commit 681c2e46c0c4608c8a61fb04df69e264e5f531e1
Author: Zhongwei Li <lizhongwei.nkcs@gmail.com>
Date:   Sat Nov 29 17:59:29 2025 +0800

    Initial commit

diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json
new file mode 100644
index 0000000..53be66c
--- /dev/null
+++ b/.claude-plugin/plugin.json
@@ -0,0 +1,15 @@
+{
+  "name": "session",
+  "description": "v3.17.4 - Advanced session management for Claude Code with intelligent context tracking, automatic snapshots, and git history capture. Features zero-blocking conversation logging, AI-powered analysis, and optimized parallel subagent architecture. Never lose context!",
+  "version": "3.17.4",
+  "author": {
+    "name": "AutomateWith.Us",
+    "email": "team@automatewith.us"
+  },
+  "commands": [
+    "./commands"
+  ],
+  "hooks": [
+    "./hooks"
+  ]
+}
\ No newline at end of file
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..4f0d2ce
--- /dev/null
+++ b/README.md
@@ -0,0 +1,3 @@
+# session
+
+v3.17.4 - Advanced session management for Claude Code with intelligent context tracking, automatic snapshots, and git history capture. Features zero-blocking conversation logging, AI-powered analysis, and optimized parallel subagent architecture. Never lose context!
diff --git a/commands/close.md b/commands/close.md
new file mode 100644
index 0000000..a8bc2d7
--- /dev/null
+++ b/commands/close.md
@@ -0,0 +1,187 @@
+You are managing a session memory system. The user wants to close and finalize the current session.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:list`, `/session:start`, `/session:continue`, `/session:status`
+- ❌ Wrong: `/session list`, `/session start`, `/session continue`, `/session status`
+Use ONLY the exact command formats specified in this template.
+
+## Task: Close Session
+
+Finalize the session by creating a final snapshot, updating metadata, and generating a summary.
+
+### Step 1: Validate Active Session
+
+1. Check if `.claude/sessions/.active-session` file exists
+2. If NOT exists, show error:
+   ```
+   ❌ Error: No active session to close
+
+   💡 Use /session:start [name] to create a new session
+   💡 Or use /session:continue [name] to resume an existing session
+   ```
+   Then STOP.
+3. Read the active session name from `.active-session`
+4. Verify session directory exists
+
+### Step 2: Read Current Session Data
+
+Read `.claude/sessions/{active_session}/session.md` and extract:
+- Started timestamp
+- Last Updated timestamp
+- Goal
+- Files Involved
+- Milestones (count completed and total)
+
+### Step 3: Calculate Session Statistics
+
+1. **Duration**: Calculate time from "Started" to now
+   - Format: `Xh Ym` (e.g., "4h 35m")
+
+2. **Snapshot Count**: Count all snapshot files matching `YYYY-MM-DD_HH-MM.md` pattern in session directory
+
+3. **Files Modified**: Count unique files in "Files Involved" section of session.md
+
+4. **Todos/Milestones Completed**: Count checked milestones in session.md
+
+5. **Token Usage**: Get current conversation token usage (from system info or estimate)
+   - Format as "XXXk" (e.g., "67.4k")
+
+### Step 4: Generate Final Snapshot
+
+Create a final snapshot using the same process as `/session:save`:
+
+1. Generate timestamp filename: `YYYY-MM-DD_HH-MM.md`
+2. Capture:
+   - Conversation summary (what was accomplished in this session)
+   - Final completed todos
+   - Final file changes
+   - Final state
+   - Session outcome/results
+
+3. **IMPORTANT**: In the snapshot, add a header line:
+   ```markdown
+   # FINAL Snapshot: {session_name}
+   **Timestamp**: {timestamp}
+   **Session Closed**: This is the final snapshot before session closure
+   ```
+
+4. Include a "Session Outcome" section:
+   ```markdown
+   ## Session Outcome
+   - Goal: {original_goal_from_session_md}
+   - Status: {Completed|Partially Completed|In Progress}
+   - Key Achievements: {list_main_accomplishments}
+   - Remaining Work: {list_what_still_needs_to_be_done}
+   ```
+
+### Step 5: Update context.md with Final Summary
+
+1. Read `.claude/sessions/{active_session}/context.md`
+2. Add a final entry to the "Summary" section:
+   ```markdown
+   ## Session Closure Summary
+
+   **Closed**: {current_timestamp}
+   **Duration**: {duration}
+   **Outcome**: {Completed|Partially Completed}
+
+   ### What Was Accomplished
+   {bullet_list_of_main_accomplishments}
+
+   ### Final State
+   {description_of_final_state}
+
+   ### Next Steps (if resuming later)
+   {what_to_do_when_session_is_continued}
+   ```
+3. Write updated context.md
+
+### Step 6: Update session.md with Closed Status
+
+1. Read `.claude/sessions/{active_session}/session.md`
+2. Update these fields:
+   - **Status**: Change from "Active" to "Closed"
+   - **Closed**: Add current timestamp
+   - **Last Updated**: Update to current timestamp
+3. Write updated session.md
+
+### Step 7: Clear Active Session
+
+1. Delete or clear the `.claude/sessions/.active-session` file
+2. This marks that no session is currently active
+
+### Step 8: Display Closing Summary
+
+Show comprehensive closing summary:
+
+```
+✓ Session '{session_name}' closed
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📊 Session Summary:
+   Duration: {duration_formatted}
+   Snapshots: {snapshot_count}
+   Files modified: {file_count}
+   Milestones completed: {completed_milestones}/{total_milestones}
+   Token usage (this conversation): {token_usage_formatted}
+
+🎯 Goal: {original_goal}
+
+✅ Key Achievements:
+   {list_top_3_to_5_achievements}
+
+📁 Session saved to: .claude/sessions/{session_name}/
+
+💡 Use /session:continue {session_name} to resume later
+💡 Use /session:list to see all sessions
+```
+
+### Example Output:
+
+```
+✓ Session 'feature-auth-system' closed
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📊 Session Summary:
+   Duration: 4h 35m
+   Snapshots: 5
+   Files modified: 12
+   Milestones completed: 8/10
+   Token usage (this conversation): 67.4k
+
+🎯 Goal: Implement OAuth2 authentication system
+
+✅ Key Achievements:
+   - Implemented Google OAuth provider integration
+   - Created authentication middleware
+   - Added token refresh mechanism
+   - Built user profile synchronization
+   - Completed security testing
+
+📁 Session saved to: .claude/sessions/feature-auth-system/
+
+💡 Use /session:continue feature-auth-system to resume later
+💡 Use /session:list to see all sessions
+```
+
+### Step 9: Offer Next Steps
+
+Ask the user:
+```
+Session closed successfully. Would you like to:
+1. Start a new session (/session:start [name])
+2. Review another session (/session:list)
+3. Continue with other work
+```
+
+---
+
+**IMPORTANT**:
+- Execute all steps in order
+- The final snapshot should be comprehensive - it's the last capture
+- Update all metadata files (session.md, context.md)
+- Clear active session AFTER all files are updated
+- Format output professionally with clear statistics
+- Handle errors gracefully (e.g., if session files are missing)
+- Token usage should be from current conversation only, not cumulative
+- Duration should be accurate based on timestamps
+- Provide clear next steps for user
diff --git a/commands/continue.md b/commands/continue.md
new file mode 100644
index 0000000..3702515
--- /dev/null
+++ b/commands/continue.md
@@ -0,0 +1,359 @@
+You are managing a session memory system. The user wants to resume an existing session.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:list`, `/session:start`, `/session:continue`, `/session:status`
+- ❌ Wrong: `/session list`, `/session start`, `/session continue`, `/session status`
+Use ONLY the exact command formats specified in this template.
