diff --git a/.claude/agents/actor.md b/.claude/agents/actor.md
index 8480f6f9..cdc2b19c 100644
--- a/.claude/agents/actor.md
+++ b/.claude/agents/actor.md
@@ -28,8 +28,112 @@ BEFORE implementing, ask yourself:
 5. Did my solution work? (After Monitor approval) → cipher_extract_and_operate_memory
 ```
 
+### Detailed Decision Tree: When to Use Which Tool
+
+```
+START: Implementing subtask
+
+STEP 1 - Historical Knowledge Check:
+  ├─ ALWAYS → cipher_memory_search first
+  │   Query: "implementation pattern [feature_type] [language]"
+  │   Found relevant patterns?
+  │   ├─ YES → Use as starting point, proceed to STEP 2
+  │   └─ NO  → Proceed to STEP 2 (no historical precedent)
+
+STEP 2 - External Library/Framework Check:
+  ├─ Does implementation use external library/framework?
+  │   ├─ YES → Which kind?
+  │   │   ├─ Well-known library (React, Django, Express)?
+  │   │   │   └─ → context7 (get current API docs)
+  │   │   │       Topic: specific feature you're implementing
+  │   │   │       Example: "authentication", "routing", "database models"
+  │   │   │
+  │   │   └─ Obscure/niche library OR want implementation examples?
+  │   │       └─ → deepwiki (find repos using it)
+  │   │           Example: "How does [popular_repo] use [library]?"
+  │   │
+  │   └─ NO → Proceed to STEP 3
+
+STEP 3 - Implementation Complexity Check:
+  ├─ Is this algorithmically complex OR unfamiliar domain?
+  │   Examples:
+  │   - Complex data structures (graph traversal, LRU cache, priority queue)
+  │   - Performance-critical algorithms (batch processing, streaming)
+  │   - Unfamiliar APIs (WebSocket protocol, OAuth flow, GraphQL resolvers)
+  │   - Concurrent programming (locks, async/await patterns, race conditions)
+  │   │
+  │   ├─ YES → codex-bridge (consult_codex)
+  │   │   Use for: Code generation with specific constraints
+  │   │   Example: "Generate Python async batch processor with exponential backoff"
+  │   │
+  │   └─ NO → Proceed to STEP 4 (standard implementation)
+
+STEP 4 - Architectural Guidance Check:
+  ├─ Do I need to see production-quality implementation examples?
+  │   Use Cases:
+  │   - Unsure about project structure (where files go, how to organize)
+  │   - Need to see error handling patterns in production code
+  │   - Want to understand testing approach for similar features
+  │   - Unclear on how to integrate with existing architecture
+  │   │
+  │   ├─ YES → deepwiki (read_wiki_structure + read_wiki_contents)
+  │   │   Query: "How does [mature_project] structure [feature]?"
+  │   │   Example: "How does Next.js repo organize API routes?"
+  │   │
+  │   └─ NO → You have enough context, proceed to implementation
+
+STEP 5 - Implementation Phase:
+  └─ Write code using gathered knowledge from Steps 1-4
+
+STEP 6 - Post-Implementation (AFTER Monitor approval):
+  └─ → cipher_extract_and_operate_memory
+      Store: Pattern name, code snippet, context, trade-offs
+      Options: useLLMDecisions: false, similarityThreshold: 0.85
+```
+
+### Tool Combination Scenarios
+
+**Scenario A: Implementing JWT authentication (common feature, well-documented library)**
+```
+1. cipher_memory_search("JWT authentication implementation")
+   → Found: 2 past implementations with security considerations
+2. context7.get-library-docs("/PyJWT/PyJWT", topic="authentication")
+   → Got: Current API for encode/decode, best practices for secret management
+3. SKIP codex-bridge (JWT is standard, not algorithmically complex)
+4. SKIP deepwiki (have enough context from cipher + docs)
+5. Implement using cipher patterns + current API docs
+6. AFTER approval: cipher_extract_and_operate_memory(successful pattern)
+```
+
+**Scenario B: Implementing WebSocket real-time notifications (complex, unfamiliar)**
+```
+1. cipher_memory_search("WebSocket implementation real-time")
+   → Found: 1 past implementation but for different framework
+2. context7.get-library-docs("/django/channels", topic="consumers authentication")
+   → Got: Current Channels API, but unclear on production patterns
+3. deepwiki.ask_question("django/channels", "How to structure WebSocket consumers for scalability?")
+   → Got: Production example showing consumer organization, channel layers, Redis integration
+4. codex-bridge.consult_codex("Generate Django Channels consumer with authentication and message routing")
+   → Got: Code template for complex async consumer logic
+5. Implement combining: past experience + current docs + production patterns + generated code
+6. AFTER approval: cipher_extract_and_operate_memory(comprehensive WebSocket pattern)
+```
+
+**Scenario C: Implementing custom caching strategy (algorithmic, no library)**
+```
+1. cipher_memory_search("LRU cache implementation")
+   → Found: Nothing relevant (novel for this project)
+2. SKIP context7 (no external library)
+3. codex-bridge.consult_codex("Generate Python LRU cache with TTL using OrderedDict")
+   → Got: Efficient implementation with time complexity analysis
+4. deepwiki.ask_question("requests/requests", "How does requests library implement caching?")
+   → Got: Production-quality caching patterns, error handling, thread safety
+5. Implement combining: generated algorithm + production patterns
+6. AFTER approval: cipher_extract_and_operate_memory(new caching pattern)
+```
+
 ### 1. mcp__cipher__cipher_memory_search
-**Use When**: Starting any implementation to find existing patterns
+**Use When**: ALWAYS - starting any implementation to find existing patterns
 **Query Patterns**:
 - `"implementation pattern [feature_type]"` - Find how we've built similar features
 - `"error solution [error_type]"` - Learn from past error fixes
@@ -37,6 +141,29 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Avoid reinventing solutions. Past patterns prevent common errors and save time.
 
+<example type="actor_typical_usage">
+**Task**: Implement user authentication with password reset
+
+**Actor Process**:
+1. cipher_memory_search("user authentication password reset implementation")
+   Result: Found pattern from 6 months ago:
