.ts:59-96
-
-MCP Server Initialization
-  src/mcp-server.ts:618-623
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-RISK ASSESSMENT
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-Single Project, Single Session:     LOW RISK ✓
-Multiple Projects with SessionIds:  MEDIUM RISK ⚠ (index contamination)
-Multiple Projects without Sessions: HIGH RISK ✗ (complete data mixing)
-
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-DOCUMENTATION FILES
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-
-GRAPH_STATE_ANALYSIS.md        ← Full technical analysis (12 sections)
-GRAPH_STATE_DIAGRAMS.md        ← Architecture and data flow diagrams (8 diagrams)
-GRAPH_STATE_FIXES.md           ← Implementation guide with code examples
-GRAPH_STATE_SUMMARY.md         ← Executive summary (this addresses your questions)
-GRAPH_STATE_QUICK_REF.txt      ← This file
-
diff --git a/docs/INTEGRATION_SUMMARY.md b/docs/INTEGRATION_SUMMARY.md
deleted file mode 100644
index fd4d9f6..0000000
--- a/docs/INTEGRATION_SUMMARY.md
+++ /dev/null
@@ -1,292 +0,0 @@
-# Integration Summary: MCP Server Documentation
-
-**All documentation consolidated into docs/ folder.**
-
----
-
-## 📚 Documentation Map
-
-```
-docs/
-├─ INTEGRATION_SUMMARY.md ........... This file
-├─ MCP_INTEGRATION_GUIDE.md ......... Complete integration guide
-├─ CLAUDE_INTEGRATION.md ........... Claude/Copilot system prompt solution
-├─ TOOL_PATTERNS.md ............... Grep → MCP replacement patterns
-├─ templates/copilot-instructions-template.md . Copy to .github/copilot-instructions.md
-├─ templates/skill-mcp-template.md .. Skill prompt template
-├─ templates/toolsets-template.jsonc . VS Code toolset template
-├─ QUICK_REFERENCE.md ............. All 39 tools reference
-├─ QUICK_START.md ................. Server deployment
-└─ ARCHITECTURE.md ................ Technical deep dive
-```
-
----
-
-## 🎯 Quick Navigation
-
-### I want to...
-
-**Make Claude use MCP in long conversations (the main problem)**
-→ Read: [docs/CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md)
-→ Then: [docs/templates/copilot-instructions-template.md](templates/copilot-instructions-template.md)
-
-**Integrate MCP into my projects**
-→ Read: [docs/MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md)
-
-**Replace grep with MCP tools**
-→ Read: [docs/TOOL_PATTERNS.md](TOOL_PATTERNS.md)
-
-**See all 39 tools**
-→ Read: [QUICK_REFERENCE.md](../QUICK_REFERENCE.md)
-
-**Deploy the server**
-→ Read: [QUICK_START.md](../QUICK_START.md)
-
----
-
-## 🔑 Key Insights
-
-### Problem: Long Conversation Instruction Drift
-
-After ~15 messages, Copilot ignores instructions and falls back to:
-
-- Reading files directly
-- Using grep patterns
-- Manual code analysis
-
-### Root Cause
-
-Instructions are overlaid suggestions. They fade in long conversations. File reads are baked into training data (default behavior).
-
-### Solution: System Prompt Engineering
-
-Make the system prompt **enforce MCP at protocol level**:
-
-- File reads become impossible (system block)
-- Grep becomes forbidden (protocol)
-- MCP becomes mandatory (only option)
-
-Result: Even at message 100, Claude uses MCP because it's the **only option**.
-
-### Implementation
-
-Edit `~/.claude_desktop_config.json`:
-
-```json
-{
-  "systemPrompt": "NEVER read files. NEVER use grep. ALWAYS use MCP tools..."
-}
-```
-
----
-
-## 📋 Setup Checklist
-
-### Infrastructure (One Time)
-
-- [ ] Docker + Docker Compose installed
-- [ ] `docker-compose up -d memgraph qdrant`
-- [ ] `npm run build && npm run start:http`
-- [ ] Verify: `curl http://localhost:9000/health`
-
-### Claude Desktop
-
-- [ ] Edit `~/.claude_desktop_config.json`
-- [ ] Add MCP server config
-- [ ] Add system prompt (enforces MCP)
-- [ ] Restart Claude
-
-### Per-Project
-
-- [ ] Copy [templates/copilot-instructions-template.md](templates/copilot-instructions-template.md) to `.github/copilot-instructions.md`
-- [ ] Update project references
-- [ ] Set `LXDIG_PROJECT_ID` in your environment or pass `projectId` to `graph_set_workspace`
-- [ ] Commit `.github/copilot-instructions.md`
-
----
-
-## 🚀 Implementation Order
-
-### Phase 1: Foundation (15 min)
-
-1. Start Docker services
-2. Build and start MCP server
-3. Update Claude Desktop config
-4. Restart Claude
-
-### Phase 2: Test (5 min)
-
-1. Ask Claude: "How does [file] work?"
-2. Verify it calls MCP tools (not file reads)
-3. Test long conversation (20+ messages)
-4. Verify no degradation
-
-### Phase 3: Rollout (Per-Project)
-
-1. Copy copilot instructions to `.github/copilot-instructions.md`
-2. Set `LXDIG_PROJECT_ID` or pass projectId to `graph_set_workspace`
-3. Commit and push
-4. Update team
-
----
-
-## 💡 Core Concepts
-
-### 39 MCP Tools Available
-
-```
-Essential 4:
-  • graph_query — Find code by natural language
-  • code_explain — Understand symbols with context
-  • impact_analyze — What breaks if I change X?
-  • test_select — Which tests should I run?
-
-Architecture 2:
-  • arch_validate — Check violations
-  • arch_suggest — Where should code go?