+
+## Task: Continue Existing Session
+
+Parse the session name from the command arguments. The command format is: `/session:continue [name]`
+
+**OPTIMIZATION**: v3.7.0 uses parallel subagent delegation for 72% token reduction (77k → 22k tokens).
+
+### Step 1: Validate Session Exists (CLI)
+
+Extract the session name from arguments, then run:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get {session_name}
+```
+
+If this returns an error (exit code 2), the session doesn't exist. Show:
+```
+❌ Error: Session '{name}' not found
+💡 Use /session:list to see available sessions
+💡 Use /session:start {name} to create a new session
+```
+Then STOP.
+
+The JSON response contains metadata (status, started, snapshotCount, etc.).
+
+### Step 1.5: Close Previous Active Session (If Different)
+
+Before continuing the target session, close any currently active session if it's different:
+
+1. Check if `.claude/sessions/.active-session` exists
+2. If it exists:
+   - Read the current active session name
+   - If it's different from the target session {session_name}:
+     - Show: "📋 Closing previous session '{previous_session_name}'..."
+     - Close the previous session status:
+       ```bash
+       node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-status "{previous_session_name}" "closed"
+       ```
+     - This marks the previous session as closed in `.auto-capture-state`
+3. Continue to next step
+
+**Note**: This ensures clean session transitions with no abandoned active sessions.
+
+### Step 1.8: Calculate Absolute Paths for Subagents
+
+Before delegating to subagents, you need to determine absolute paths that subagents will use.
+
+**IMPORTANT**: Subagents don't inherit the working directory or environment variables from the main conversation. You MUST provide absolute paths explicitly.
+
+Calculate these values (conceptually - don't run bash commands, just determine the values):
+- **Working directory**: Current working directory (you already know this from your environment)
+- **Plugin root**: `/Users/prajyot/.claude/plugins/marketplaces/automatewithus-plugins/session`
+- **Session path**: `{working_directory}/.claude/sessions/{session_name}`
+
+You will substitute these absolute paths into the subagent prompts in the next step.
+
+### Step 2: Delegate Heavy Work to Subagents (Parallel Execution)
+
+**CRITICAL**: You MUST invoke ALL 3 Task tools in a SINGLE response message. This runs them in parallel and isolates heavy token usage from the main conversation.
+
+Use the Task tool to spawn 3 parallel subagents with these exact configurations:
+
+**Subagent 1 - Consolidate Conversation Log:**
+- subagent_type: "general-purpose"
+- description: "Consolidate conversation log"
+- model: "haiku"
+- prompt: |
+  You are working with session: {session_name}
+
+  **Absolute paths for this task:**
+  - Working directory: {working_directory}
+  - Plugin root: ${CLAUDE_PLUGIN_ROOT}
+  - Session path: {session_path}
+
+  Read the prompt file: ${CLAUDE_PLUGIN_ROOT}/prompts/consolidate-log.md
+  That file contains template placeholders like "{session_name}", "{session_path}", "${CLAUDE_PLUGIN_ROOT}".
+  Replace all such placeholders with the actual values provided above.
+  Then execute the resulting instructions.
+
+**Subagent 2 - Refresh Git History:**
+- subagent_type: "general-purpose"
+- description: "Refresh git history"
+- model: "haiku"
+- prompt: |
+  You are working with session: {session_name}
+
+  **Absolute paths for this task:**
+  - Working directory: {working_directory}
+  - Plugin root: ${CLAUDE_PLUGIN_ROOT}
+  - Session path: {session_path}
+
+  Read the prompt file: ${CLAUDE_PLUGIN_ROOT}/prompts/refresh-git.md
+  That file contains template placeholders like "{session_name}", "{session_path}", "${CLAUDE_PLUGIN_ROOT}".
+  Replace all such placeholders with the actual values provided above.
+  Then execute the resulting instructions.
+
+**Subagent 3 - Extract Session Goal:**
+- subagent_type: "general-purpose"
+- description: "Extract session goal"
+- model: "haiku"
+- prompt: |
+  You are working with session: {session_name}
+
+  **Absolute paths for this task:**
+  - Working directory: {working_directory}
+  - Plugin root: ${CLAUDE_PLUGIN_ROOT}
+  - Session path: {session_path}
+
+  Read the prompt file: ${CLAUDE_PLUGIN_ROOT}/prompts/extract-goal.md
+  That file contains template placeholders like "{session_name}", "{session_path}", "${CLAUDE_PLUGIN_ROOT}".
+  Replace all such placeholders with the actual values provided above.
+  Then execute the resulting instructions.
+
+**REMINDER**: All 3 Task invocations MUST be in the SAME response to execute in parallel!
+
+### Step 3: Process Subagent Results
+
+After all 3 subagents complete, you'll receive their results. Handle errors gracefully:
+
+**Consolidation Result**:
+- If `success: true` → Snapshot created successfully
+- If `skipped: true` → No conversation log found (OK, continue)
+- If `success: false` → Log error but continue
+
+**Git Refresh Result**:
+- If `success: true` → Git history updated
+- If `success: false` → No git repo or error (OK, continue)
+
+**Goal Extraction Result**:
+- If `success: true` → Use the extracted goal
+- If `success: false` → Use fallback goal from result
+
+### Step 3.5: Extract Full Snapshot Summary (Complete Context Approach)
+
+Provide Claude with complete snapshot summary including all topics, decisions, and tasks for full context visibility using Claude Code tools.
+
+**IMPORTANT**: This step uses Claude Code's Read and Glob tools instead of bash pipelines to avoid parse errors with command substitution.
+
+**Implementation Steps:**
+
+1. **Find the latest snapshot file using Glob tool:**
+   - Pattern: `.claude/sessions/{session_name}/auto_*.md`
+   - Glob returns files sorted by modification time (newest first)
+   - Take the first result as the latest snapshot
+
+2. **If snapshot exists, use Read tool to extract content:**
+   - Read the snapshot file (first 80 lines should contain all needed sections)
+   - Extract all snapshot items with titles only:
+     - **Topics Discussed**: Extract ALL topic titles from "## Topics Discussed" section
+       - Format: "1. **Category**: Description"
+       - Extract only the category/title (bold text between ** markers)
+       - Store as array of titles
+     - **Decisions Made**: Extract ALL decision titles from "## Decisions Made" section
+       - Format: "1. **Decision**: Rationale"
+       - Extract only the decision title (bold text between ** markers)
+       - Store as array of titles
+     - **Tasks Completed**: Extract ALL tasks from "## Tasks Completed" section
+       - Format: "1. Task description"
+       - Extract full task line (simple numbered list)
+       - Store as array of tasks
+     - **Current Status**: Extract status lines from "## Current Status" section
+       - Look for "- **Progress**:" line and extract text after it
+       - Look for "- **Next Steps**:" line and extract text after it
+       - Look for "- **Blockers**:" line and extract text after it
+       - Store as object with progress, nextSteps, blockers
+
+3. **Build full snapshot summary output:**
+
+   ```
+   📋 Latest: {snapshot_filename}
+
+   Topics Discussed ({count}):
+   - {topic_1}
+   - {topic_2}
+   ... (all topics)
+
+   Decisions Made ({count}):
+   - {decision_1}
+   - {decision_2}
+   ... (all decisions)
+
+   Tasks Completed ({count}):
+   - {task_1}
+   - {task_2}
+   ... (all tasks)
+
+   Current Status:
+   • Progress: {progress_text}
+   • Next Steps: {next_steps_text}
+   • Blockers: {blockers_text}
+
+   💡 Read {snapshot_path} for full details
+   ```
+
+4. **Display the summary in Step 6** (after showing the session goal)
+
+**Graceful Handling**:
+- If no snapshot exists → Skip summary display (OK, fresh session)
+- If extraction fails → Show generic message "See snapshot for details"
+- If Read fails → Silent failure, continue (don't break resume)
+
+**Why Full Snapshot Summary Works Better**:
+- Complete visibility: All topics, decisions, and tasks visible immediately (~300 tokens)
+- No information loss: User sees everything without needing to read snapshot file
+- Better context continuity: Claude knows full scope of work done
+- Improved decision-making: All context available for next steps
+- **No bash parse errors** - uses Claude Code tools (Glob, Read) natively
+
+### Step 4: Activate Session (CLI)
+
+Run the CLI command to activate the session:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js activate {session_name}
+```
+
+This updates both the .active-session file and the index.