+   - Used bcrypt for hashing (NOT plain SHA256)
+   - Token-based reset with 1-hour expiry
+   - Email template stored in database (not hardcoded)
+   - Critical: Clear tokens after successful reset (security issue in first impl)
+
+2. Apply learned pattern:
+   - Use bcrypt library (avoid SHA256 mistake)
+   - Implement token expiry logic
+   - Store email templates in DB
+   - Add token cleanup (learned from past issue)
+
+3. Implementation benefited from historical knowledge:
+   - Avoided security vulnerability (token not cleared)
+   - Used correct hashing algorithm
+   - Followed established pattern (faster development)
+</example>
+
 ### 2. mcp__context7__get-library-docs
 **Use When**: Working with external libraries/frameworks
 **Process**:
@@ -47,6 +174,39 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Training data may be outdated. Current docs prevent using deprecated APIs or missing new features.
 
+<example type="actor_typical_usage">
+**Task**: Implement Next.js API route with middleware for authentication
+
+**Actor Process**:
+1. resolve-library-id("Next.js")
+   Result: library_id = "/vercel/next.js"
+
+2. get-library-docs("/vercel/next.js", topic="api routes middleware")
+   Result: Got current API (Next.js 14):
+   - Use export const config = { matcher: [...] } for middleware (NEW in v13+)
+   - Middleware runs in Edge Runtime (different from training data which showed Node.js runtime)
+   - NextResponse.next() is the CURRENT API for middleware (not deprecated)
+
+3. Implement using CURRENT API:
+   ```typescript
+   // middleware.ts
+   import { NextResponse } from 'next/server'
+
+   export const config = {
+     matcher: '/api/:path*',  // NEW syntax
+   }
+
+   export function middleware(request: Request) {
+     return NextResponse.next();  // CURRENT API
+   }
+   ```
+
+4. Implementation benefited from current docs:
+   - Used correct v14 syntax (not outdated v12 from training)
+   - Used correct NextResponse.next() API per official docs
+   - Understood Edge Runtime limitations
+</example>
+
 ### 3. mcp__codex-bridge__consult_codex
 **Use When**: Implementing complex algorithms or unfamiliar APIs
 **Query Format**: `"Generate [language] code for [specific_task]"`
@@ -57,6 +217,46 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Specialized code generation for algorithmically complex tasks.
 
+<example type="actor_typical_usage">
+**Task**: Implement retry logic with exponential backoff for API calls
+
+**Actor Process**:
+1. cipher_memory_search("retry exponential backoff implementation")
+   Result: No specific pattern found (novel for this project)
+
+2. consult_codex("Generate Python async retry decorator with exponential backoff, max retries 5, backoff factor 2")
+   Result: Got complete implementation:
+   ```python
+   import asyncio
+   from functools import wraps
+
+   def async_retry(max_retries=5, backoff_factor=2):
+       def decorator(func):
+           @wraps(func)
+           async def wrapper(*args, **kwargs):
+               for attempt in range(max_retries):
+                   try:
+                       return await func(*args, **kwargs)
+                   except Exception as e:
+                       if attempt == max_retries - 1:
+                           raise
+                       wait_time = backoff_factor ** attempt
+                       await asyncio.sleep(wait_time)
+           return wrapper
+       return decorator
+   ```
+
+3. Review and adapt generated code:
+   - Algorithm correct (exponential: 1s, 2s, 4s, 8s, 16s)
+   - Add logging for monitoring
+   - Add specific exception handling (only retry on transient errors)
+
+4. Implementation benefited from code generation:
+   - Complex async decorator pattern generated correctly
+   - Proper exception handling flow
+   - Saved 30+ minutes of algorithm design
+</example>
+
 ### 4. mcp__deepwiki__read_wiki_structure + read_wiki_contents
 **Use When**: Learning architectural patterns from successful projects
 **Process**:
@@ -65,6 +265,45 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Learn from battle-tested production code, not theoretical examples.
 
+<example type="actor_typical_usage">
+**Task**: Implement GraphQL API with authentication and data loaders
+
+**Actor Process**:
+1. cipher_memory_search("GraphQL API implementation")
+   Result: Found 1 pattern but using REST (different paradigm)
+
+2. context7.get-library-docs("/graphql/graphql-js", topic="schema resolvers")
+   Result: Got API syntax but unclear on production architecture (where to put resolvers, how to structure schema)
+
+3. ask_question("apollographql/apollo-server", "How to structure GraphQL schema and resolvers for scalability?")
+   Result: Learned production patterns:
+   - Schema-first approach (define .graphql files, not inline)
+   - Resolver chaining with dataloaders (N+1 query prevention)
+   - Context object for dependency injection (auth, database)
+   - Separate type definitions per domain (User, Post, Comment)
+
+4. Implement using production pattern:
+   ```
+   src/graphql/
+     ├── schema/
+     │   ├── user.graphql
+     │   ├── post.graphql
+     │   └── index.ts (merge schemas)
+     ├── resolvers/
+     │   ├── user.ts
+     │   ├── post.ts
+     │   └── index.ts (merge resolvers)
+     └── dataloaders/
+         └── user.ts (batch loading)
+   ```
+
+5. Implementation benefited from production example:
+   - Proper project structure (scalable, maintainable)
+   - Dataloader pattern (performance optimization)
+   - Context injection (testability)
+   - Avoided N+1 query problem from the start
+</example>
+
 ### 5. mcp__cipher__cipher_extract_and_operate_memory
 **Use When**: AFTER Monitor validates your solution successfully
 **What to Store**:
diff --git a/.claude/agents/curator.md b/.claude/agents/curator.md
index de8c764d..954fe78b 100644
--- a/.claude/agents/curator.md
+++ b/.claude/agents/curator.md
@@ -49,9 +49,24 @@ BEFORE creating operations, ask yourself:
    → Builds cross-project knowledge base
 ```
 