-
-Search 3:
-  • semantic_search — Search by meaning
-  • find_pattern — Detect violations
-  • find_similar_code — Find implementations
-
-Testing 3:
-  • test_categorize — Group tests
-  • suggest_tests — Tests needed for symbol
-  • test_run — Execute tests
-
-Memory 4:
-  • episode_add — Record decisions
-  • decision_query — Recall decisions
-  • reflect — Synthesize learnings
-  • [coordination tools]
-
-+ 18 more specialized tools
-```
-
-See [docs/TOOL_PATTERNS.md](TOOL_PATTERNS.md) for pattern matching.
-
-### Multi-Project Architecture
-
-```
-Claude → MCP Server → Memgraph + Qdrant
-                      ↓
-                    Project A (isolated)
-                    Project B (isolated)
-                    Project C (isolated)
-```
-
-Each project: isolated by `projectId` + `workspaceRoot`
-Shared: Memgraph + Qdrant infrastructure
-
-### Session Re-anchoring
-
-```
-Message 1:    graph_set_workspace() → session starts
-Message 1-4:  Normal MCP queries
-Message 5:    graph_health() → verify still ready
-Message 6-9:  Normal MCP queries
-Message 10:   graph_health() → re-anchor
-...continues indefinitely without degradation
-```
-
----
-
-## 📊 Performance Gains
-
-| Task                | Grep/Manual | MCP   | Improvement  |
-| ------------------- | ----------- | ----- | ------------ |
-| Find symbol         | 450ms       | 50ms  | 9x faster    |
-| Understand function | 5 min       | 200ms | 1500x faster |
-| Impact analysis     | 10 min      | 100ms | 6000x faster |
-| Search by meaning   | 2 min       | 150ms | 800x faster  |
-| False positives     | High        | <1%   | 100x better  |
-
----
-
-## ✅ Success Criteria
-
-After full implementation:
-
-- ✅ Claude uses MCP in **every** conversation
-- ✅ Long conversations (100+ messages) **never** degrade
-- ✅ Zero file reads across entire session
-- ✅ Zero grep patterns used
-- ✅ Full dependency context always available
-- ✅ All projects share MCP infrastructure
-- ✅ Heavy MCP dependency, zero fallback
-
----
-
-## 🔍 Before vs After
-
-### Before (Grep/File Reads)
-
-```
-User: "How does auth work?"
-Claude:
-  1. Opens src/auth/service.ts
-  2. Reads 200 lines
-  3. Opens 5 imported files
-  4. Manually traces dependencies
-  Result: Takes 1+ minutes, incomplete context
-
-User (message 20): "Refactor this"
-Claude:
-  1. Falls back to grep
-  2. Misses some usages
-  3. Suggests incomplete refactor
-  Result: Broken code, manual fixing needed
-```
-
-### After (MCP)
-
-```
-User: "How does auth work?"
-Claude:
-  1. graph_set_workspace()
-  2. code_explain('AuthService')
-  3. graph_query('show call graph')
-  Result: 200ms, complete dependency context, perfect
-
-User (message 20): "Refactor this"
-Claude:
-  1. graph_health() [re-anchor]
-  2. impact_analyze(['src/auth/service.ts'])
-  3. Suggests safe refactor with impact analysis
-  Result: Correct refactor, safe changes
-```
-
----
-
-## 📚 Related Files
-
-- **Root**: [QUICK_START.md](../QUICK_START.md), [QUICK_REFERENCE.md](../QUICK_REFERENCE.md), [ARCHITECTURE.md](../ARCHITECTURE.md)
-- **Docs**: [MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md), [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md), [TOOL_PATTERNS.md](TOOL_PATTERNS.md)
-- **Template**: [templates/copilot-instructions-template.md](templates/copilot-instructions-template.md)
-
----
-
-## 🎓 For Your Team
-
-1. **Tech Lead**: Read [docs/CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md) + [docs/MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md)
-2. **Developers**: Read [docs/templates/copilot-instructions-template.md](templates/copilot-instructions-template.md) + [docs/TOOL_PATTERNS.md](TOOL_PATTERNS.md)
-3. **New Team Members**: Follow setup checklist + read `.github/copilot-instructions.md`
-
----
-
-## 🆘 Troubleshooting
-
-| Issue                    | Solution                 | Doc                                            |
-| ------------------------ | ------------------------ | ---------------------------------------------- |
-| Claude reads files       | Update system prompt     | [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md) |
-| Long conversations break | Add graph_health() calls | [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md) |
-| Don't know which tool    | Check TOOL_PATTERNS      | [TOOL_PATTERNS.md](TOOL_PATTERNS.md)           |
-| Server won't start       | Check Docker             | [QUICK_START.md](../QUICK_START.md)            |
-| Need tool reference      | See all 39 tools         | [QUICK_REFERENCE.md](../QUICK_REFERENCE.md)    |
-
----
-
-## 🎯 One-Liner Summary
-
-**System prompt engineering (not instructions) solves Copilot instruction drift. Use MCP exclusively for code intelligence. Scale to all projects with shared infrastructure. 🚀**
diff --git a/docs/MCP_INTEGRATION_GUIDE.md b/docs/MCP_INTEGRATION_GUIDE.md
index 880648b..d346ae5 100644
--- a/docs/MCP_INTEGRATION_GUIDE.md
+++ b/docs/MCP_INTEGRATION_GUIDE.md
@@ -7,7 +7,7 @@ Complete guide for integrating lxDIG MCP across projects.
 ### 1. Start Infrastructure
 
 ```bash
-cd /home/alex_rod/code-graph-server
+cd /home/alex_rod/projects/lxDIG-MCP
 docker-compose up -d memgraph qdrant
 npm install && npm run build
 npm run start:http  # Listens on http://localhost:9000
@@ -22,7 +22,7 @@ Edit `~/.claude_desktop_config.json`:
   "mcpServers": {
     "lxdig": {
       "command": "node",
-      "args": ["/home/alex_rod/code-graph-server/dist/server.js"],
+      "args": ["/home/alex_rod/projects/lxDIG-MCP/dist/server.js"],
       "env": {
         "MCP_TRANSPORT": "stdio",
         "MEMGRAPH_HOST": "localhost",
@@ -44,7 +44,7 @@ Create `.vscode/mcp.json`:
     "lxdig": {
       "type": "stdio",
       "command": "node",
-      "args": ["/home/alex_rod/code-graph-server/dist/server.js"]
+      "args": ["/home/alex_rod/projects/lxDIG-MCP/dist/server.js"]
     }
   }
 }
@@ -81,7 +81,7 @@ For each project, add `.mcp-config.json`:
 }
 ```
 
-## 38 Tools Quick Reference
+## 39 MCP Tools Quick Reference
 
 ### Essential (Use First)
 
@@ -200,7 +200,7 @@ await mcp.initialize();
 await mcp.query("find all HTTP handlers");
 ```
 
-See docs/CLIENT_EXAMPLES.md for Python, bash, React.