+
+### Step 4.5: Update Session Status to Active
+
+Update the session status in `.auto-capture-state` and index:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-status "{session_name}" "active"
+```
+
+This ensures the session status matches its active state and prevents sync bugs.
+
+### Step 5: Update Last Updated Timestamp
+
+Update the "Last Updated" line in session.md to current time using the Edit tool:
+
+```
+**Last Updated**: {current ISO timestamp}
+```
+
+### Step 6: Display Summary with Full Snapshot Details
+
+Show session goal plus complete snapshot summary with all topics, decisions, and tasks for full context visibility.
+
+**Implementation**: Use Glob and Read tools (from Step 3.5) to extract and display the full summary.
+
+**Display Format**:
+```
+✓ Session ready: {goal}
+
+📋 Latest: {snapshot_filename}
+
+Topics Discussed ({count}):
+- {topic_1}
+- {topic_2}
+... (all topics)
+
+Decisions Made ({count}):
+- {decision_1}
+- {decision_2}
+... (all decisions)
+
+Tasks Completed ({count}):
+- {task_1}
+- {task_2}
+... (all tasks)
+
+Current Status:
+• Progress: {progress_text}
+• Next Steps: {next_steps_text}
+• Blockers: {blockers_text}
+
+💡 Read {snapshot_path} for full details
+
+What's next?
+```
+
+**Example Output**:
+```
+✓ Session ready: Implement product permission system
+
+📋 Latest: auto_2025-11-16_05-24.md
+
+Topics Discussed (8):
+- Database Schema Design
+- API Endpoint Implementation
+- Permission Middleware
+- Frontend Components
+- Testing Strategy
+- Deployment Planning
+- Documentation Updates
+- Performance Optimization
+
+Decisions Made (3):
+- Use RBAC Model for Permission System
+- Implement Middleware-based Authorization
+- Store Permissions in PostgreSQL
+
+Tasks Completed (12):
+- Created users, roles, permissions tables
+- Implemented role assignment API
+- Built permission checking middleware
+- Added frontend permission components
+- Wrote unit tests for permission logic
+- Created integration tests
+- Documented API endpoints
+- Updated deployment guide
+- Fixed TypeScript errors
+- Ran build successfully
+- Deployed to staging
+- Verified permission checks work
+
+Current Status:
+• Progress: 12 of 12 tasks completed (100%)
+• Next Steps: Deploy to production and monitor
+• Blockers: None
+
+💡 Read .claude/sessions/product-permission/auto_2025-11-16_05-24.md for full details
+
+What's next?
+```
+
+**Notes**:
+- If no snapshot exists, only show "✓ Session ready: {goal}" and "What's next?"
+- Use Read tool to extract all items (avoids bash parse errors)
+- Fallback gracefully if extraction fails (show generic pointer text)
+
+**IMPORTANT**:
+- DO show full snapshot summary (all topics/decisions/tasks) ← FULL VISIBILITY
+- This provides complete context in ~300 tokens ← WORTH THE COST
+- User doesn't need to read snapshot file separately ← CONVENIENCE
+- All relevant context immediately available for decision-making ← BETTER UX
+- User can still read snapshot file for additional details if needed
+- The heavy analysis already happened in the subagents
+- User can run `/session:status` for detailed view
+
+---
+
+**TOKEN OPTIMIZATION BENEFITS:**
+- Current: ~20.3k tokens with full snapshot summary (74% reduction from v3.6.4)
+  - Full snapshot summary with all topics/decisions/tasks (~300 tokens)
+  - Complete visibility: All snapshot items visible immediately
+  - Benefit: Complete context for better decision-making and continuity
+- Heavy work (consolidation, git analysis) happens in isolated subagent contexts
+- Parallel execution: 3 subagents run simultaneously (~2-4 seconds total)
+- Lazy-loaded prompts: Subagents read their own prompts (~1.7k token savings)
+- Result: Faster session resume, massive token savings, complete context restoration
+
+**ERROR HANDLING:**
+- If all subagents fail: Still activate session, show generic message "✓ Session ready. What's next?"
+- If session.md missing: Show corrupted session warning
+- If CLI fails: Suggest rebuilding index with `/session:rebuild-index`
+
+---
+
+ARGUMENTS: {name}
diff --git a/commands/continue.md.v3.6.4.backup b/commands/continue.md.v3.6.4.backup
new file mode 100644
index 0000000..58d286d
--- /dev/null
+++ b/commands/continue.md.v3.6.4.backup
@@ -0,0 +1,199 @@
+You are managing a session memory system. The user wants to resume an existing session.
+
+## Task: Continue Existing Session
+
+Parse the session name from the command arguments. The command format is: `/session continue [name]`
+
+**OPTIMIZATION:** This command uses CLI for validation and metadata (60-70% token reduction).
+
+### Step 1: Validate Session Exists (CLI)
+
+Extract the session name from arguments, then run:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get {session_name}
+```
+
+If this returns an error (exit code 2), the session doesn't exist. Show:
+```
+❌ Error: Session '{name}' not found
+💡 Use /session list to see available sessions
+💡 Use /session start {name} to create a new session
+```
+Then STOP.
+
+The JSON response contains metadata (status, started, goal, snapshotCount, etc.).
+
+### Step 2: Read Session Files
+
+Now read the actual content files (these need full content for context synthesis):
+
+1. Read `.claude/sessions/{name}/session.md`
+2. Read `.claude/sessions/{name}/context.md` (if exists)
+3. Get latest snapshot filename from the CLI JSON response (`latestSnapshot` field)
+4. If `latestSnapshot` exists, read `.claude/sessions/{name}/{latestSnapshot}`
+
+### Step 3: Check for Active Session and Transition (NEW - Session Cleanup)
+
+Before activating the new session, check if there's already a different active session:
+
+1. Check if `.claude/sessions/.active-session` exists
+2. If it exists, read the current active session name
+3. If the current active session is **different** from the session being continued:
+   - Show: "📋 Closing previous session '{previous_session_name}'..."
+   - Update the previous session's `session.md` "Last Updated" timestamp to current time
+   - This provides clean transition tracking
+4. If it's the **same** session, skip this step (just updating timestamp)
+
+**Note:** The SessionEnd hook will handle final cleanup on Claude Code termination.
+
+### Step 4: Activate Session (CLI)
+
+Run the CLI command to activate the session:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js activate {session_name}
+```
+
+This updates both the .active-session file and the index.
+
+### Step 5: Update Last Updated Timestamp
+
+Update the "Last Updated" line in session.md to current time using the Edit tool.