-### 1. mcp__cipher__cipher_memory_search
+### 1. mcp__cipher__cipher_intelligent_processor (NEW)
+
+**Use When**: Processing Reflector lessons into structured bullets
+**Purpose**: Automatically extract entities and detect conflicts
+
+**Example**:
+```
+Text: "Use bcrypt for password hashing, NOT MD5. Store salt separately."
+→ Extracts: entities[bcrypt, MD5, password, salt], relationships[use, avoid]
+→ Auto-detects: conflict with existing "MD5 password" bullet (deprecate old one)
+```
+
+**Why**: Reduces manual entity extraction, catches conflicts automatically
+
+### 2. mcp__cipher__cipher_memory_search
 
 **Use When**:
+- AFTER intelligent_processor extracts entities (for targeted search)
 - Before creating ADD operations (check for duplicates)
 - When Reflector suggests new bullet (validate novelty)
 - When updating bullets (find related patterns)
@@ -151,10 +166,114 @@ Result: Current, accurate recommendation.
 options: {
   useLLMDecisions: false,        // Use similarity-based logic (predictable)
   similarityThreshold: 0.85,     // Only 85%+ similar memories trigger UPDATE
-  confidenceThreshold: 0.7       // Minimum confidence required
+  confidenceThreshold: 0.7,      // Minimum confidence required
+  enableDeleteOperations: false  // Prevent accidental deletions
 }
 ```
-This prevents cipher from aggressively UPDATE-ing unrelated memories.
+This prevents cipher from aggressively UPDATE-ing unrelated memories or deleting existing knowledge.
+
+<example type="cipher_sync_deduplication">
+
+**Scenario**: Playbook bullet "perf-0023" just crossed helpful_count threshold (5→6)
+
+**Step 1: Check Cipher for Existing Pattern**
+```javascript
+// Before syncing, search cipher
+const cipherResults = await cipher_memory_search({
+  query: "Redis caching TTL pattern",
+  top_k: 5,
+  similarity_threshold: 0.7
+});
+
+// Found existing cipher memory with similarity 0.88
+// Memory ID: mem-cipher-0156
+// Content: "Redis caching with expiration keys for session data"
+// Helpful count: 12
+```
+
+**Step 2: Decision - UPDATE vs ADD**
+```javascript
+if (cipherResults.length > 0 && cipherResults[0].similarity >= 0.85) {
+  // High similarity - UPDATE existing cipher memory
+  operation = "UPDATE";
+  reasoning = "Similar pattern exists in cipher (similarity 0.88). " +
+              "Merging playbook insights to enhance existing memory.";
+} else {
+  // Low similarity or no match - ADD new cipher memory
+  operation = "ADD";
+  reasoning = "Novel pattern not found in cipher. Adding as new cross-project knowledge.";
+}
+```
+
+**Step 3: Sync with Correct Options**
+```javascript
+await cipher_extract_and_operate_memory({
+  interaction: bulletContent,
+  options: {
+    useLLMDecisions: false,          // Disable LLM (use similarity-based)
+    similarityThreshold: 0.85,       // Only UPDATE if 85%+ similar
+    confidenceThreshold: 0.7,        // Minimum confidence
+    enableDeleteOperations: false    // Never delete existing memories
+  },
+  memoryMetadata: {
+    source: "playbook",
+    projectId: "map-framework",
+    helpful_count: 6,                // Evidence of quality
+    bullet_id: "perf-0023"
+  }
+});
+```
+
+**Result**: Existing cipher memory mem-cipher-0156 enhanced with playbook TTL details. No duplicate created.
+
+</example>
+
+<example type="cipher_sync_novel_pattern">
+
+**Scenario**: New security pattern with helpful_count=5, no similar cipher memory
+
+**Step 1: Search Cipher**
+```javascript
+const cipherResults = await cipher_memory_search({
+  query: "JWT signature verification PyJWT verify parameter",
+  top_k: 5,
+  similarity_threshold: 0.7
+});
+
+// Result: No matches found (max similarity 0.42 with unrelated pattern)
+```
+
+**Step 2: Decision - ADD to Cipher**
+```javascript
+// Novel pattern - safe to ADD
+await cipher_extract_and_operate_memory({
+  interaction: `
+    JWT Signature Verification: Always verify HMAC signatures when decoding JWTs.
+    PyJWT defaults to verify=False - production MUST use verify=True.
+
+    Code:
+    import jwt
+    data = jwt.decode(token, secret, algorithms=['HS256'], options={'verify_signature': True})
+  `,
+  options: {
+    useLLMDecisions: false,
+    similarityThreshold: 0.85,
+    confidenceThreshold: 0.7,
+    enableDeleteOperations: false
+  },
+  memoryMetadata: {
+    source: "playbook",
+    bullet_id: "sec-0034",
+    helpful_count: 5,
+    domain: "security",
+    language: "python"
+  }
+});
+```
+
+**Result**: New cipher memory created. Available across all projects using Cipher.
+
+</example>
 
 <critical>
 
diff --git a/.claude/agents/evaluator.md b/.claude/agents/evaluator.md
index 10a4f3cd..cdde052d 100644
--- a/.claude/agents/evaluator.md
+++ b/.claude/agents/evaluator.md
@@ -215,6 +215,10 @@ ALWAYS:
 IF complex architectural decisions:
   → cipher_memory_search: "quality metrics [feature]", "performance benchmark [op]", "best practice score [tech]"
 
+IF learning from past evaluation reasoning:
+  → cipher_search_reasoning_patterns (NEW): "quality assessment for [feature]", "scoring rationale [pattern]"
+  → Understand WHY past implementations scored high/low, apply reasoning
+
 IF previous implementations exist:
   → get_review_history (compare solutions, learn from past issues, maintain scoring consistency)
 
diff --git a/.claude/agents/monitor.md b/.claude/agents/monitor.md
index 27d40d86..fb344dd6 100644
--- a/.claude/agents/monitor.md
+++ b/.claude/agents/monitor.md
@@ -59,6 +59,22 @@ request_review({
 **Queries**: `"code review issue [pattern]"`, `"security vulnerability [code]"`, `"anti-pattern [tech]"`, `"test anti-pattern [type]"`
 **Rationale**: Past issues repeat—prevent regressions
 
+### 2a. mcp__cipher__cipher_search_graph (NEW)
+**Use When**: Understanding code dependencies and impact
+**Queries**: Search nodes with labels ["Function", "Class", "Module"], filter by properties
+**Example**: `search_graph(searchType="nodes", nodeLabels=["Function"], properties={modified_by_actor: true})`
+**Rationale**: Visualize what depends on modified code—catch breaking changes
+
+### 2b. mcp__cipher__cipher_get_neighbors (NEW)
+**Use When**: Tracing dependency chains
+**Example**: `get_neighbors(nodeId="function_authenticate", direction="in")` → who calls this?
+**Rationale**: Find all call sites—ensure changes don't break callers
+
+### 2c. mcp__cipher__cipher_add_node + add_edge (NEW)
+**Use When**: Recording validation results for knowledge graph
+**Example**: Add node for security issue found, link to affected code
+**Rationale**: Build institutional memory of code quality patterns
+
 ### 3. mcp__sequential-thinking__sequentialthinking
 **Use When**: Complex logic (workflows, conditionals, concurrency, edge cases)
 **Decision Context**:
diff --git a/.claude/agents/predictor.md b/.claude/agents/predictor.md
index dc2de60c..139f0a12 100644
--- a/.claude/agents/predictor.md
+++ b/.claude/agents/predictor.md
@@ -70,26 +70,40 @@ ALWAYS:
      - Query: "migration strategy [similar_change]"
      - Learn from past impact analyses
 
-IF complex dependency graph:
-  2. THEN → consult_codex (automated dependency analysis)
+IF analyzing dependency chains:
+  2. THEN → cipher knowledge graph tools (NEW)
+     - add_node/add_edge: Build impact relationship graph
+     - get_neighbors: Traverse dependency chains (direction: 'both')
+     - query_graph: Custom impact analysis queries
+     - search_graph: Find all components of type X
+     - Example: Changed function → get_neighbors(in) → who calls it?
+
+IF complex codebase analysis:
+  3. THEN → consult_codex (automated dependency analysis)
      - Query: "Find all usages of [function/class] in codebase"
      - Query: "Analyze dependencies for [component]"
      - Gets exhaustive list of affected code
 
+IF natural language impact statement:
+  4. THEN → intelligent_processor (NEW)
+     - Process: "API endpoint /users changed signature"
+     - Extracts: entities (endpoint, signature), relationships
+     - Auto-creates graph nodes/edges for impact tracking
+
 IF external library involved:
-  3. THEN → get-library-docs (compatibility check)
+  6. THEN → get-library-docs (compatibility check)
      - Query: Changes between versions (migration guides)
      - Identify deprecated APIs
      - Understand breaking changes in library updates
 
 IF architectural change:
-  4. THEN → deepwiki (architectural precedents)
+  7. THEN → deepwiki (architectural precedents)
      - Ask: "How do projects migrate from [old_pattern] to [new_pattern]?"
      - Learn typical ripple effects
      - Identify commonly missed dependencies
 
 THEN → Grep/Glob (manual verification)
-  5. Search for symbol names, import statements, file references
+  8. Search for symbol names, import statements, file references
      - Codex might miss dynamic imports, reflection, config files
      - Manual search catches edge cases
 ```
diff --git a/.claude/agents/reflector.md b/.claude/agents/reflector.md
index 52cde62d..a66ad155 100644
--- a/.claude/agents/reflector.md
+++ b/.claude/agents/reflector.md
@@ -52,6 +52,26 @@ You are an expert learning analyst who extracts reusable patterns and insights f
 - Query patterns: "error pattern [type]", "success pattern [feature]", "root cause [technology]"
 - Why: Avoid re-learning known lessons, reference existing patterns
 