+See [QUICK_REFERENCE.md](../QUICK_REFERENCE.md) for all 39 tools.
 
 ## Rollout Phases
 
@@ -232,6 +232,5 @@ See docs/CLIENT_EXAMPLES.md for Python, bash, React.
 - QUICK_START.md — Deployment details
 - QUICK_REFERENCE.md — All 39 tools
 - ARCHITECTURE.md — Technical deep dive
-- docs/CLIENT_EXAMPLES.md — Code snippets
 - docs/CLAUDE_INTEGRATION.md — System prompt details
 - docs/TOOL_PATTERNS.md — Before/after examples
diff --git a/docs/PROJECT_FEATURES_CAPABILITIES.md b/docs/PROJECT_FEATURES_CAPABILITIES.md
index 977a077..0901fd1 100644
--- a/docs/PROJECT_FEATURES_CAPABILITIES.md
+++ b/docs/PROJECT_FEATURES_CAPABILITIES.md
@@ -2,7 +2,7 @@
 
 ## Executive Summary
 
-lexDIG-MCP is an MCP server focused on **architecture-aware code intelligence** and **agent-ready task coordination**. It combines:
+lxDIG-MCP is an MCP server focused on **architecture-aware code intelligence** and **agent-ready task coordination**. It combines:
 
 - A graph plane for structural understanding.
 - A semantic retrieval plane for relevance ranking.
diff --git a/docs/README.md b/docs/README.md
index 3e5c493..305f059 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -1,149 +1,107 @@
-# Documentation Index
+# Documentation
 
-## Start Here 👇
+## Where to Start
 
-**Your MCP server is production-ready. Here's how to use it.**
+| Goal                                           | Document                                             |
+| ---------------------------------------------- | ---------------------------------------------------- |
+| Deploy the server                              | [QUICK_START.md](../QUICK_START.md)                  |
+| See all 39 tools at a glance                   | [QUICK_REFERENCE.md](../QUICK_REFERENCE.md)          |
+| Integrate into a project                       | [MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md) |
+| Stop Claude falling back to grep in long chats | [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md)       |
+| Replace grep/file reads with MCP equivalents   | [TOOL_PATTERNS.md](TOOL_PATTERNS.md)                 |
+| Understand architecture and internals          | [ARCHITECTURE.md](../ARCHITECTURE.md)                |
 
-### Quick Overview (5 min)
+---
 
-→ [INTEGRATION_SUMMARY.md](INTEGRATION_SUMMARY.md)
+## Guides
 
-### Fix Copilot Instructions Drift (15 min)
+### [MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md)
 
-→ [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md) ⭐ THE SOLUTION
+Complete project integration: server setup, VS Code and Claude Desktop config, per-project
+`.github/copilot-instructions.md`, multi-project architecture.
 
-### Copy to Your Projects
+### [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md)
 
-→ [templates/copilot-instructions-template.md](templates/copilot-instructions-template.md)
+How to enforce MCP usage at the **system prompt level** so Claude never falls back to
+grep or file reads — even in 100+ message conversations. Includes ready-to-paste config.
 
-### Complete Integration Guide (30 min)
+### [TOOL_PATTERNS.md](TOOL_PATTERNS.md)
 
-→ [MCP_INTEGRATION_GUIDE.md](MCP_INTEGRATION_GUIDE.md)
+Side-by-side before/after patterns: grep → `graph_query`, file reads → `code_explain`,
+manual impact tracing → `impact_analyze`, test discovery → `test_select`, and more.
 
-### Grep → MCP Patterns (15 min)
+---
 
-→ [TOOL_PATTERNS.md](TOOL_PATTERNS.md)
+## Reference
 
-### Code Comment Conventions
+### [TOOLS_INFORMATION_GUIDE.md](TOOLS_INFORMATION_GUIDE.md)
 
-→ [CODE_COMMENT_STANDARD.md](CODE_COMMENT_STANDARD.md)
+Full 39-tool inventory by category, tool-selection cheatsheet, runtime notes (session
+scoping, rebuild async model, profile-driven output), and output contract reference.
 
-### Consolidated Tool Information
+### [PROJECT_FEATURES_CAPABILITIES.md](PROJECT_FEATURES_CAPABILITIES.md)
 
-→ [TOOLS_INFORMATION_GUIDE.md](TOOLS_INFORMATION_GUIDE.md)
+Feature overview by capability area: graph intelligence, semantic retrieval, testing and
+change impact, architecture governance, agent coordination and memory.
 
-### Project Features & Capabilities
+---
 
-→ [PROJECT_FEATURES_CAPABILITIES.md](PROJECT_FEATURES_CAPABILITIES.md)
+## Development Standards
 
-### Audits & Evaluations Summary
+### [CODE_COMMENT_STANDARD.md](CODE_COMMENT_STANDARD.md)
 
-→ [AUDITS_EVALUATIONS_SUMMARY.md](AUDITS_EVALUATIONS_SUMMARY.md)
+TSDoc format for file headers, exported APIs, and internal helpers. Required for all core
+modules; includes scope guidance and style rules.
