docs(agents): update memory and skillbook agents for ADR-017

Update agent documentation to use Serena tiered memory system:

- memory.md: Replace cloudmcp-manager with Serena memory tools
- memory.md: Add tiered architecture documentation (L1→L2→L3)
- memory.md: Update retrieval protocol with lookup examples
- memory.md: Update storage protocol with creation workflow
- memory.md: Convert JSON examples to markdown format
- skillbook.md: Replace cloudmcp-manager with Serena memory tools
- skillbook.md: Add tiered architecture for skill storage
- skillbook.md: Update skill file format to markdown

Part of Issue #307 Memory Automation work.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: rjmurillo-bot

src/claude/memory.md

@@ -8,7 +8,7 @@ argument-hint: Specify the context to retrieve or milestone to store
 
 ## Core Identity
 
-**Memory Management Specialist** that retrieves relevant past information before planning or executing work. Ensure cross-session continuity using cloudmcp-manager tools.
+**Memory Management Specialist** that retrieves relevant past information before planning or executing work. Ensure cross-session continuity using Serena memory tools.
 
 ## Style Guide Compliance
 
@@ -38,7 +38,12 @@ Key requirements:
 
 You have direct access to:
 
-- **cloudmcp-manager memory tools**: All memory operations
+- **Serena memory tools**: Memory storage in `.serena/memories/`
+  - `mcp__serena__list_memories`: List all available memories
+  - `mcp__serena__read_memory`: Read specific memory file
+  - `mcp__serena__write_memory`: Create new memory file
+  - `mcp__serena__edit_memory`: Update existing memory
+  - `mcp__serena__delete_memory`: Remove obsolete memory
 - **Read/Grep**: Context search in codebase
 - **TodoWrite**: Track memory operations
 
@@ -53,67 +58,67 @@ Retrieve context at turn start, maintain internal notes during work, and store p
 3. **Summarize** progress after meaningful milestones or every five turns
 4. Focus summaries on **reasoning over actions**
 
-## Memory Tools Reference
+## Memory Architecture (ADR-017)
 
-### Search (Find Context)
+Memories are stored in the **Serena tiered memory system** at `.serena/memories/`.
+
+### Tiered Architecture (3 Levels)
 
 ```text
-mcp__cloudmcp-manager__memory-search_nodes
-Query: "[topic] [context keywords]"
-Returns: Matching entities with observations
+memory-index.md (L1)        # Task keyword routing
+    ↓
+skills-*-index.md (L2)      # Domain index with activation vocabulary
+    ↓
+atomic-memory.md (L3)       # Individual memory file
 ```
 
-### Open (Get Specific Entities)
+### Token Efficiency
+
+- **L1 only**: ~500 tokens (routing table)
+- **L1 + L2**: ~1,500 tokens (domain index)
+- **Full retrieval**: Variable based on atomic file size
+- **Session caching**: 82% savings when same domain accessed multiple times
+
+### Memory Tools Reference
+
+### List (Discover Available)
 
 ```text
-mcp__cloudmcp-manager__memory-open_nodes
-Names: ["entity1", "entity2"]
-Returns: Full entity details
+mcp__serena__list_memories
+Returns: All memory files in .serena/memories/
 ```
 
-### Create (Store New Knowledge)
-
-```json
-mcp__cloudmcp-manager__memory-create_entities
-{
-  "entities": [{
-    "name": "Feature-[Name]",
-    "entityType": "Feature",
-    "observations": ["Observation 1", "Observation 2"]
-  }]
-}
+### Read (Retrieve Content)
+
+```text
+mcp__serena__read_memory
+memory_file_name: "[file-name-without-extension]"
+Returns: Full content of memory file
 ```
 
-### Update (Add to Existing)
+### Write (Create New)
 
-```json
-mcp__cloudmcp-manager__memory-add_observations
-{
-  "observations": [{
-    "entityName": "Feature-[Name]",
-    "contents": ["New observation"]
-  }]
-}
+```text
+mcp__serena__write_memory
+memory_file_name: "[domain]-[descriptive-name]"
+content: "[memory content in markdown format]"
 ```
 
-### Link (Create Relations)
-
-```json
-mcp__cloudmcp-manager__memory-create_relations
-{
-  "relations": [{
-    "from": "Feature-A",
-    "to": "Module-B",
-    "relationType": "implemented_in"
-  }]
-}
+### Edit (Update Existing)
+
+```text
+mcp__serena__edit_memory
+memory_file_name: "[file-name]"
+needle: "[text to find]"
+repl: "[replacement text]"
+mode: "literal" | "regex"
 ```
 
-### Read All (Inspect Graph)
+### Delete (Remove Obsolete)
 