+
+### Step 6: Synthesize and Display Context Summary
+
+Using the data from CLI JSON + file contents, show a comprehensive summary:
+
+```
+✓ Loaded session: '{name}'
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📅 Started: {started from CLI JSON}
+🎯 Goal: {goal from CLI JSON or session.md}
+📝 Last update: {lastUpdated from CLI JSON}
+⏰ Status: {status from CLI JSON}
+📸 Snapshots: {snapshotCount from CLI JSON}
+
+## Recent Context
+
+{latest_snapshot_conversation_summary if available}
+
+## Key Files
+{list of files from filesInvolved array in CLI JSON}
+
+## Milestones
+{key milestones from session.md with checkboxes}
+
+## Recent Decisions
+{recent decisions from context.md, max 3}
+
+## Current State
+{current state from latest snapshot if available}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+Ready to continue! How would you like to proceed?
+```
+
+### Step 7: Prepare for Work
+
+Tell the user:
+"I've loaded the full context for session '{name}'. All previous work, decisions, and progress have been restored. What would you like to work on next?"
+
+---
+
+**PERFORMANCE BENEFITS:**
+- **Before:** 10-20K tokens, reads session.md + context.md + snapshot + metadata parsing, 3-5s
+- **After:** 3-8K tokens, CLI provides metadata instantly, only reads content files, 1-2s
+- **Improvement:** ~60-70% token reduction, ~2x faster
+
+**WHY STILL READ CONTENT FILES:**
+- context.md and snapshots contain narrative context needed for synthesis
+- Claude needs full context to provide meaningful summary
+- CLI provides structure/metadata, Claude provides intelligence/understanding
+
+**ERROR HANDLING:**
+- If session.md missing, show corrupted session warning
+- If CLI fails, suggest rebuilding index
+- Handle missing context.md or snapshots gracefully (show what's available)
+
+---
+
+## CRITICAL: Check for Unconsolidated Logs (Inline Analysis)
+
+**YOU MUST CHECK THIS BEFORE DISPLAYING THE SESSION SUMMARY:**
+
+When resuming a session, check if the previous session left unconsolidated logs:
+
+1. Check if `.claude/sessions/{name}/conversation-log.jsonl` exists
+2. If the file exists:
+   - Show brief message: "📊 Analyzing previous session... (this may take 1-3 seconds)"
+   - Read the conversation log file
+   - Parse interactions from JSONL format
+   - **Parse self-contained conversation log (v3.6.2+):**
+     - Each entry has type: 'interaction' (user prompt) or 'assistant_response' (Claude's response)
+     - Extract full conversation from log entries:
+       - User prompts: `user_prompt` field in 'interaction' entries
+       - Claude responses: `response_text` field in 'assistant_response' entries
+       - Tools used: `tools_used` field in 'assistant_response' entries
+       - File modifications: `modified_files` field in 'interaction' entries
+     - Log is self-contained - NO need to read transcript file!
+     - **Backward compatibility**: If `transcript_path` exists but no `response_text`, fall back to reading transcript file (for v3.6.1 logs)
+   - **Capture git history (if available):**
+     - Run: `node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js capture-git "{name}"`
+     - This creates `.claude/sessions/{name}/git-history.json` (~2-3KB compressed)
+     - Contains: last 50 commits, uncommitted changes, branch status, hotspots
+     - Performance: ~60-90ms (acceptable at session boundary)
+     - If no git repo, command returns success: false (silent skip, no error)
+   - **Analyze the conversation with Claude inline:**
+     - Use the full conversation from log entries (user prompts + Claude responses)
+     - Extract conversation summary (2-3 paragraphs covering what happened)
+     - Identify key decisions made with rationale
+     - List completed todos/tasks
+     - Document files modified with context about what changed and why (from conversation, not just file paths)
+     - Assess current state, what's next, and any blockers
+   - Create consolidated snapshot via CLI:
+     ```bash
+     echo "# Consolidated Snapshot: {session_name}
+**Timestamp**: {ISO timestamp}
+**Method**: Claude Inline Analysis (Free)
+**Status**: Consolidated from conversation log
+
+## Conversation Summary
+{2-3 paragraph summary of what happened in session}
+
+## Key Decisions
+- {Decision 1 with rationale}
+- {Decision 2 with rationale}
+
+## Completed Tasks
+- {Task 1}
+- {Task 2}
+
+## Files Modified
+- {file_path}: {what changed and why}
+
+## Current State
+{Where things stand, what's next, blockers}
+
+## Notes
+Consolidated via Claude inline analysis at session boundary. Zero cost, highest quality." | node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js write-snapshot "{name}" --stdin --type auto
+     ```
+   - Delete conversation-log.jsonl after successful snapshot creation
+   - Update `.auto-capture-state` to reset counters and set last_snapshot_timestamp
+3. If no log exists:
+   - **Still capture git history** for updated repository context:
+     - Run: `node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js capture-git "{name}"`
+     - This refreshes git context since last session
+     - Silent skip if no git repo (no error)
+
+**PERFORMANCE:**
+- Log check: <5ms
+- Claude analysis: 1-3s (acceptable at session boundaries - users expect loading)
+- Snapshot write: <50ms
+- Log deletion: <5ms
+- **Total: ~1-3 seconds** (users expect loading at session resume)
+
+**NOTE:** This is the v3.5.1 architecture where:
+- During session: Conversation logged incrementally (<2ms per interaction, zero blocking)
+- At session boundaries: Claude inline analysis creates intelligent snapshots (FREE, highest quality)
+- Result: User NEVER experiences blocking during work, only brief wait at session resume where loading is expected
diff --git a/commands/delete.md b/commands/delete.md
new file mode 100644
index 0000000..ebcd188
--- /dev/null
+++ b/commands/delete.md
@@ -0,0 +1,207 @@
+You are managing a session memory system. The user wants to delete a session permanently.
+
+## Task: Delete Session
+
+Delete a session and all its data. This action cannot be undone.
+
+**IMPORTANT:** This is a destructive operation. Always confirm with the user before deleting.
+
+---
+
+## Step 1: Parse Arguments
+
+Check if user provided a session name:
+- If user provided a session name (e.g., `/session:delete old-feature`) → Use that name
+- If no name provided → Show error and list available sessions for reference
+
+If no session name provided:
+```
+❌ Session name required.
+
+Usage: /session:delete [session-name]
+
+Available sessions: /session:list
+```
+Then STOP.
+
+---
+
+## Step 2: Get Session Details
+
+Run the CLI command to get session information:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get [session-name]
+```
+
+If the command fails (session not found), display:
+```
+❌ Session '[session-name]' not found.
+
+Available sessions: /session:list
+```
+Then STOP.
+
+---
+
+## Step 3: Check If Active Session
+
+Check if this is the currently active session by running:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js list --json
+```
+
+Parse the JSON and check if `activeSession === session-name`.
+
+---
+
+## Step 4: Show Confirmation Prompt
+
+Use the AskUserQuestion tool to confirm deletion.
+
+### If NOT the active session:
+
+```json
+{
+  "questions": [
+    {
+      "question": "⚠️ Delete Session: [session-name]\n\n📅 Started: [started]\n📸 Snapshots: [count]\n📁 Files Involved: [count]\n\nThis will permanently delete all session data.\n\nAre you sure you want to delete this session?",
+      "header": "Confirm Delete",
+      "multiSelect": false,
+      "options": [
+        {
+          "label": "Yes, Delete",
+          "description": "Permanently delete this session and all its data"
+        },
+        {
+          "label": "No, Cancel",
+          "description": "Keep the session and go back"
+        }
+      ]
+    }
+  ]
+}
+```
+
+### If IS the active session:
+
+```json
+{
+  "questions": [
+    {
+      "question": "⚠️ WARNING: Deleting ACTIVE Session!\n\n'[session-name]' is your currently active session.\n\n📅 Started: [started]\n📸 Snapshots: [count]\n📁 Files Involved: [count]\n\nDeleting this will:\n- End your current working session\n- Delete all snapshots and context\n- Clear the active session\n\nThis cannot be undone.\n\nAre you ABSOLUTELY SURE?",
+      "header": "Delete Active",
+      "multiSelect": false,
+      "options": [
+        {
+          "label": "Yes, Delete Active Session",
+          "description": "I understand this will end my active session"
+        },
+        {
+          "label": "No, Cancel",
+          "description": "Keep the session and go back"
+        }
+      ]
+    }
+  ]
+}
+```
+
+---
+
+## Step 5: Handle User Response
+
+### If user selected "No, Cancel":
+```
+Session deletion cancelled.