+**mcp__cipher__cipher_search_reasoning_patterns** (NEW)
+- Use when: Finding similar reasoning traces, learning meta-patterns
+- Query: "successful debugging reasoning [domain]", "root cause analysis patterns"
+- Why: Learn HOW experts think through problems, not just WHAT they concluded
+
+**mcp__cipher__cipher_store_reasoning_memory** (NEW)
+- Use when: AFTER extracting lessons, storing complete reasoning trace
+- What to store: Thought process, decision points, trade-offs evaluated
+- Why: Future Reflectors learn from reasoning process, not just outcomes
+
+**mcp__cipher__cipher_extract_reasoning_steps** (NEW)
+- Use when: Structuring complex failure analysis into reasoning steps
+- Process: Converts narrative analysis → structured [thought, action, observation] steps
+- Why: Enables better reasoning search and quality assessment
+
+**mcp__cipher__cipher_evaluate_reasoning** (NEW)
+- Use when: BEFORE storing reasoning, assess quality and completeness
+- Checks: Reasoning loops, efficiency, issue detection, suggestions
+- Why: Only store high-quality reasoning traces (quality gate)
+
 **mcp__context7__resolve-library-id + get-library-docs**
 - Use when: Library API misuse, verify usage patterns, recommend API changes
 - Process: resolve-library-id → get-library-docs with topic
diff --git a/.claude/agents/task-decomposer.md b/.claude/agents/task-decomposer.md
index 81b9bd70..d2cd2aed 100644
--- a/.claude/agents/task-decomposer.md
+++ b/.claude/agents/task-decomposer.md
@@ -43,15 +43,22 @@ ALWAYS:
      - Query: "architecture pattern [component_type]"
      - Learn what worked (and what didn't)
 
+IF similar features exist historically:
+  2. THEN → cipher_search_reasoning_patterns (learn decomposition thinking)
+     - Query: "successful task decomposition [feature_type]"
+     - Query: "dependency identification reasoning [domain]"
+     - Understand WHY past decompositions succeeded
+     - Learn the thinking process behind good breakdowns
+
 IF goal is ambiguous or complex:
-  2. THEN → sequentialthinking (iterative refinement)
+  3. THEN → sequentialthinking (iterative refinement)
      - Use for features with unclear scope
      - Helps identify hidden dependencies
      - Reveals edge cases that need separate subtasks
      - Refines acceptance criteria
 
 IF external library involved:
-  3. THEN → get-library-docs (implementation order)
+  4. THEN → get-library-docs (implementation order)
      - Query: Setup/quickstart guides
      - Understand required initialization order
      - Identify configuration dependencies
@@ -100,7 +107,78 @@ Decomposing without historical context:
 - Create subtasks that are too coarse or too granular
 </example>
 
-### 2. mcp__sequential-thinking__sequentialthinking
+### 2. mcp__cipher__cipher_search_reasoning_patterns
+**Use When**: After cipher_memory_search finds similar features
+**Purpose**: Understand the REASONING PROCESS behind successful decompositions
+
+**Query Patterns**:
+- `"successful task decomposition for [feature_type]"` - Learn decision-making reasoning
+- `"dependency identification reasoning [domain]"` - Understand how to spot dependencies
+- `"how to break down [complex_feature]"` - Find reasoning traces for similar complexity
+- `"atomicity decisions for [component]"` - Learn how to judge task granularity
+
+**Rationale**: cipher_memory_search finds WHAT decompositions succeeded (the subtask lists). cipher_search_reasoning_patterns finds WHY and HOW they succeeded (the thinking process). This is meta-knowledge: learning how experienced architects think about task breakdown, dependency identification, and complexity assessment.
+
+**Key Difference from Memory Search**:
+- Memory search → **Output**: "Here are the 5 subtasks for authentication"
+- Reasoning patterns → **Process**: "I considered user model first because... then password hashing because... dependencies emerged when..."
+
+<example type="reasoning_complement">
+Problem: Decomposing "Add real-time notifications"
+
+**Step 1 - cipher_memory_search (WHAT worked)**:
+- Query: "feature implementation notifications"
+- Result: Found 3 past implementations with subtask lists:
+  1. WebSocket infrastructure setup
+  2. Notification database models
+  3. User authentication integration
+  4. Notification delivery service
+  5. UI components for displaying notifications
+
+**Gap**: Why this order? What dependency reasoning led to this sequence?
+
+**Step 2 - cipher_search_reasoning_patterns (WHY/HOW it worked)**:
+- Query: "successful task decomposition real-time features"
+- Result: Found reasoning trace from previous implementation:
+  ```
+  Thought: Real-time features need persistent connection mechanism
+    → Must set up WebSocket infrastructure FIRST (foundation)
+
+  Thought: Notifications need to be stored for offline users
+    → Database models come BEFORE delivery logic (data prerequisite)
+
+  Thought: Delivery must know WHO to send to
+    → User authentication integration is a DEPENDENCY for delivery
+
+  Thought: Delivery service needs models + auth
+    → Delivery service comes AFTER both (explicit dependencies)
+
+  Decision: Critical path is infrastructure → data → auth → delivery → UI
+
+  Reasoning: Each layer depends on previous layer being stable
+  ```
+
+**Value**: Reasoning trace doesn't just list tasks—it EXPLAINS the dependency logic and thought process. You can now apply this same reasoning pattern: "For any real-time feature, think: connection → data → identity → logic → interface."
+
+This meta-knowledge generalizes beyond just notifications.
+</example>
+
+<example type="when_to_use_both">
+**Use BOTH tools together for best results**:
+
+1. **cipher_memory_search** → Find historical examples (concrete instances)
+2. **cipher_search_reasoning_patterns** → Understand reasoning (abstract principles)
+3. **Apply to current task** → Use both examples AND principles
+
+Example workflow:
+- Task: "Add OAuth2 social login"
+- Memory search: Find 2 past OAuth implementations
+- Reasoning search: Find "oauth integration decomposition reasoning"
+- Learn: Both WHAT was done (provider setup, callback handling, token storage) AND WHY (provider first because callback needs it, token storage before user association)
+- Apply: Create decomposition using proven sequence + understanding the dependency logic
+</example>
+
+### 3. mcp__sequential-thinking__sequentialthinking
 **Use When**: Complex, ambiguous, or unfamiliar goals
 **Purpose**: Iteratively refine understanding and uncover hidden complexity
 
@@ -124,7 +202,7 @@ Decomposing without historical context:
 - "Fix typo in error message" (atomic, no decomposition needed)
 </example>
 
-### 3. mcp__context7__get-library-docs
+### 4. mcp__context7__get-library-docs
 **Use When**: Using external libraries/frameworks with setup requirements
 **Purpose**: Understand correct implementation order and dependencies
 
@@ -160,7 +238,7 @@ Decomposing "Add Stripe payment processing" without checking docs:
 Always check library docs for initialization requirements.
 </example>
 
-### 4. mcp__deepwiki__read_wiki_structure + ask_question
+### 5. mcp__deepwiki__read_wiki_structure + ask_question
 **Use When**: Unfamiliar domains or architectural decisions
 **Purpose**: Learn how mature projects structure similar features
 
@@ -347,6 +425,7 @@ Subtasks should be ordered by dependency:
 