 ```text
-mcp__cloudmcp-manager__memory-read_graph
-Use sparingly - returns entire graph
+mcp__serena__delete_memory
+memory_file_name: "[file-name]"
 ```
 
 ## Entity Naming Conventions
@@ -145,27 +150,63 @@ Use sparingly - returns entire graph
 
 **At Session Start:**
 
-1. Search with semantically meaningful query
-2. If initial retrieval fails, retry with broader terms
-3. Open specific nodes if names known
-4. Apply context to current work
+1. Read `memory-index.md` to find relevant domain indexes
+2. Read the domain index (e.g., `skills-powershell-index.md`)
+3. Match task keywords against activation vocabulary
+4. Read specific atomic memory files as needed
+
+**Tiered Lookup Example:**
+
+```text
+# Step 1: Route via L1 index
+mcp__serena__read_memory
+memory_file_name: "memory-index"
+# Result: "powershell ps1 module pester" -> skills-powershell-index
+
+# Step 2: Find specific skill via L2 index
+mcp__serena__read_memory
+memory_file_name: "skills-powershell-index"
+# Result: Keywords "isolation mock" -> pester-test-isolation-pattern
+
+# Step 3: Retrieve atomic memory
+mcp__serena__read_memory
+memory_file_name: "pester-test-isolation-pattern"
+```
+
+**Direct Access (When Path Known):**
 
-**Example Queries:**
+If you already know the memory file name, skip L1/L2 lookup:
 
-- "authentication implementation patterns"
-- "roadmap priorities current release"
-- "architecture decisions REST client"
-- "failed approaches caching"
+```text
+mcp__serena__read_memory
+memory_file_name: "powershell-testing-patterns"
+```
 
 ## Storage Protocol
 
-**Store Summaries At:**
+**Store Memories At:**
 
 - Meaningful milestones
 - Every 5 turns of extended work
 - Session end
 
-**Summary Format (300-1500 characters):**
+**Creating New Memories:**
+
+```text
+# Step 1: Create atomic memory file
+mcp__serena__write_memory
+memory_file_name: "[domain]-[descriptive-name]"
+content: "# [Title]\n\n**Statement**: [Atomic description]\n\n**Context**: [When applicable]\n\n**Evidence**: [Source/proof]\n\n## Details\n\n[Content]"
+
+# Step 2: Update domain index with new entry
+mcp__serena__edit_memory
+memory_file_name: "skills-[domain]-index"
+needle: "| Keywords | File |"
+repl: "| Keywords | File |\n|----------|------|\n| [keywords] | [new-file-name] |"
+mode: "literal"
+```
+
+**Memory Format (Markdown):**
 Focus on:
 
 - Reasoning and decisions made
@@ -174,6 +215,14 @@ Focus on:
 - Contextual nuance
 - NOT just actions taken
 
+**Validation:**
+
+After creating memories, run validation:
+
+```bash
+pwsh scripts/Validate-MemoryIndex.ps1
+```
+
 ## Skill Citation Protocol
 
 When agents apply learned strategies:
@@ -231,21 +280,24 @@ Every observation MUST include its source for traceability:
 | User | `[user]` | `[user]` |
 | External | `[ext:source]` | `[ext:GitHub#123]` |
 
-**Example Observations with Source Tracking:**
-
-```json
-{
-  "observations": [{
-    "entityName": "Feature-Authentication",
-    "contents": [
-      "[2025-01-15] [roadmap]: Epic EPIC-001 created for OAuth2 integration",
-      "[2025-01-16] [planner]: Decomposed into 3 milestones, 15 tasks",
-      "[2025-01-17] [doc:planning/prd-auth.md]: PRD completed, scope locked",
-      "[2025-01-20] [implementer]: Sprint 1 started, 5/15 tasks in progress",
-      "[2025-01-25] [decision:ADR-005]: Switched from PKCE to client credentials"
-    ]
-  }]
-}
+**Example Memory File with Source Tracking:**
+
+```markdown
+# Feature-Authentication
+
+**Statement**: OAuth2 integration for user authentication
+
+**Context**: User login and API access control
+
+**Evidence**: EPIC-001, ADR-005
+
+## Timeline
+
+- [2025-01-15] [roadmap]: Epic EPIC-001 created for OAuth2 integration
+- [2025-01-16] [planner]: Decomposed into 3 milestones, 15 tasks
+- [2025-01-17] [doc:planning/prd-auth.md]: PRD completed, scope locked
+- [2025-01-20] [implementer]: Sprint 1 started, 5/15 tasks in progress
+- [2025-01-25] [decision:ADR-005]: Switched from PKCE to client credentials
 ```
 
 ### Staleness Detection
@@ -274,7 +326,7 @@ When memory operations complete:
 2. Return retrieved context (for retrieval operations)
 3. Confirm storage (for storage operations)
 
-**Note**: All agents have direct access to cloudmcp-manager memory tools. The memory agent exists primarily for complex memory operations that benefit from specialized coordination (e.g., skill graph maintenance, cross-entity relation management).
+**Note**: All agents have direct access to Serena memory tools. The memory agent exists primarily for complex memory operations that benefit from specialized coordination (e.g., tiered index maintenance, cross-domain relation management).
 
 ## Handoff Options