+
+---
 
-### Plans & Pending Actions
+## Templates
 
-→ [PLANS_PENDING_ACTIONS_SUMMARY.md](PLANS_PENDING_ACTIONS_SUMMARY.md)
+| File                                                                                     | Usage                                                    |
+| ---------------------------------------------------------------------------------------- | -------------------------------------------------------- |
+| [templates/copilot-instructions-template.md](templates/copilot-instructions-template.md) | Copy to `.github/copilot-instructions.md` in any project |
+| [templates/GRAPH_EXPERT_AGENT.md](templates/GRAPH_EXPERT_AGENT.md)                       | System prompt for an AI agent operating this repo        |
+| [templates/skill-mcp-template.md](templates/skill-mcp-template.md)                       | Skill prompt template for tool-specific tasks            |
+| [templates/toolsets-template.jsonc](templates/toolsets-template.jsonc)                   | VS Code toolset configuration                            |
 
 ---
 
-## File Structure
+## File Index
 
 ```
 docs/
-├─ README.md (you are here)
-├─ INTEGRATION_SUMMARY.md ........... Quick reference + navigation
-├─ CLAUDE_INTEGRATION.md ........... System prompt solution ⭐
-├─ MCP_INTEGRATION_GUIDE.md ........ Complete setup guide
-├─ TOOL_PATTERNS.md ............... Before/after patterns
-├─ TOOLS_INFORMATION_GUIDE.md ...... Consolidated tool inventory
-├─ PROJECT_FEATURES_CAPABILITIES.md  Features and capability map
-├─ AUDITS_EVALUATIONS_SUMMARY.md ... Consolidated findings
-├─ PLANS_PENDING_ACTIONS_SUMMARY.md  Prioritized execution plan
-├─ templates/
-│  ├─ copilot-instructions-template.md  Copy to projects
-│  ├─ skill-mcp-template.md .......... Skill prompt template
-│  └─ toolsets-template.jsonc ........ VS Code toolset template
-└─ (other docs...)
+  README.md                          — This file (navigation hub)
+  CLAUDE_INTEGRATION.md              — Enforce MCP via system prompt (fixes instruction drift)
+  CODE_COMMENT_STANDARD.md           — TSDoc comment conventions
+  MCP_INTEGRATION_GUIDE.md           — Full project setup and integration guide
+  PROJECT_FEATURES_CAPABILITIES.md   — Feature and capability map
+  TOOL_PATTERNS.md                   — Grep → MCP replacement patterns
+  TOOLS_INFORMATION_GUIDE.md         — 39-tool inventory, cheatsheet, runtime notes
+  templates/
+    copilot-instructions-template.md — Copy to any project
+    GRAPH_EXPERT_AGENT.md            — AI agent system prompt for this repo
+    skill-mcp-template.md            — Skill prompt template
+    toolsets-template.jsonc          — VS Code toolsets
 
 Root:
-├─ .github/copilot-instructions.md . For this project (ready to use)
-├─ QUICK_REFERENCE.md ............. All 39 tools
-├─ QUICK_START.md ................. Server deployment
-├─ ARCHITECTURE.md ................ Technical details
-└─ README.md ...................... Project overview
+  QUICK_START.md      — Server deployment (Docker + npm)
+  QUICK_REFERENCE.md  — All 39 tools with params and examples
+  ARCHITECTURE.md     — Technical deep dive (graph pipeline, parsers, engines)
+  README.md           — Project overview and setup
+  .github/copilot-instructions.md — Active Copilot instructions for this repo
 ```
 
 ---
 
-## By Use Case
-
-### I need to fix "Copilot ignores my instructions"
-
-1. Read: **CLAUDE_INTEGRATION.md** (15 min)
-2. Update: `~/.claude_desktop_config.json`
-3. Restart: Claude Desktop
-4. Test: Ask code question
-
-### I want to integrate into my projects
-
-1. Start: **INTEGRATION_SUMMARY.md** (5 min)
-2. Copy: **templates/copilot-instructions-template.md** (1 min)
-3. Setup: Follow checklist (10 min)
-4. Test: Long conversation (5 min)
-
-### I want to replace grep with MCP
-
-1. Learn: **TOOL_PATTERNS.md** (15 min)
-2. Apply: Use pattern in your code
-3. Test: Verify faster + more accurate
-
-### I need complete integration details
-
-1. Read: **MCP_INTEGRATION_GUIDE.md** (30 min)
-2. Follow: Setup phases
-3. Deploy: To all projects
-
-### I need all tool references
-
-1. See: **QUICK_REFERENCE.md** (root)
-2. See: **TOOL_PATTERNS.md** (quick lookup)
-
----
-
-## Key Insight
-
-**System Prompt Engineering (not instructions) solves instruction drift.**
-
-- Instructions fade in long conversations
-- System prompt is protocol-level (never fades)
-- File reads become impossible (not a suggestion)
-- Grep becomes forbidden (not a suggestion)
-
-See: [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md)
-
----
-
-## Performance Gains
+## Performance Reference
 
-| Task              | Before | After | Gain         |
-| ----------------- | ------ | ----- | ------------ |
-| Find symbol       | 450ms  | 50ms  | 9x faster    |
-| Understand        | 5 min  | 200ms | 1500x faster |
-| Impact analysis   | 10 min | 100ms | 6000x faster |
-| Search by concept | 2 min  | 150ms | 800x faster  |
-
----
-
-## 39 Tools at a Glance
-
-**Essential 4:**
-
-- `graph_query` - Find code
-- `code_explain` - Understand symbols
-- `impact_analyze` - What breaks?
-- `test_select` - Which tests?