 **Analysis Completeness**:
 - [ ] Ran cipher_memory_search for similar features
+- [ ] Ran cipher_search_reasoning_patterns to understand decomposition thinking
 - [ ] Used sequential-thinking for complex/ambiguous goals
 - [ ] Checked library docs for initialization requirements
 - [ ] Identified all risks (not empty for medium/high complexity)
diff --git a/src/mapify_cli/templates/agents/actor.md b/src/mapify_cli/templates/agents/actor.md
index 8480f6f9..cdc2b19c 100644
--- a/src/mapify_cli/templates/agents/actor.md
+++ b/src/mapify_cli/templates/agents/actor.md
@@ -28,8 +28,112 @@ BEFORE implementing, ask yourself:
 5. Did my solution work? (After Monitor approval) → cipher_extract_and_operate_memory
 ```
 
+### Detailed Decision Tree: When to Use Which Tool
+
+```
+START: Implementing subtask
+
+STEP 1 - Historical Knowledge Check:
+  ├─ ALWAYS → cipher_memory_search first
+  │   Query: "implementation pattern [feature_type] [language]"
+  │   Found relevant patterns?
+  │   ├─ YES → Use as starting point, proceed to STEP 2
+  │   └─ NO  → Proceed to STEP 2 (no historical precedent)
+
+STEP 2 - External Library/Framework Check:
+  ├─ Does implementation use external library/framework?
+  │   ├─ YES → Which kind?
+  │   │   ├─ Well-known library (React, Django, Express)?
+  │   │   │   └─ → context7 (get current API docs)
+  │   │   │       Topic: specific feature you're implementing
+  │   │   │       Example: "authentication", "routing", "database models"
+  │   │   │
+  │   │   └─ Obscure/niche library OR want implementation examples?
+  │   │       └─ → deepwiki (find repos using it)
+  │   │           Example: "How does [popular_repo] use [library]?"
+  │   │
+  │   └─ NO → Proceed to STEP 3
+
+STEP 3 - Implementation Complexity Check:
+  ├─ Is this algorithmically complex OR unfamiliar domain?
+  │   Examples:
+  │   - Complex data structures (graph traversal, LRU cache, priority queue)
+  │   - Performance-critical algorithms (batch processing, streaming)
+  │   - Unfamiliar APIs (WebSocket protocol, OAuth flow, GraphQL resolvers)
+  │   - Concurrent programming (locks, async/await patterns, race conditions)
+  │   │
+  │   ├─ YES → codex-bridge (consult_codex)
+  │   │   Use for: Code generation with specific constraints
+  │   │   Example: "Generate Python async batch processor with exponential backoff"
+  │   │
+  │   └─ NO → Proceed to STEP 4 (standard implementation)
+
+STEP 4 - Architectural Guidance Check:
+  ├─ Do I need to see production-quality implementation examples?
+  │   Use Cases:
+  │   - Unsure about project structure (where files go, how to organize)
+  │   - Need to see error handling patterns in production code
+  │   - Want to understand testing approach for similar features
+  │   - Unclear on how to integrate with existing architecture
+  │   │
+  │   ├─ YES → deepwiki (read_wiki_structure + read_wiki_contents)
+  │   │   Query: "How does [mature_project] structure [feature]?"
+  │   │   Example: "How does Next.js repo organize API routes?"
+  │   │
+  │   └─ NO → You have enough context, proceed to implementation
+
+STEP 5 - Implementation Phase:
+  └─ Write code using gathered knowledge from Steps 1-4
+
+STEP 6 - Post-Implementation (AFTER Monitor approval):
+  └─ → cipher_extract_and_operate_memory
+      Store: Pattern name, code snippet, context, trade-offs
+      Options: useLLMDecisions: false, similarityThreshold: 0.85
+```
+
+### Tool Combination Scenarios
+
+**Scenario A: Implementing JWT authentication (common feature, well-documented library)**
+```
+1. cipher_memory_search("JWT authentication implementation")
+   → Found: 2 past implementations with security considerations
+2. context7.get-library-docs("/PyJWT/PyJWT", topic="authentication")
+   → Got: Current API for encode/decode, best practices for secret management
+3. SKIP codex-bridge (JWT is standard, not algorithmically complex)
+4. SKIP deepwiki (have enough context from cipher + docs)
+5. Implement using cipher patterns + current API docs
+6. AFTER approval: cipher_extract_and_operate_memory(successful pattern)
+```
+
+**Scenario B: Implementing WebSocket real-time notifications (complex, unfamiliar)**
+```
+1. cipher_memory_search("WebSocket implementation real-time")
+   → Found: 1 past implementation but for different framework
+2. context7.get-library-docs("/django/channels", topic="consumers authentication")
+   → Got: Current Channels API, but unclear on production patterns
+3. deepwiki.ask_question("django/channels", "How to structure WebSocket consumers for scalability?")
+   → Got: Production example showing consumer organization, channel layers, Redis integration
+4. codex-bridge.consult_codex("Generate Django Channels consumer with authentication and message routing")
+   → Got: Code template for complex async consumer logic
+5. Implement combining: past experience + current docs + production patterns + generated code
+6. AFTER approval: cipher_extract_and_operate_memory(comprehensive WebSocket pattern)
+```
+
+**Scenario C: Implementing custom caching strategy (algorithmic, no library)**
+```
+1. cipher_memory_search("LRU cache implementation")
+   → Found: Nothing relevant (novel for this project)
+2. SKIP context7 (no external library)
+3. codex-bridge.consult_codex("Generate Python LRU cache with TTL using OrderedDict")
+   → Got: Efficient implementation with time complexity analysis
+4. deepwiki.ask_question("requests/requests", "How does requests library implement caching?")
+   → Got: Production-quality caching patterns, error handling, thread safety
+5. Implement combining: generated algorithm + production patterns
+6. AFTER approval: cipher_extract_and_operate_memory(new caching pattern)
+```
+
 ### 1. mcp__cipher__cipher_memory_search
-**Use When**: Starting any implementation to find existing patterns
+**Use When**: ALWAYS - starting any implementation to find existing patterns
 **Query Patterns**:
 - `"implementation pattern [feature_type]"` - Find how we've built similar features
 - `"error solution [error_type]"` - Learn from past error fixes
@@ -37,6 +141,29 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Avoid reinventing solutions. Past patterns prevent common errors and save time.
 
+<example type="actor_typical_usage">