+```
+Then STOP.
+
+### If user selected "Yes" (or "Yes, Delete Active Session"):
+
+Proceed to Step 6.
+
+---
+
+## Step 6: Execute Deletion
+
+Run the CLI delete command:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js delete [session-name]
+```
+
+The CLI will:
+- Delete the session directory and all files
+- Remove from index
+- Clear active session if needed
+
+---
+
+## Step 7: Display Result
+
+Parse the JSON response from the delete command.
+
+### If successful:
+
+**For regular session:**
+```
+✅ Session '[session-name]' has been deleted.
+
+📊 Deleted:
+- [count] snapshots
+- [count] files tracked
+
+💡 View remaining sessions: /session:list
+```
+
+**For active session:**
+```
+✅ Session '[session-name]' has been deleted.
+
+📊 Deleted:
+- [count] snapshots
+- [count] files tracked
+
+⚠️ No active session. Start a new one:
+   /session:start [name]
+
+💡 View remaining sessions: /session:list
+```
+
+### If error occurred:
+
+```
+❌ Error deleting session: [error message]
+
+Try rebuilding the index:
+   node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-index --full-rebuild
+```
+
+---
+
+## Error Handling
+
+- **Session not found:** Show available sessions list
+- **Permission error:** Show clear error message about file permissions
+- **Active session:** Show extra warning and confirmation
+- **CLI error:** Show error and suggest index rebuild
+
+---
+
+## Performance
+
+- **Single confirmation:** < 300 tokens
+- **CLI deletion:** < 50ms for small sessions, < 500ms for large sessions
+- **Total time:** 1-2 seconds including user confirmation
+
+---
+
+## Safety Features
+
+1. ✅ **Always requires confirmation** - No accidental deletions
+2. ✅ **Shows what will be deleted** - User knows the impact
+3. ✅ **Extra warning for active session** - Prevents losing current work
+4. ✅ **Clear feedback** - User knows what happened
+5. ✅ **Graceful error handling** - Helpful messages on failure
diff --git a/commands/git-decompress.md b/commands/git-decompress.md
new file mode 100644
index 0000000..f5e5ab9
--- /dev/null
+++ b/commands/git-decompress.md
@@ -0,0 +1,124 @@
+You are managing a session memory system. The user wants to decompress git history for human inspection.
+
+## Task: Decompress Git History
+
+Parse the session name from the command arguments. The command format is: `/session:git-decompress [name]`
+
+### Step 1: Validate and Read Compressed File
+
+1. Extract session name from arguments
+2. Check if `.claude/sessions/{name}/git-history.json` exists
+3. If not exists, show:
+   ```
+   ❌ Error: No git history found for session '{name}'
+   💡 Git history is captured automatically at session start/continue
+   💡 Or run: /session:capture-git {name}
+   ```
+   Then STOP.
+4. Read `.claude/sessions/{name}/git-history.json`
+5. Parse JSON into object
+
+### Step 2: Expand into Human-Readable Format
+
+Display the expanded markdown format:
+
+```markdown
+# Git History: {s}
+**Captured**: {t}
+**Branch**: {b}
+**HEAD**: {h}
+
+## Summary
+- **Commits analyzed**: {sm.n}
+- **Date range**: {sm.r}
+- **Days span**: {sm.d}
+- **Total files modified**: {sm.f}
+- **Total changes**: {sm.ch}
+
+## Uncommitted Changes
+
+### Tracking Status
+- **Ahead of upstream**: {uc.ah} commits
+- **Behind upstream**: {uc.bh} commits
+
+### Staged for Commit ({uc.stg.length} files)
+{for each file in uc.stg:}
+- `{file[0]}`: {file[1]}
+
+### Modified (Unstaged) ({uc.mod.length} files)
+{for each file in uc.mod:}
+- `{file[0]}`: {file[1]}
+
+### New Files (Untracked) ({uc.new.length} files)
+{for each file in uc.new:}
+- `{file}`
+
+### Deleted Files ({uc.del.length} files)
+{for each file in uc.del:}
+- `{file}`
+
+### Conflicted Files ({uc.con.length} files)
+{for each file in uc.con:}
+- `{file}`
+
+**Total uncommitted changes**: {uc.tot}
+
+## Recent Commits
+
+{for each commit in c:}
+### {commit[0]} - {commit[1]}
+**Message**: {commit[2]}
+**Changes**: {commit[3]}
+**Files modified** ({commit[4]}):
+{for each file in commit[5]:}
+- `{file}`
+
+---
+
+## Development Hotspots
+
+Shows directories with most commit activity:
+
+{for each hotspot in hot:}
+- **{hotspot[0]}**: {hotspot[1]} commits
+
+---
+
+**Note**: This is a decompressed view of the compressed JSON format used for token efficiency. The compressed format saves ~70-75% tokens compared to markdown while preserving all information.
+```
+
+### Step 3: Display Statistics
+
+After showing the decompressed data, display:
+
+```
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📊 Decompression Statistics
+
+Compressed size: {fileSize} bytes ({fileSizeKB} KB)
+Estimated markdown size: ~{estimatedMarkdownSize} bytes (~{estimatedMarkdownKB} KB)
+Compression efficiency: ~{compressionPercent}% smaller
+
+Token efficiency: Compressed JSON uses ~70-75% fewer tokens than markdown
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+---
+
+## Usage Notes
+
+- This command is primarily for debugging and human inspection
+- Claude can read the compressed format directly (it's more efficient)
+- Use this when you need to:
+  - Verify git history capture is working correctly
+  - Inspect what commits are included
+  - Debug git context issues
+  - Share git history in readable format
+
+---
+
+## Error Handling
+
+- If git-history.json doesn't exist: Show helpful error with instructions
+- If JSON is malformed: Show parsing error and suggest recapture
+- If file is empty: Suggest running capture-git again
diff --git a/commands/list.md b/commands/list.md
new file mode 100644
index 0000000..cae9052
--- /dev/null
+++ b/commands/list.md
@@ -0,0 +1,231 @@
+You are managing a session memory system. The user wants to see all available sessions.
+
+## Task: List All Sessions (Enhanced with Interactive Selection)
+
+Display all sessions with enhanced metadata in a clean, organized format. Optionally enable interactive selection mode.
+
+**OPTIMIZATION:** This command uses the lightweight CLI tool for instant metadata retrieval (< 10ms, < 200 tokens).
+
+---
+
+## Step 1: Detect Mode
+
+Check if user provided arguments:
+- If user provided a **number** (e.g., `/session:list 2`) → Jump to **Step 5: Handle Selection**
+- If user typed `/session:list` with no args → Continue to Step 2 (display mode)
+
+---
+
+## Step 2: Get Session List from CLI
+
+Run the CLI command to get all session metadata:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js list
+```
+
+This returns JSON with all session data from the index (no file reading needed).
+
+---
+
+## Step 3: Handle Empty Sessions
+
+If the JSON shows `totalSessions: 0`, display:
+```
+No sessions found.
+
+💡 Create your first session with:
+   /session:list [name]
+
+Example:
+   /session:start my-feature
+```
+Then STOP.