-
-**+ 35 more** (see QUICK_REFERENCE.md)
+| Task                | Manual | MCP    | Improvement   |
+| ------------------- | ------ | ------ | ------------- |
+| Find symbol         | 450 ms | 50 ms  | 9× faster     |
+| Understand function | 5 min  | 200 ms | 1 500× faster |
+| Impact analysis     | 10 min | 100 ms | 6 000× faster |
+| Search by concept   | 2 min  | 150 ms | 800× faster   |
+| False positive rate | High   | < 1%   | —             |
 
 ---
 
@@ -172,18 +130,3 @@ After implementation:
 ✅ Zero file reads or grep
 ✅ Full dependency context always
 ✅ Heavy MCP dependency, zero fallback
-
----
-
-## Start Now
-
-1. **5 min**: [INTEGRATION_SUMMARY.md](INTEGRATION_SUMMARY.md)
-2. **15 min**: [CLAUDE_INTEGRATION.md](CLAUDE_INTEGRATION.md)
-3. **10 min**: Setup using checklist
-4. **5 min**: Test with code question
-
-**Total: 35 minutes to full setup**
-
----
-
-**Everything you need is here. Let's go! 🚀**
diff --git a/docs/TOOLS_INFORMATION_GUIDE.md b/docs/TOOLS_INFORMATION_GUIDE.md
index 1140080..cc131b9 100644
--- a/docs/TOOLS_INFORMATION_GUIDE.md
+++ b/docs/TOOLS_INFORMATION_GUIDE.md
@@ -20,10 +20,10 @@ Based on the built runtime registry (`dist/tools/registry.js`), the server curre
 | Category     |  Count |
 | ------------ | -----: |
 | graph        |      4 |
-| utility      |      3 |
+| utility      |      2 |
 | code         |      7 |
 | test         |      5 |
-| coordination |      5 |
+| coordination |      6 |
 | setup        |      2 |
 | arch         |      2 |
 | docs         |      2 |
@@ -43,7 +43,6 @@ Based on the built runtime registry (`dist/tools/registry.js`), the server curre
 
 #### Utility
 
-- `diff_since`
 - `tools_list`
 - `contract_validate`
 
@@ -67,6 +66,7 @@ Based on the built runtime registry (`dist/tools/registry.js`), the server curre
 
 #### Coordination
 
+- `diff_since`
 - `context_pack`
 - `agent_claim`
 - `agent_release`
diff --git a/docs/TOOL_PATTERNS.md b/docs/TOOL_PATTERNS.md
index fa8454f..7999065 100644
--- a/docs/TOOL_PATTERNS.md
+++ b/docs/TOOL_PATTERNS.md
@@ -5,6 +5,7 @@ Quick reference for replacing grep/file reads with MCP tools.
 ## Discovery: Find Code
 
 ### ❌ Grep Approach
+
 ```bash
 grep -r "MyClass" src/ --include="*.ts"
 grep -r "import.*AuthService" src/
@@ -12,10 +13,11 @@ grep -r "function.*handle" src/api/
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.query('find all references to MyClass');
-await mcp.query('find all imports of AuthService');
-await mcp.query('find all HTTP request handlers');
+await mcp.query("find all references to MyClass");
+await mcp.query("find all imports of AuthService");
+await mcp.query("find all HTTP request handlers");
 ```
 
 **Benefits**: 10x faster, zero false positives, full context
@@ -25,6 +27,7 @@ await mcp.query('find all HTTP request handlers');
 ## Understanding: Explain Code
 
 ### ❌ File Read Approach
+
 ```bash
 cat src/auth/service.ts  # Read entire file
 grep -n "class AuthService" src/auth/service.ts
@@ -32,8 +35,9 @@ grep -n "validateToken" src/auth/service.ts
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.explain('AuthService');
+await mcp.explain("AuthService");
 // Returns: definition + all methods + dependencies + callers
 ```
 
@@ -44,6 +48,7 @@ await mcp.explain('AuthService');
 ## Impact: What Breaks?
 
 ### ❌ Manual Approach
+
 ```bash
 git diff --name-only HEAD~1
 # Then manually check affected files and tests
@@ -52,9 +57,10 @@ git diff --name-only HEAD~1
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('impact_analyze', {
-  changedFiles: ['src/auth/service.ts']
+await mcp.call("impact_analyze", {
+  changedFiles: ["src/auth/service.ts"],
 });
 // Returns: direct dependents, indirect dependents, affected tests, risk level
 ```
@@ -66,6 +72,7 @@ await mcp.call('impact_analyze', {
 ## Testing: Which Tests to Run?
 
 ### ❌ Manual Approach
+
 ```bash
 find . -name "*.test.ts" | xargs grep -l "AuthService"
 # Time: 5+ minutes
@@ -73,9 +80,10 @@ find . -name "*.test.ts" | xargs grep -l "AuthService"
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('test_select', {
-  changedFiles: ['src/auth/service.ts']
+await mcp.call("test_select", {
+  changedFiles: ["src/auth/service.ts"],
 });
 // Returns: exact test files affected
 ```
@@ -87,6 +95,7 @@ await mcp.call('test_select', {
 ## Patterns: Find Violations
 
 ### ❌ Grep Approach
+
 ```bash
 grep -r "console\.log" src/
 grep -r "\.any()" src/
@@ -94,8 +103,9 @@ grep -r "hardcoded.*password" src/
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('arch_validate', { profile: 'strict' });
+await mcp.call("arch_validate", { profile: "strict" });
 // Returns: architecture violations with severity
 ```
 
@@ -106,6 +116,7 @@ await mcp.call('arch_validate', { profile: 'strict' });
 ## Search by Meaning
 
 ### ❌ Grep Approach
+
 ```bash
 grep -r "validate" src/  # Returns 500+ results
 grep -r "error.*handling" src/  # Returns 1000+ results
@@ -113,10 +124,11 @@ grep -r "error.*handling" src/  # Returns 1000+ results
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('semantic_search', {
-  query: 'input validation patterns',
-  limit: 10
+await mcp.call("semantic_search", {
+  query: "input validation patterns",
+  limit: 10,
 });
 // Returns: 10 most relevant results by meaning
 ```
@@ -128,16 +140,18 @@ await mcp.call('semantic_search', {
 ## Similar Code: Find Patterns
 
 ### ❌ Manual Approach
+
 ```bash
 grep -r "class.*Service" src/
 # Manual comparison of 50+ results
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('find_similar_code', {
-  symbol: 'AuthService',
-  limit: 5
+await mcp.call("find_similar_code", {
+  symbol: "AuthService",
+  limit: 5,
 });