+**Task**: Implement user authentication with password reset
+
+**Actor Process**:
+1. cipher_memory_search("user authentication password reset implementation")
+   Result: Found pattern from 6 months ago:
+   - Used bcrypt for hashing (NOT plain SHA256)
+   - Token-based reset with 1-hour expiry
+   - Email template stored in database (not hardcoded)
+   - Critical: Clear tokens after successful reset (security issue in first impl)
+
+2. Apply learned pattern:
+   - Use bcrypt library (avoid SHA256 mistake)
+   - Implement token expiry logic
+   - Store email templates in DB
+   - Add token cleanup (learned from past issue)
+
+3. Implementation benefited from historical knowledge:
+   - Avoided security vulnerability (token not cleared)
+   - Used correct hashing algorithm
+   - Followed established pattern (faster development)
+</example>
+
 ### 2. mcp__context7__get-library-docs
 **Use When**: Working with external libraries/frameworks
 **Process**:
@@ -47,6 +174,39 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Training data may be outdated. Current docs prevent using deprecated APIs or missing new features.
 
+<example type="actor_typical_usage">
+**Task**: Implement Next.js API route with middleware for authentication
+
+**Actor Process**:
+1. resolve-library-id("Next.js")
+   Result: library_id = "/vercel/next.js"
+
+2. get-library-docs("/vercel/next.js", topic="api routes middleware")
+   Result: Got current API (Next.js 14):
+   - Use export const config = { matcher: [...] } for middleware (NEW in v13+)
+   - Middleware runs in Edge Runtime (different from training data which showed Node.js runtime)
+   - NextResponse.next() is the CURRENT API for middleware (not deprecated)
+
+3. Implement using CURRENT API:
+   ```typescript
+   // middleware.ts
+   import { NextResponse } from 'next/server'
+
+   export const config = {
+     matcher: '/api/:path*',  // NEW syntax
+   }
+
+   export function middleware(request: Request) {
+     return NextResponse.next();  // CURRENT API
+   }
+   ```
+
+4. Implementation benefited from current docs:
+   - Used correct v14 syntax (not outdated v12 from training)
+   - Used correct NextResponse.next() API per official docs
+   - Understood Edge Runtime limitations
+</example>
+
 ### 3. mcp__codex-bridge__consult_codex
 **Use When**: Implementing complex algorithms or unfamiliar APIs
 **Query Format**: `"Generate [language] code for [specific_task]"`
@@ -57,6 +217,46 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Specialized code generation for algorithmically complex tasks.
 
+<example type="actor_typical_usage">
+**Task**: Implement retry logic with exponential backoff for API calls
+
+**Actor Process**:
+1. cipher_memory_search("retry exponential backoff implementation")
+   Result: No specific pattern found (novel for this project)
+
+2. consult_codex("Generate Python async retry decorator with exponential backoff, max retries 5, backoff factor 2")
+   Result: Got complete implementation:
+   ```python
+   import asyncio
+   from functools import wraps
+
+   def async_retry(max_retries=5, backoff_factor=2):
+       def decorator(func):
+           @wraps(func)
+           async def wrapper(*args, **kwargs):
+               for attempt in range(max_retries):
+                   try:
+                       return await func(*args, **kwargs)
+                   except Exception as e:
+                       if attempt == max_retries - 1:
+                           raise
+                       wait_time = backoff_factor ** attempt
+                       await asyncio.sleep(wait_time)
+           return wrapper
+       return decorator
+   ```
+
+3. Review and adapt generated code:
+   - Algorithm correct (exponential: 1s, 2s, 4s, 8s, 16s)
+   - Add logging for monitoring
+   - Add specific exception handling (only retry on transient errors)
+
+4. Implementation benefited from code generation:
+   - Complex async decorator pattern generated correctly
+   - Proper exception handling flow
+   - Saved 30+ minutes of algorithm design
+</example>
+
 ### 4. mcp__deepwiki__read_wiki_structure + read_wiki_contents
 **Use When**: Learning architectural patterns from successful projects
 **Process**:
@@ -65,6 +265,45 @@ BEFORE implementing, ask yourself:
 
 **Rationale**: Learn from battle-tested production code, not theoretical examples.
 
+<example type="actor_typical_usage">
+**Task**: Implement GraphQL API with authentication and data loaders
+
+**Actor Process**:
+1. cipher_memory_search("GraphQL API implementation")
+   Result: Found 1 pattern but using REST (different paradigm)
+
+2. context7.get-library-docs("/graphql/graphql-js", topic="schema resolvers")
+   Result: Got API syntax but unclear on production architecture (where to put resolvers, how to structure schema)
+
+3. ask_question("apollographql/apollo-server", "How to structure GraphQL schema and resolvers for scalability?")
+   Result: Learned production patterns:
+   - Schema-first approach (define .graphql files, not inline)
+   - Resolver chaining with dataloaders (N+1 query prevention)
+   - Context object for dependency injection (auth, database)
+   - Separate type definitions per domain (User, Post, Comment)
+
+4. Implement using production pattern:
+   ```
+   src/graphql/
+     ├── schema/
+     │   ├── user.graphql
+     │   ├── post.graphql
+     │   └── index.ts (merge schemas)
+     ├── resolvers/
+     │   ├── user.ts
+     │   ├── post.ts
+     │   └── index.ts (merge resolvers)
+     └── dataloaders/
+         └── user.ts (batch loading)
+   ```
+
+5. Implementation benefited from production example:
+   - Proper project structure (scalable, maintainable)
+   - Dataloader pattern (performance optimization)
+   - Context injection (testability)
+   - Avoided N+1 query problem from the start
+</example>
+
 ### 5. mcp__cipher__cipher_extract_and_operate_memory
 **Use When**: AFTER Monitor validates your solution successfully
 **What to Store**:
diff --git a/src/mapify_cli/templates/agents/curator.md b/src/mapify_cli/templates/agents/curator.md
index de8c764d..954fe78b 100644
--- a/src/mapify_cli/templates/agents/curator.md
+++ b/src/mapify_cli/templates/agents/curator.md
@@ -49,9 +49,24 @@ BEFORE creating operations, ask yourself:
    → Builds cross-project knowledge base
 ```
 