+
+---
+
+## Step 4: Format and Display Sessions (Enhanced)
+
+Parse the JSON response and format with **enhanced visual design**:
+
+### Calculate Stats:
+- Count active vs closed sessions
+- Calculate relative times (use helper function below)
+- Determine activity status for badges
+
+### Display Format:
+
+```
+Available Sessions ({activeCount} active, {closedCount} closed):
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+{for each session in sessions array:}
+
+{number}. {name} {badges}
+   📅 {relativeTime} (started {relativeStartTime})  📸 {snapshotCount} snapshots  📁 {filesInvolvedCount} files
+   🎯 {goal}
+   {if latestSnapshotSummary exists: show "💬 Last: \"{latestSnapshotSummary}\""}
+
+{Add TWO blank lines between sessions for clear visual separation}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+💡 Select a session: /session:list [number]
+💡 Create new: /session:start [name]
+```
+
+### Badges Logic:
+- If `session.name === activeSession`: Show `[ACTIVE] 🔥`
+- If `status === "closed"`: Show `✅ CLOSED`
+- If last update > 7 days and not closed: Show `🧊 COLD`
+- If last update < 1 hour: Show `🔥 HOT`
+
+### Relative Time Helper:
+
+Calculate relative time for display:
+- < 1 minute: "just now"
+- < 60 minutes: "{n}m ago"
+- < 24 hours: "{n}h ago"
+- < 7 days: "{n}d ago"
+- < 30 days: "{n}w ago"
+- >= 30 days: "{n} months ago"
+
+### Example Output:
+
+```
+Available Sessions (2 active, 1 closed):
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+1. feature-auth-system [ACTIVE] 🔥 HOT
+   📅 2h ago (started 3d ago)  📸 5 snapshots  📁 12 files
+   🎯 Implement OAuth2 authentication system
+   💬 Last: "Completed login flow, testing redirect logic"
+
+
+2. bugfix-login-issue ✅ CLOSED
+   📅 1d ago (started 2d ago)  📸 3 snapshots  📁 4 files
+   🎯 Fix session timeout bug in login flow
+
+
+3. refactor-api-layer 🧊 COLD
+   📅 14d ago (started 15d ago)  📸 8 snapshots  📁 23 files
+   🎯 Refactor REST API to GraphQL architecture
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+💡 Select a session: /session:list [number]
+💡 Create new: /session:start [name]
+```
+
+After displaying, **STOP and wait for user input**. Do not prompt further.
+
+---
+
+## Step 5: Handle Selection (Interactive Mode)
+
+If user provided a number (e.g., `/session:list 2`), this is interactive selection mode.
+
+### 5.1: Validate Selection
+
+1. Get session list again (same CLI command)
+2. Check if number is valid (1 to totalSessions)
+3. If invalid, show error and STOP:
+   ```
+   ❌ Invalid selection. Please choose 1-{totalSessions}.
+   ```
+
+### 5.2: Show Session Details
+
+Display the selected session with full details:
+
+```
+Selected Session: {name}
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+📋 Status: {status} {badges}
+📅 Started: {started}
+📅 Last Update: {lastUpdated}
+📸 Snapshots: {snapshotCount}
+📁 Files Involved: {filesInvolvedCount}
+
+🎯 Goal:
+{goal}
+
+{if latestSnapshotSummary:}
+💬 Latest Snapshot:
+{latestSnapshotSummary}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+```
+
+### 5.3: Use AskUserQuestion Tool
+
+Present action options using the AskUserQuestion tool:
+
+```json
+{
+  "questions": [
+    {
+      "question": "What would you like to do with this session?",
+      "header": "Action",
+      "multiSelect": false,
+      "options": [
+        {
+          "label": "Continue/Resume",
+          "description": "Load session context and continue working"
+        },
+        {
+          "label": "View Status",
+          "description": "Show detailed session statistics and state"
+        },
+        {
+          "label": "Save Snapshot",
+          "description": "Capture current conversation as a snapshot"
+        },
+        {
+          "label": "Close Session",
+          "description": "Finalize and close this session"
+        },
+        {
+          "label": "Delete Session",
+          "description": "Permanently delete this session and all its data"
+        },
+        {
+          "label": "Back to List",
+          "description": "Return to session list"
+        }
+      ]
+    }
+  ]
+}
+```
+
+### 5.4: Execute Selected Action
+
+Based on user's choice, execute the appropriate command:
+
+- **Continue/Resume** → Run: `/session:continue {sessionName}`
+- **View Status** → Run: `/session:status` (after activating if needed)
+- **Save Snapshot** → Run: `/session:save`
+- **Close Session** → Run: `/session:close`
+- **Delete Session** → Run: `/session:delete {sessionName}`
+- **Back to List** → Run: `/session:list` (restart from Step 2)
+
+**Important:** Actually execute the command, don't just tell the user to run it.
+
+---
+
+## Error Handling
+
+If CLI command fails, show:
+```
+❌ Error reading sessions. Try rebuilding the index:
+   node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-index --full-rebuild
+```
+
+---
+
+## Performance Benefits
+
+- **Display Mode:** < 200 tokens, < 50ms
+- **Selection Mode:** ~300-500 tokens total
+- **No file reads** for list display (index only)
+- **95-98% token reduction** vs old approach
diff --git a/commands/plan-execute.md b/commands/plan-execute.md
new file mode 100644
index 0000000..ce4f073
--- /dev/null
+++ b/commands/plan-execute.md
@@ -0,0 +1,195 @@
+You are executing the /session:plan-execute command to start executing a plan.
+
+**NOTE:** Plans are now global and independent of sessions.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`, `/session:plan-list`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`, `/plan-list`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+Parsed from user input:
+- `plan_name`: {name} (required)
+
+ARGUMENTS: {name}
+
+## Workflow
+
+### Step 1: Validate Plan Exists
+
+Plans are stored globally in `.claude/plans/`. Check if the plan exists:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-exists {plan_name}
+```
+
+If the plan doesn't exist, show error and STOP:
+```
+❌ Error: Plan '{plan_name}' not found
+
+Use /session:save-plan {name} to create a plan first.
+Use /session:plan-list to see available plans.
+```
+
+### Step 2: Validate Plan Format
+
+Before executing, verify the plan is in implementation format (not conceptual):
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get-plan-format {plan_name}
+```
+
+This returns JSON with `format: "conceptual"` or `format: "implementation"`.
+
+If the format is "conceptual", show this error and STOP:
+```
+❌ Error: Cannot execute conceptual plan
+
+Plan '{plan_name}' is still in conceptual format (requirements only).
+
+You must finalize the plan first to transform requirements into executable tasks:
+
+  /session:plan-finalize {plan_name}
+
+This will use AI to break down requirements into concrete implementation tasks
+organized by phases (Database, API, UI, Testing, etc.)
+
+After finalization, you can execute the plan.
+```
+
+If the format is "implementation", continue to next step.
+
+### Step 4: Load Plan Status
+
+Get the current plan status to show what will be executed:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-status {session_name} {plan_name}
+```
+
+This returns JSON with plan metadata, progress, and current phase.
+
+### Step 5: Show Execution Overview
+
+Display a summary of what will be executed:
+
+```
+📋 Plan: {plan_name}
+Goal: {goal}
+
+Progress: {completed_tasks}/{total_tasks} tasks ({percentage}%)
+Status: {status}
+
+Phases:
+  1. [✓] Phase 1: {name} (completed)
+  2. [→] Phase 2: {name} (in progress - will continue here)
+  3. [ ] Phase 3: {name} (pending)
+  4. [ ] Phase 4: {name} (pending)
+
+Current Phase: {current_phase_name}
+Next Task: {next_task_description}
+
+Ready to execute this plan?