 // Returns: 5 most similar implementations
 ```
@@ -149,6 +163,7 @@ await mcp.call('find_similar_code', {
 ## Architecture: Where Does Code Go?
 
 ### ❌ Manual Approach
+
 ```bash
 # Read architecture docs
 # Manually check layer rules
@@ -157,9 +172,10 @@ await mcp.call('find_similar_code', {
 ```
 
 ### ✅ MCP Approach
+
 ```typescript
-await mcp.call('arch_suggest', {
-  filePath: 'new-feature.ts'
+await mcp.call("arch_suggest", {
+  filePath: "new-feature.ts",
 });
 // Returns: recommended layer + reasoning
 ```
@@ -168,56 +184,56 @@ await mcp.call('arch_suggest', {
 
 ---
 
-## All 38 Tools Quick Lookup
-
-| Use Case | Tool | Example |
-|----------|------|---------|
-| **Find code** | `graph_query` | find all HTTP handlers |
-| **Understand symbol** | `code_explain` | AuthService |
-| **Impact of change** | `impact_analyze` | [files] |
-| **Tests to run** | `test_select` | [files] |
-| **Architecture violations** | `arch_validate` | {} |
-| **Where to put code** | `arch_suggest` | filePath |
-| **Search by concept** | `semantic_search` | "validation" |
-| **Similar patterns** | `find_similar_code` | symbol |
-| **Detect violations** | `find_pattern` | "pattern name" |
-| **Categorize tests** | `test_categorize` | {} |
-| **Test coverage gaps** | `suggest_tests` | symbol |
-| **Get context** | `context_pack` | task, profile |
-| **Code snippets** | `semantic_slice` | symbol |
-| **Historical changes** | `diff_since` | timestamp |
-| **Record decision** | `episode_add` | type, content, agentId |
-| **Recall decisions** | `decision_query` | agentId |
-| **Claim task** | `agent_claim` | agentId, taskName |
-| **Release task** | `agent_release` | agentId, taskName |
-| **Check coordination** | `agent_status` | {} |
-| **Graph health** | `graph_health` | {} |
-| **Rebuild graph** | `graph_rebuild` | mode |
-| **Cypher query** | `graph_query` | query, language:'cypher' |
-| **Set workspace** | `graph_set_workspace` | workspaceRoot, projectId |
-| **Code clusters** | `code_clusters` | {} |
-| **Semantic diff** | `semantic_diff` | symbol1, symbol2 |
-| **Test run** | `test_run` | testFiles |
-| **Progress query** | `progress_query` | task |
-| **Task update** | `task_update` | taskId, status |
-| **Feature status** | `feature_status` | feature |
-| **Blocking issues** | `blocking_issues` | {} |
-| **Reference repo** | `ref_query` | query |
-| **Setup project** | `init_project_setup` | workspaceRoot |
-| **Setup copilot** | `setup_copilot_instructions` | {} |
-| **Reflect on session** | `reflect` | agentId |
-| **Coordination overview** | `coordination_overview` | {} |
-| **Search docs** | `search_docs` | query |
-| **Index docs** | `index_docs` | {} |
+## All 39 Tools Quick Lookup
+
+| Use Case                    | Tool                         | Example                  |
+| --------------------------- | ---------------------------- | ------------------------ |
+| **Find code**               | `graph_query`                | find all HTTP handlers   |
+| **Understand symbol**       | `code_explain`               | AuthService              |
+| **Impact of change**        | `impact_analyze`             | [files]                  |
+| **Tests to run**            | `test_select`                | [files]                  |
+| **Architecture violations** | `arch_validate`              | {}                       |
+| **Where to put code**       | `arch_suggest`               | filePath                 |
+| **Search by concept**       | `semantic_search`            | "validation"             |
+| **Similar patterns**        | `find_similar_code`          | symbol                   |
+| **Detect violations**       | `find_pattern`               | "pattern name"           |
+| **Categorize tests**        | `test_categorize`            | {}                       |
+| **Test coverage gaps**      | `suggest_tests`              | symbol                   |
+| **Get context**             | `context_pack`               | task, profile            |
+| **Code snippets**           | `semantic_slice`             | symbol                   |
+| **Historical changes**      | `diff_since`                 | timestamp                |
+| **Record decision**         | `episode_add`                | type, content, agentId   |
+| **Recall decisions**        | `decision_query`             | agentId                  |
+| **Claim task**              | `agent_claim`                | agentId, taskName        |
+| **Release task**            | `agent_release`              | agentId, taskName        |
+| **Check coordination**      | `agent_status`               | {}                       |
+| **Graph health**            | `graph_health`               | {}                       |
+| **Rebuild graph**           | `graph_rebuild`              | mode                     |
+| **Cypher query**            | `graph_query`                | query, language:'cypher' |
+| **Set workspace**           | `graph_set_workspace`        | workspaceRoot, projectId |
+| **Code clusters**           | `code_clusters`              | {}                       |
+| **Semantic diff**           | `semantic_diff`              | symbol1, symbol2         |
+| **Test run**                | `test_run`                   | testFiles                |
+| **Progress query**          | `progress_query`             | task                     |
+| **Task update**             | `task_update`                | taskId, status           |
+| **Feature status**          | `feature_status`             | feature                  |
+| **Blocking issues**         | `blocking_issues`            | {}                       |
+| **Reference repo**          | `ref_query`                  | query                    |