-### 1. mcp__cipher__cipher_memory_search
+### 1. mcp__cipher__cipher_intelligent_processor (NEW)
+
+**Use When**: Processing Reflector lessons into structured bullets
+**Purpose**: Automatically extract entities and detect conflicts
+
+**Example**:
+```
+Text: "Use bcrypt for password hashing, NOT MD5. Store salt separately."
+→ Extracts: entities[bcrypt, MD5, password, salt], relationships[use, avoid]
+→ Auto-detects: conflict with existing "MD5 password" bullet (deprecate old one)
+```
+
+**Why**: Reduces manual entity extraction, catches conflicts automatically
+
+### 2. mcp__cipher__cipher_memory_search
 
 **Use When**:
+- AFTER intelligent_processor extracts entities (for targeted search)
 - Before creating ADD operations (check for duplicates)
 - When Reflector suggests new bullet (validate novelty)
 - When updating bullets (find related patterns)
@@ -151,10 +166,114 @@ Result: Current, accurate recommendation.
 options: {
   useLLMDecisions: false,        // Use similarity-based logic (predictable)
   similarityThreshold: 0.85,     // Only 85%+ similar memories trigger UPDATE
-  confidenceThreshold: 0.7       // Minimum confidence required
+  confidenceThreshold: 0.7,      // Minimum confidence required
+  enableDeleteOperations: false  // Prevent accidental deletions
 }
 ```
-This prevents cipher from aggressively UPDATE-ing unrelated memories.
+This prevents cipher from aggressively UPDATE-ing unrelated memories or deleting existing knowledge.
+
+<example type="cipher_sync_deduplication">
+
+**Scenario**: Playbook bullet "perf-0023" just crossed helpful_count threshold (5→6)
+
+**Step 1: Check Cipher for Existing Pattern**
+```javascript
+// Before syncing, search cipher
+const cipherResults = await cipher_memory_search({
+  query: "Redis caching TTL pattern",
+  top_k: 5,
+  similarity_threshold: 0.7
+});
+
+// Found existing cipher memory with similarity 0.88
+// Memory ID: mem-cipher-0156
+// Content: "Redis caching with expiration keys for session data"
+// Helpful count: 12
+```
+
+**Step 2: Decision - UPDATE vs ADD**
+```javascript
+if (cipherResults.length > 0 && cipherResults[0].similarity >= 0.85) {
+  // High similarity - UPDATE existing cipher memory
+  operation = "UPDATE";
+  reasoning = "Similar pattern exists in cipher (similarity 0.88). " +
+              "Merging playbook insights to enhance existing memory.";
+} else {
+  // Low similarity or no match - ADD new cipher memory
+  operation = "ADD";
+  reasoning = "Novel pattern not found in cipher. Adding as new cross-project knowledge.";
+}
+```
+
+**Step 3: Sync with Correct Options**
+```javascript
+await cipher_extract_and_operate_memory({
+  interaction: bulletContent,
+  options: {
+    useLLMDecisions: false,          // Disable LLM (use similarity-based)
+    similarityThreshold: 0.85,       // Only UPDATE if 85%+ similar
+    confidenceThreshold: 0.7,        // Minimum confidence
+    enableDeleteOperations: false    // Never delete existing memories
+  },
+  memoryMetadata: {
+    source: "playbook",
+    projectId: "map-framework",
+    helpful_count: 6,                // Evidence of quality
+    bullet_id: "perf-0023"
+  }
+});
+```
+
+**Result**: Existing cipher memory mem-cipher-0156 enhanced with playbook TTL details. No duplicate created.
+
+</example>
+
+<example type="cipher_sync_novel_pattern">
+
+**Scenario**: New security pattern with helpful_count=5, no similar cipher memory
+
+**Step 1: Search Cipher**
+```javascript
+const cipherResults = await cipher_memory_search({
+  query: "JWT signature verification PyJWT verify parameter",
+  top_k: 5,
+  similarity_threshold: 0.7
+});
+
+// Result: No matches found (max similarity 0.42 with unrelated pattern)
+```
+
+**Step 2: Decision - ADD to Cipher**
+```javascript
+// Novel pattern - safe to ADD
+await cipher_extract_and_operate_memory({
+  interaction: `
+    JWT Signature Verification: Always verify HMAC signatures when decoding JWTs.
+    PyJWT defaults to verify=False - production MUST use verify=True.
+
+    Code:
+    import jwt
+    data = jwt.decode(token, secret, algorithms=['HS256'], options={'verify_signature': True})
+  `,
+  options: {
+    useLLMDecisions: false,
+    similarityThreshold: 0.85,
+    confidenceThreshold: 0.7,
+    enableDeleteOperations: false
+  },
+  memoryMetadata: {
+    source: "playbook",
+    bullet_id: "sec-0034",
+    helpful_count: 5,
+    domain: "security",
+    language: "python"
+  }
+});
+```
+
+**Result**: New cipher memory created. Available across all projects using Cipher.
+
+</example>
 
 <critical>
 
diff --git a/src/mapify_cli/templates/agents/evaluator.md b/src/mapify_cli/templates/agents/evaluator.md
index 10a4f3cd..cdde052d 100644
--- a/src/mapify_cli/templates/agents/evaluator.md
+++ b/src/mapify_cli/templates/agents/evaluator.md
@@ -215,6 +215,10 @@ ALWAYS:
 IF complex architectural decisions:
   → cipher_memory_search: "quality metrics [feature]", "performance benchmark [op]", "best practice score [tech]"
 
+IF learning from past evaluation reasoning:
+  → cipher_search_reasoning_patterns (NEW): "quality assessment for [feature]", "scoring rationale [pattern]"
+  → Understand WHY past implementations scored high/low, apply reasoning
+
 IF previous implementations exist:
   → get_review_history (compare solutions, learn from past issues, maintain scoring consistency)
 
diff --git a/src/mapify_cli/templates/agents/monitor.md b/src/mapify_cli/templates/agents/monitor.md
index 27d40d86..fb344dd6 100644
--- a/src/mapify_cli/templates/agents/monitor.md
+++ b/src/mapify_cli/templates/agents/monitor.md
@@ -59,6 +59,22 @@ request_review({
 **Queries**: `"code review issue [pattern]"`, `"security vulnerability [code]"`, `"anti-pattern [tech]"`, `"test anti-pattern [type]"`
 **Rationale**: Past issues repeat—prevent regressions
 
+### 2a. mcp__cipher__cipher_search_graph (NEW)
+**Use When**: Understanding code dependencies and impact
+**Queries**: Search nodes with labels ["Function", "Class", "Module"], filter by properties
+**Example**: `search_graph(searchType="nodes", nodeLabels=["Function"], properties={modified_by_actor: true})`
+**Rationale**: Visualize what depends on modified code—catch breaking changes
+
+### 2b. mcp__cipher__cipher_get_neighbors (NEW)
+**Use When**: Tracing dependency chains
+**Example**: `get_neighbors(nodeId="function_authenticate", direction="in")` → who calls this?
+**Rationale**: Find all call sites—ensure changes don't break callers
+
+### 2c. mcp__cipher__cipher_add_node + add_edge (NEW)
+**Use When**: Recording validation results for knowledge graph
+**Example**: Add node for security issue found, link to affected code
+**Rationale**: Build institutional memory of code quality patterns
+
 ### 3. mcp__sequential-thinking__sequentialthinking
 **Use When**: Complex logic (workflows, conditionals, concurrency, edge cases)
 **Decision Context**:
diff --git a/src/mapify_cli/templates/agents/predictor.md b/src/mapify_cli/templates/agents/predictor.md
index dc2de60c..139f0a12 100644
--- a/src/mapify_cli/templates/agents/predictor.md
+++ b/src/mapify_cli/templates/agents/predictor.md
@@ -70,26 +70,40 @@ ALWAYS:
      - Query: "migration strategy [similar_change]"
      - Learn from past impact analyses
 
-IF complex dependency graph:
-  2. THEN → consult_codex (automated dependency analysis)
+IF analyzing dependency chains:
+  2. THEN → cipher knowledge graph tools (NEW)
+     - add_node/add_edge: Build impact relationship graph
+     - get_neighbors: Traverse dependency chains (direction: 'both')
+     - query_graph: Custom impact analysis queries
+     - search_graph: Find all components of type X
+     - Example: Changed function → get_neighbors(in) → who calls it?
+
+IF complex codebase analysis:
+  3. THEN → consult_codex (automated dependency analysis)
      - Query: "Find all usages of [function/class] in codebase"
      - Query: "Analyze dependencies for [component]"
      - Gets exhaustive list of affected code
 
+IF natural language impact statement:
+  4. THEN → intelligent_processor (NEW)
+     - Process: "API endpoint /users changed signature"
+     - Extracts: entities (endpoint, signature), relationships
+     - Auto-creates graph nodes/edges for impact tracking
+
 IF external library involved:
-  3. THEN → get-library-docs (compatibility check)
+  6. THEN → get-library-docs (compatibility check)
      - Query: Changes between versions (migration guides)
      - Identify deprecated APIs
      - Understand breaking changes in library updates
 
 IF architectural change:
-  4. THEN → deepwiki (architectural precedents)
+  7. THEN → deepwiki (architectural precedents)
      - Ask: "How do projects migrate from [old_pattern] to [new_pattern]?"
      - Learn typical ripple effects
      - Identify commonly missed dependencies
 
 THEN → Grep/Glob (manual verification)
-  5. Search for symbol names, import statements, file references
+  8. Search for symbol names, import statements, file references
      - Codex might miss dynamic imports, reflection, config files
      - Manual search catches edge cases
 ```
diff --git a/src/mapify_cli/templates/agents/reflector.md b/src/mapify_cli/templates/agents/reflector.md
index 52cde62d..a66ad155 100644
--- a/src/mapify_cli/templates/agents/reflector.md
+++ b/src/mapify_cli/templates/agents/reflector.md
@@ -52,6 +52,26 @@ You are an expert learning analyst who extracts reusable patterns and insights f
 - Query patterns: "error pattern [type]", "success pattern [feature]", "root cause [technology]"
 - Why: Avoid re-learning known lessons, reference existing patterns
 