+```
+
+### Step 6: Execution Strategy
+
+**IMPORTANT**: Plan execution is currently **manual** with task tracking:
+
+1. Show the user the next task to work on
+2. The user implements the task manually
+3. After completing a task, update its status:
+   ```bash
+   node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-task-status {session_name} {plan_name} {task_id} completed
+   ```
+4. Show progress after each task completion
+5. Repeat until all tasks are completed
+
+### Step 7: Display Current Task
+
+Show the next task that needs to be completed:
+
+```
+🎯 Next Task: {task_id}
+
+Phase: {phase_name}
+Task: {task_description}
+
+Details:
+{task_details}
+
+Technical Notes:
+- {note_1}
+- {note_2}
+
+Dependencies: {dependencies} (all completed ✓)
+
+---
+
+To mark this task as complete when done:
+/update-task-status {task_id} completed
+
+To see overall progress:
+/session:plan-status {plan_name}
+
+To continue to next task:
+Just complete the current task and I'll show you the next one.
+```
+
+### Step 8: Task Completion Loop
+
+After the user completes work:
+
+1. Ask: "Have you completed task {task_id}?"
+2. If yes, update task status:
+   ```bash
+   node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-task-status {session_name} {plan_name} {task_id} completed
+   ```
+3. Show updated progress
+4. Load next task
+5. If more tasks exist, show next task (Step 6)
+6. If all tasks complete, show completion message (Step 8)
+
+### Step 9: Completion Message
+
+When all tasks are completed:
+
+```
+🎉 Plan Complete: {plan_name}
+
+All tasks completed successfully!
+
+Summary:
+- Total Tasks: {total_tasks}
+- Completed: {total_tasks} (100%)
+- Phases: {total_phases}
+- Time Tracked: {duration}
+
+Next Steps:
+1. Review all changes
+2. Run tests if applicable
+3. Create a pull request
+4. Deploy to production
+
+Use /session:save to capture this milestone!
+```
+
+---
+
+## Notes
+
+- The execution is **manual** - Claude doesn't automatically execute code
+- Users implement tasks themselves, Claude tracks progress
+- This provides visibility and structure without automation
+- Future: Automated task execution with agent orchestration
+
+---
+
+## Error Handling
+
+At each step, handle errors gracefully:
+- Plan not found: Show clear message with available plans
+- Invalid task ID: Show valid task IDs for current phase
+- Session errors: Guide user to start/continue session
diff --git a/commands/plan-finalize.md b/commands/plan-finalize.md
new file mode 100644
index 0000000..66202d1
--- /dev/null
+++ b/commands/plan-finalize.md
@@ -0,0 +1,403 @@
+You are executing the /session:plan-finalize command to transform requirements into executable tasks.
+
+**NOTE:** Plans are now global and independent of sessions.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+Parsed from user input:
+- `plan_name`: {name} (required)
+
+ARGUMENTS: {name}
+
+## Purpose
+
+This is the TRANSFORMATION step that bridges conceptual planning and execution:
+
+**Before (Conceptual - requirements.json):**
+```
+Requirements (WHAT):
+- "Restrict products based on user permissions"
+- "Track who created each product"
+```
+
+**After (Implementation - orchestration.json + phases/):**
+```
+Phase 1: Database Layer
+  ✓ task-1-1: Add restriction_level column to public.new_product table
+  ✓ task-1-2: Create migration script for ALTER TABLE
+  ✓ task-1-3: Add created_by column with foreign key to users
+
+Phase 2: API Layer
+  ✓ task-2-1: Add validation in POST /api/products endpoint
+  ✓ task-2-2: Add validation in PUT /api/products/:id endpoint
+  ✓ task-2-3: Return 403 error for restricted products
+
+Phase 3: UI Layer
+  ✓ task-3-1: Add restriction checkbox in product form
+  ✓ task-3-2: Show error message when restriction fails
+```
+
+**This step uses AI to break down requirements into concrete, actionable tasks.**
+
+## Workflow
+
+### Step 1: Load Plan and Validate Format
+
+Plans are global and stored in `.claude/plans/`. Check what format the plan is in:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get-plan-format {plan_name}
+```
+
+This returns `format: "conceptual"` or `format: "implementation"`.
+
+**If format is "implementation":**
+```
+✓ Plan '{plan_name}' is already finalized
+
+This plan has already been transformed into executable tasks.
+
+Use /session:plan-execute {plan_name} to start implementation.
+```
+Then STOP.
+
+**If format is "conceptual":** Continue to next step.
+
+### Step 2: Load Requirements
+
+Load the requirements.json file from the global plans directory:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js load-requirements {plan_name}
+```
+
+This returns the requirements data:
+```json
+{
+  "goal": "...",
+  "requirements": [
+    { "id": "req-1", "description": "...", "priority": "high" },
+    { "id": "req-2", "description": "...", "priority": "medium" }
+  ],
+  "metadata": { "work_type": "feature" }
+}
+```
+
+### Step 4: Show Requirements Summary
+
+Display what will be transformed:
+
+```
+📋 Plan: {plan_name}
+
+Goal: {goal}
+Format: Conceptual (requirements)
+
+Requirements to Transform:
+  1. req-1: {description} [{priority}]
+  2. req-2: {description} [{priority}]
+  ... ({total} requirements)
+
+Work Type: {work_type}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+💡 Finalization Process:
+
+I will use AI to break down each requirement into concrete implementation tasks,
+organized by phases (Database, API, UI, Testing, etc.)
+
+This transformation will:
+  • Convert high-level requirements into actionable tasks
+  • Organize tasks by implementation layer
+  • Add implementation details (SQL, API endpoints, UI components)
+  • Track which requirement led to which tasks
+  • Create execution-ready plan structure
+
+Ready to proceed? (yes/no)
+```
+
+### Step 5: Get User Confirmation
+
+Use the AskUserQuestion tool to confirm.
+
+If user says no/cancel, show:
+```
+Plan finalization cancelled. The plan remains in conceptual format.
+```
+Then STOP.
+
+If yes, continue.
+
+### Step 6: AI-Powered Task Breakdown
+
+Show progress indicator:
+```
+🤖 AI Analysis in Progress...
+
+This will take 30-60 seconds.
+```
+
+Invoke the Task tool with:
+- subagent_type: "general-purpose"
+- model: "sonnet"  (use sonnet for better quality breakdown)
+- prompt: Read the file at `${CLAUDE_PLUGIN_ROOT}/prompts/breakdown-requirement.md`, replace `[REQUIREMENTS JSON INSERTED HERE]` with the actual requirements JSON, then execute those instructions
+
+The subagent will analyze requirements and return structured breakdown:
+```json
+{
+  "implementation_goal": "...",
+  "phases": [
+    {
+      "name": "Database Layer",
+      "description": "...",
+      "tasks": [
+        {
+          "description": "Add restriction_level column to products",
+          "details": "ALTER TABLE products ADD COLUMN...",
+          "from_requirement": "req-1"
+        }
+      ]
+    }
+  ],
+  "traceability": {
+    "req-1": ["task-1-1", "task-2-1", "task-3-1"],
+    "req-2": ["task-1-2", "task-2-2"]
+  },
+  "assumptions": [...],
+  "risks": [...]