+| **Setup project**           | `init_project_setup`         | workspaceRoot            |
+| **Setup copilot**           | `setup_copilot_instructions` | {}                       |
+| **Reflect on session**      | `reflect`                    | agentId                  |
+| **Coordination overview**   | `coordination_overview`      | {}                       |
+| **Search docs**             | `search_docs`                | query                    |
+| **Index docs**              | `index_docs`                 | {}                       |
 
 ## Performance Comparison
 
-| Task | Grep | MCP | Improvement |
-|------|------|-----|---|
-| Find symbol | 450ms | 50ms | 9x |
-| Understand function | 5 min manual | 200ms | 1500x |
-| Impact analysis | 10 min manual | 100ms | 6000x |
-| Search by meaning | 2 min grep | 150ms | 800x |
+| Task                | Grep          | MCP   | Improvement |
+| ------------------- | ------------- | ----- | ----------- |
+| Find symbol         | 450ms         | 50ms  | 9x          |
+| Understand function | 5 min manual  | 200ms | 1500x       |
+| Impact analysis     | 10 min manual | 100ms | 6000x       |
+| Search by meaning   | 2 min grep    | 150ms | 800x        |
 
 ---
 
@@ -237,11 +253,13 @@ await mcp.call('arch_suggest', {
 ## Token Efficiency (Long Conversations)
 
 Use these for compact responses:
+
 - `profile: 'compact'` — for token-light answers
 - `semantic_slice` — get only relevant code lines
 - `context_pack` — multi-file context under budget
 
 Avoid:
+
 - Full file reads (use `semantic_slice` instead)
 - Long lists (use `limit` parameter)
 - Multiple separate queries (combine when possible)
diff --git a/docs/templates/GRAPH_EXPERT_AGENT.md b/docs/templates/GRAPH_EXPERT_AGENT.md
index 9a9c788..c6da1c2 100644
--- a/docs/templates/GRAPH_EXPERT_AGENT.md
+++ b/docs/templates/GRAPH_EXPERT_AGENT.md
@@ -8,7 +8,7 @@ You are the **Graph Expert Agent** for this project. Your goal is to produce acc
 
 ## Ground Truth About This Project
 
-- Runtime: Node/TypeScript MCP server in `src/server.ts` (33 tools)
+- Runtime: Node/TypeScript MCP server in `src/server.ts` (39 tools)
 - Storage: Memgraph MAGE (`memgraph/memgraph-mage:latest`) + Qdrant
 - Transport: MCP HTTP (`POST /` and `POST /mcp`) and health at `GET /health`
 - Workspace context is **session-scoped** (per MCP session), not process-global
diff --git a/src/graph/orchestrator.ts b/src/graph/orchestrator.ts
index 98be248..dce5b75 100644
--- a/src/graph/orchestrator.ts
+++ b/src/graph/orchestrator.ts
@@ -1210,7 +1210,6 @@ export class GraphOrchestrator {
             f.priority = $priority,
             f.projectId = $projectId,
             f.createdAt = timestamp()
-          ON MATCH DO NOTHING
         `,
         params: {
           id: `${projectId}:feature:${feature.id}`,
diff --git a/src/tools/__tests__/tool-handlers.contract.test.ts b/src/tools/__tests__/tool-handlers.contract.test.ts
index 3678131..95d6426 100644
--- a/src/tools/__tests__/tool-handlers.contract.test.ts
+++ b/src/tools/__tests__/tool-handlers.contract.test.ts
@@ -116,7 +116,7 @@ describe("ToolHandlers contract normalization", () => {
     const parsed = JSON.parse(response);
 
     expect(parsed.ok).toBe(true);
-    expect(parsed.contractWarnings).toContain("mapped changedFiles -> files");
+    expect(parsed.contractWarnings ?? []).not.toContain("mapped changedFiles -> files");
     expect(selectAffectedTests).toHaveBeenCalledWith(["src/baz.ts"], true, 2);
   });
 
diff --git a/src/tools/__tests__/tool-handlers.integration.test.ts b/src/tools/__tests__/tool-handlers.integration.test.ts
index 958b9da..aa128bc 100644
--- a/src/tools/__tests__/tool-handlers.integration.test.ts
+++ b/src/tools/__tests__/tool-handlers.integration.test.ts
@@ -715,8 +715,8 @@ describe("INCONSISTENCY: arch_suggest parameter naming", () => {
   });
 });
 
-describe("INCONSISTENCY: impact_analyze changedFiles normalization", () => {
-  it("normalizes changedFiles -> files with contract warning via callTool", async () => {
+describe("impact_analyze changedFiles normalization", () => {
+  it("passes changedFiles through natively (no contract warning)", async () => {
     const { handlers } = createHandlers();
 
     (handlers as any).testEngine = {
@@ -734,7 +734,7 @@ describe("INCONSISTENCY: impact_analyze changedFiles normalization", () => {
     const parsed = parseResponse(response);
 
     expect(parsed.ok).toBe(true);
-    expect(parsed.contractWarnings).toContain("mapped changedFiles -> files");
+    expect(parsed.contractWarnings ?? []).not.toContain("mapped changedFiles -> files");
   });
 
   it("works directly with files parameter (no warning)", async () => {
diff --git a/src/tools/handlers/core-graph-tools.ts b/src/tools/handlers/core-graph-tools.ts
index 1efe304..202b63e 100644
--- a/src/tools/handlers/core-graph-tools.ts
+++ b/src/tools/handlers/core-graph-tools.ts
@@ -702,9 +702,11 @@ export const coreGraphToolDefinitions: ToolDefinition[] = [
         const memgraphFileCount = ctx.toSafeNumber(stats.fileCount) ?? 0;
         const memgraphFuncCount = ctx.toSafeNumber(stats.funcCount) ?? 0;
         const memgraphClassCount = ctx.toSafeNumber(stats.classCount) ?? 0;
-        const memgraphImportCount = ctx.toSafeNumber(stats.importCount) ?? 0;
-        const memgraphIndexableCount =
-          memgraphFileCount + memgraphFuncCount + memgraphClassCount + memgraphImportCount;
+        // memgraphIndexableCount counts the same three symbol types tracked by the
+        // in-memory index (FILE, FUNCTION, CLASS) so both sides of the drift check
+        // are symmetric.  IMPORT, VARIABLE, TEST_* etc. are deliberately excluded
+        // because the in-memory index may not carry all of them after every sync.