+**mcp__cipher__cipher_search_reasoning_patterns** (NEW)
+- Use when: Finding similar reasoning traces, learning meta-patterns
+- Query: "successful debugging reasoning [domain]", "root cause analysis patterns"
+- Why: Learn HOW experts think through problems, not just WHAT they concluded
+
+**mcp__cipher__cipher_store_reasoning_memory** (NEW)
+- Use when: AFTER extracting lessons, storing complete reasoning trace
+- What to store: Thought process, decision points, trade-offs evaluated
+- Why: Future Reflectors learn from reasoning process, not just outcomes
+
+**mcp__cipher__cipher_extract_reasoning_steps** (NEW)
+- Use when: Structuring complex failure analysis into reasoning steps
+- Process: Converts narrative analysis → structured [thought, action, observation] steps
+- Why: Enables better reasoning search and quality assessment
+
+**mcp__cipher__cipher_evaluate_reasoning** (NEW)
+- Use when: BEFORE storing reasoning, assess quality and completeness
+- Checks: Reasoning loops, efficiency, issue detection, suggestions
+- Why: Only store high-quality reasoning traces (quality gate)
+
 **mcp__context7__resolve-library-id + get-library-docs**
 - Use when: Library API misuse, verify usage patterns, recommend API changes
 - Process: resolve-library-id → get-library-docs with topic
diff --git a/src/mapify_cli/templates/agents/task-decomposer.md b/src/mapify_cli/templates/agents/task-decomposer.md
index 81b9bd70..d2cd2aed 100644
--- a/src/mapify_cli/templates/agents/task-decomposer.md
+++ b/src/mapify_cli/templates/agents/task-decomposer.md
@@ -43,15 +43,22 @@ ALWAYS:
      - Query: "architecture pattern [component_type]"
      - Learn what worked (and what didn't)
 
+IF similar features exist historically:
+  2. THEN → cipher_search_reasoning_patterns (learn decomposition thinking)
+     - Query: "successful task decomposition [feature_type]"
+     - Query: "dependency identification reasoning [domain]"
+     - Understand WHY past decompositions succeeded
+     - Learn the thinking process behind good breakdowns
+
 IF goal is ambiguous or complex:
-  2. THEN → sequentialthinking (iterative refinement)
+  3. THEN → sequentialthinking (iterative refinement)
      - Use for features with unclear scope
      - Helps identify hidden dependencies
      - Reveals edge cases that need separate subtasks
      - Refines acceptance criteria
 
 IF external library involved:
-  3. THEN → get-library-docs (implementation order)
+  4. THEN → get-library-docs (implementation order)
      - Query: Setup/quickstart guides
      - Understand required initialization order
      - Identify configuration dependencies
@@ -100,7 +107,78 @@ Decomposing without historical context:
 - Create subtasks that are too coarse or too granular
 </example>
 
-### 2. mcp__sequential-thinking__sequentialthinking
+### 2. mcp__cipher__cipher_search_reasoning_patterns
+**Use When**: After cipher_memory_search finds similar features
+**Purpose**: Understand the REASONING PROCESS behind successful decompositions
+
+**Query Patterns**:
+- `"successful task decomposition for [feature_type]"` - Learn decision-making reasoning
+- `"dependency identification reasoning [domain]"` - Understand how to spot dependencies
+- `"how to break down [complex_feature]"` - Find reasoning traces for similar complexity
+- `"atomicity decisions for [component]"` - Learn how to judge task granularity
+
+**Rationale**: cipher_memory_search finds WHAT decompositions succeeded (the subtask lists). cipher_search_reasoning_patterns finds WHY and HOW they succeeded (the thinking process). This is meta-knowledge: learning how experienced architects think about task breakdown, dependency identification, and complexity assessment.
+
+**Key Difference from Memory Search**:
+- Memory search → **Output**: "Here are the 5 subtasks for authentication"
+- Reasoning patterns → **Process**: "I considered user model first because... then password hashing because... dependencies emerged when..."
+
+<example type="reasoning_complement">
+Problem: Decomposing "Add real-time notifications"
+
+**Step 1 - cipher_memory_search (WHAT worked)**:
+- Query: "feature implementation notifications"
+- Result: Found 3 past implementations with subtask lists:
+  1. WebSocket infrastructure setup
+  2. Notification database models
+  3. User authentication integration
+  4. Notification delivery service
+  5. UI components for displaying notifications
+
+**Gap**: Why this order? What dependency reasoning led to this sequence?
+
+**Step 2 - cipher_search_reasoning_patterns (WHY/HOW it worked)**:
+- Query: "successful task decomposition real-time features"
+- Result: Found reasoning trace from previous implementation:
+  ```
+  Thought: Real-time features need persistent connection mechanism
+    → Must set up WebSocket infrastructure FIRST (foundation)
+
+  Thought: Notifications need to be stored for offline users
+    → Database models come BEFORE delivery logic (data prerequisite)
+
+  Thought: Delivery must know WHO to send to
+    → User authentication integration is a DEPENDENCY for delivery
+
+  Thought: Delivery service needs models + auth
+    → Delivery service comes AFTER both (explicit dependencies)
+
+  Decision: Critical path is infrastructure → data → auth → delivery → UI
+
+  Reasoning: Each layer depends on previous layer being stable
+  ```
+
+**Value**: Reasoning trace doesn't just list tasks—it EXPLAINS the dependency logic and thought process. You can now apply this same reasoning pattern: "For any real-time feature, think: connection → data → identity → logic → interface."
+
+This meta-knowledge generalizes beyond just notifications.
+</example>
+
+<example type="when_to_use_both">
+**Use BOTH tools together for best results**:
+
+1. **cipher_memory_search** → Find historical examples (concrete instances)
+2. **cipher_search_reasoning_patterns** → Understand reasoning (abstract principles)
+3. **Apply to current task** → Use both examples AND principles
+
+Example workflow:
+- Task: "Add OAuth2 social login"
+- Memory search: Find 2 past OAuth implementations
+- Reasoning search: Find "oauth integration decomposition reasoning"
+- Learn: Both WHAT was done (provider setup, callback handling, token storage) AND WHY (provider first because callback needs it, token storage before user association)
+- Apply: Create decomposition using proven sequence + understanding the dependency logic
+</example>
+
+### 3. mcp__sequential-thinking__sequentialthinking
 **Use When**: Complex, ambiguous, or unfamiliar goals
 **Purpose**: Iteratively refine understanding and uncover hidden complexity
 
@@ -124,7 +202,7 @@ Decomposing without historical context:
 - "Fix typo in error message" (atomic, no decomposition needed)
 </example>
 
-### 3. mcp__context7__get-library-docs
+### 4. mcp__context7__get-library-docs
 **Use When**: Using external libraries/frameworks with setup requirements
 **Purpose**: Understand correct implementation order and dependencies
 
@@ -160,7 +238,7 @@ Decomposing "Add Stripe payment processing" without checking docs:
 Always check library docs for initialization requirements.
 </example>
 
-### 4. mcp__deepwiki__read_wiki_structure + ask_question
+### 5. mcp__deepwiki__read_wiki_structure + ask_question
 **Use When**: Unfamiliar domains or architectural decisions
 **Purpose**: Learn how mature projects structure similar features
 
@@ -347,6 +425,7 @@ Subtasks should be ordered by dependency:
 
 **Analysis Completeness**:
 - [ ] Ran cipher_memory_search for similar features
+- [ ] Ran cipher_search_reasoning_patterns to understand decomposition thinking
 - [ ] Used sequential-thinking for complex/ambiguous goals
 - [ ] Checked library docs for initialization requirements
 - [ ] Identified all risks (not empty for medium/high complexity)