+        const memgraphIndexableCount = memgraphFileCount + memgraphFuncCount + memgraphClassCount;
 
         const indexStats = ctx.context.index.getStatistics();
         const indexFileCount = ctx.context.index.getNodesByType("FILE").length;
@@ -740,7 +742,10 @@ export const coreGraphToolDefinitions: ToolDefinition[] = [
               )
             : 0;
 
-        const indexDrift = Math.abs(indexStats.totalNodes - memgraphIndexableCount) > 3;
+        // Compare same types on both sides: in-memory FILE+FUNC+CLASS vs Memgraph FILE+FUNC+CLASS
+        const cachedIndexableCount = indexFileCount + indexFuncCount + indexClassCount;
+        const indexDrift =
+          memgraphIndexableCount > 0 && Math.abs(cachedIndexableCount - memgraphIndexableCount) > 3;
         const embeddingDrift = embeddingCount < indexedSymbols;
 
         const txMetadataResult = await ctx.context.memgraph.executeCypher(
@@ -793,7 +798,9 @@ export const coreGraphToolDefinitions: ToolDefinition[] = [
               driftDetected: indexDrift,
               memgraphNodes: memgraphNodeCount,
               memgraphIndexableNodes: memgraphIndexableCount,
-              cachedNodes: indexStats.totalNodes,
+              cachedNodes: cachedIndexableCount,
+              cachedNodeNote:
+                "FILE+FUNCTION+CLASS counts compared; other types (VARIABLE, TEST_*, SECTION, etc.) excluded from drift check",
               memgraphRels: memgraphRelCount,
               cachedRels: indexStats.totalRelationships,
               recommendation: indexDrift
diff --git a/src/tools/handlers/core-utility-tools.ts b/src/tools/handlers/core-utility-tools.ts
index 7f52ff9..ca1d536 100644
--- a/src/tools/handlers/core-utility-tools.ts
+++ b/src/tools/handlers/core-utility-tools.ts
@@ -4,7 +4,7 @@
  */
 
 import * as z from "zod";
-import type { HandlerBridge, ToolDefinition , ToolArgs } from "../types.js";
+import type { HandlerBridge, ToolDefinition, ToolArgs } from "../types.js";
 
 export const coreUtilityToolDefinitions: ToolDefinition[] = [
   {
@@ -26,13 +26,7 @@ export const coreUtilityToolDefinitions: ToolDefinition[] = [
 
       const KNOWN_CATEGORIES: Record<string, string[]> = {
         setup: ["init_project_setup", "setup_copilot_instructions"],
-        graph: [
-          "graph_set_workspace",
-          "graph_rebuild",
-          "graph_query",
-          "graph_health",
-          "ref_query",
-        ],
+        graph: ["graph_set_workspace", "graph_rebuild", "graph_query", "graph_health", "ref_query"],
         architecture: ["arch_validate", "arch_suggest"],
         semantic: [
           "semantic_search",
@@ -50,6 +44,7 @@ export const coreUtilityToolDefinitions: ToolDefinition[] = [
         progress: ["progress_query", "task_update", "feature_status"],
         coordination: [
           "agent_claim",
+          "agent_status",
           "agent_release",
           "coordination_overview",
           "diff_since",
diff --git a/src/tools/handlers/test-tools.ts b/src/tools/handlers/test-tools.ts
index ed70b59..d4bc03f 100644
--- a/src/tools/handlers/test-tools.ts
+++ b/src/tools/handlers/test-tools.ts
@@ -13,7 +13,8 @@ import * as path from "path";
 import type { GraphNode, GraphRelationship } from "../../graph/index.js";
 import { execWithTimeout } from "../../utils/exec-utils.js";
 import * as z from "zod";
-import type { HandlerBridge, ToolDefinition , ToolArgs } from "../types.js";
+import type { HandlerBridge, ToolDefinition, ToolArgs } from "../types.js";
+import { logger } from "../../utils/logger.js";
 
 /**
  * Determine the command and arguments used to execute tests.
@@ -98,7 +99,9 @@ async function resolveDirectImpact(ctx: HandlerBridge, changedFiles: string[]):
         { projectId, changedPaths: changedFiles },
       );
 
-      const paths: string[] = result.data.map((row: Record<string, unknown>) => String(row.path ?? "")).filter(Boolean);
+      const paths: string[] = result.data
+        .map((row: Record<string, unknown>) => String(row.path ?? ""))
+        .filter(Boolean);
 
       if (paths.length > 0) {
         return paths;
@@ -376,7 +379,7 @@ export const testToolDefinitions: ToolDefinition[] = [
           );
         }
 
-        const cwd = process.cwd();
+        const cwd = ctx.getActiveProjectContext?.().workspaceRoot ?? process.cwd();
 
         // Resolve runner: config > auto-detect by extension > vitest fallback
         const { cmd, env: runnerEnv } = resolveTestRunner(
@@ -385,12 +388,12 @@ export const testToolDefinitions: ToolDefinition[] = [
           ctx.context.config?.testing,
         );
 
-        console.error(`[ToolHandlers] Executing: ${cmd}`);
+        logger.debug(`[test_run] Executing in ${cwd}: ${cmd}`);
 
         try {
           const augmentedEnv = { ...process.env, ...(runnerEnv ?? {}) };
           const output = execWithTimeout(cmd, {
-            cwd: process.cwd(),
+            cwd,
             encoding: "utf-8",
             stdio: ["pipe", "pipe", "pipe"],
             env: augmentedEnv,
diff --git a/src/tools/tool-handler-base.ts b/src/tools/tool-handler-base.ts
index c759657..a77c761 100644
--- a/src/tools/tool-handler-base.ts
+++ b/src/tools/tool-handler-base.ts
@@ -377,18 +377,7 @@ export abstract class ToolHandlerBase extends SessionManager {
     const warnings: string[] = [];
     const normalized = { ...(rawArgs || {}) };
 
-    if (toolName === "impact_analyze") {
-      const files = Array.isArray(normalized.files)
-        ? normalized.files
-        : Array.isArray(normalized.changedFiles)
-          ? normalized.changedFiles
-          : [];
-      if (Array.isArray(normalized.changedFiles) && !Array.isArray(normalized.files)) {
-        warnings.push("mapped changedFiles -> files");
-      }
-      normalized.files = files;
-      delete normalized.changedFiles;
-    }
+    // impact_analyze: both 'files' and 'changedFiles' are native Zod params — no alias mapping needed.
 
     if (toolName === "progress_query") {
       if (typeof normalized.type !== "string") {