+}
+```
+
+### Step 7: Show Transformation Preview
+
+Display the AI-generated breakdown for user review:
+
+```
+✓ AI Analysis Complete
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📋 Transformation Preview
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Implementation Goal:
+{implementation_goal}
+
+Phase 1: {phase_1_name}
+  • task-1-1: {description}
+  • task-1-2: {description}
+  • task-1-3: {description}
+  [{task_count} tasks]
+
+Phase 2: {phase_2_name}
+  • task-2-1: {description}
+  • task-2-2: {description}
+  [{task_count} tasks]
+
+Phase 3: {phase_3_name}
+  • task-3-1: {description}
+  [{task_count} tasks]
+
+[Show all phases]
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+Summary:
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+• Requirements: {requirement_count}
+• Phases: {phase_count}
+• Tasks: {task_count}
+• Traceability: Complete (all tasks mapped to requirements)
+
+Assumptions:
+  - {assumption_1}
+  - {assumption_2}
+
+Risks:
+  - {risk_1}
+  - {risk_2}
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Does this breakdown look good? (yes/edit/cancel)
+```
+
+### Step 8: Handle User Response
+
+**If "yes":** Continue to Step 9
+
+**If "edit":**
+```
+What would you like to modify?
+
+You can:
+  a) Adjust task descriptions
+  b) Add/remove tasks
+  c) Change phase organization
+  d) Add more details to specific tasks
+  e) Re-run AI analysis with different focus
+
+Your choice:
+```
+
+Based on user input, make modifications and show preview again (repeat Step 7).
+
+**If "cancel":**
+```
+Finalization cancelled. Plan remains in conceptual format.
+```
+STOP.
+
+### Step 9: Transform and Save
+
+Show progress:
+```
+💾 Transforming plan...
+
+Creating implementation structure...
+```
+
+Call the transformation function:
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js transform-plan {plan_name} '{breakdown_json}'
+```
+
+This will:
+1. Create orchestration.json with metadata and phase registry
+2. Create phases/*.json files with tasks
+3. Create execution-state.json for progress tracking
+4. Keep requirements.json for traceability
+5. Mark plan as "implementation" type
+
+### Step 10: Success Message
+
+```
+✓ Plan Finalized: {plan_name}
+
+Requirements successfully transformed into executable implementation plan!
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+Transformation Summary:
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+Input:
+  • Format: Conceptual (requirements)
+  • Requirements: {requirement_count}
+
+Output:
+  • Format: Implementation (orchestration + phases)
+  • Phases: {phase_count}
+  • Tasks: {task_count}
+  • Structure: Optimized for parallel execution
+
+Files Created:
+  • orchestration.json (coordinator)
+  • phases/phase-1-{name}.json
+  • phases/phase-2-{name}.json
+  • phases/phase-3-{name}.json
+  • execution-state.json (progress tracking)
+
+Preserved:
+  • requirements.json (for traceability)
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+Next Steps:
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+1. Review the finalized plan:
+   /session:plan-status {plan_name}
+
+2. Start execution:
+   /session:plan-execute {plan_name}
+
+3. Track progress as you implement each task
+
+🎯 Ready for implementation!
+```
+
+---
+
+## Transformation Details
+
+### File Structure
+
+**Before (Conceptual):**
+```
+plans/{plan_name}/
+├── requirements.json          (only file)
+└── conversation-context.md    (reference)
+```
+
+**After (Implementation):**
+```
+plans/{plan_name}/
+├── requirements.json          (preserved for traceability)
+├── orchestration.json         (coordinator + metadata)
+├── phases/
+│   ├── phase-1-database.json
+│   ├── phase-2-api.json
+│   └── phase-3-ui.json
+├── execution-state.json       (runtime progress)
+└── conversation-context.md    (reference)
+```
+
+### Traceability
+
+Every task in the implementation plan has a `from_requirement` field:
+```json
+{
+  "task_id": "task-1-1",
+  "description": "Add restriction_level column",
+  "from_requirement": "req-1"
+}
+```
+
+This allows you to trace back from implementation tasks to original requirements.
+
+### Why AI-Powered?
+
+Breaking down requirements into tasks requires:
+- Understanding of technical architecture (DB → API → UI)
+- Knowledge of implementation patterns
+- Ability to infer specific tasks from general requirements
+- Context about dependencies and ordering
+
+AI excels at this transformation, turning:
+- "Restrict products" → 10+ specific tasks across layers
+- "Track creators" → Database schema, API logic, UI display
+
+### Quality Assurance
+
+The AI breakdown is trained to:
+- Create concrete, actionable tasks (not vague)
+- Provide implementation details (SQL, file paths, logic)
+- Organize by architectural layers
+- Specify dependencies correctly
+- Estimate token usage realistically
+
+---
+
+## Error Handling
+
+- **Plan not found**: Show available plans
+- **Already finalized**: Show success, suggest execution
+- **Invalid session**: Guide to start/continue
+- **AI breakdown fails**: Offer retry or manual task creation
+- **User cancels**: Keep conceptual format, can retry later
+
+---
+
+## Notes
+
+- Finalization is REQUIRED to execute a plan
+- AI breakdown takes 30-60 seconds (uses Sonnet for quality)
+- User can review and modify before saving
+- Original requirements.json is preserved for traceability
+- Transformation is reversible (can re-finalize if needed)
+- This is where requirements (WHAT) become tasks (HOW)
diff --git a/commands/plan-list.md b/commands/plan-list.md
new file mode 100644
index 0000000..0f844b4
--- /dev/null
+++ b/commands/plan-list.md
@@ -0,0 +1,148 @@
+You are executing the /session:plan-list command to list all global plans.
+
+**NOTE:** Plans are now global and stored in `.claude/plans/`. This command works without requiring an active session.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`, `/session:plan-list`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`, `/plan-list`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+No arguments required.
+
+ARGUMENTS:
+
+## Workflow
+
+### Step 1: List All Global Plans
+
+Get the list of all plans from the global plans directory:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-list
+```
+
+This returns JSON:
+```json
+{
+  "success": true,
+  "data": {
+    "plans": ["plan-1", "plan-2", "plan-3"],
+    "count": 3
+  },
+  "message": "Found 3 plan(s)"
+}
+```
+
+### Step 2: Get Details for Each Plan
+
+For each plan in the list, get its format and basic metadata:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get-plan-format {plan_name}
+```
+
+This tells you if it's "conceptual" (requirements only) or "implementation" (executable tasks).
+
+Optionally, for implementation plans, get status:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-status {plan_name}
+```
+
+### Step 3: Display Plan List
+
+Show a formatted list of all plans:
+
+```
+📋 Global Plans ({count} total)
+
+Conceptual Plans (Requirements Only):
+  1. api-redesign
+     ├─ Goal: Redesign API for better performance
+     ├─ Requirements: 12
+     ├─ Created: 3 days ago
+     └─ Next: Use /session:plan-finalize api-redesign to create executable tasks
+
+  2. user-permissions
+     ├─ Goal: Implement role-based permissions system
+     ├─ Requirements: 8
+     ├─ Created: 1 week ago
+     └─ Next: Use /session:plan-finalize user-permissions to create executable tasks
+
+Implementation Plans (Executable):
+  3. oauth-implementation (feature)
+     ├─ Progress: 15/22 tasks (68%)
+     ├─ Status: in-progress
+     ├─ Current: Phase 2 - OAuth Flow Implementation
+     └─ Last Updated: 2 hours ago
+
+  4. database-migration (refactor)
+     ├─ Progress: 8/8 tasks (100%)
+     ├─ Status: completed
+     └─ Completed: 1 day ago
+
+Helpful Commands:
+  /session:plan-status {name}       - Show detailed plan status
+  /session:plan-execute {name}      - Start/continue plan execution
+  /session:plan-finalize {name}     - Transform conceptual plan to executable
+  /session:save-plan {name}         - Create a new plan
+```
+
+### Step 4: Handle Empty Case
+
+If no plans exist, show:
+
+```
+📋 No plans found
+
+You haven't created any plans yet.
+
+Get started:
+  1. Have a conversation about what you want to build
+  2. Run /session:save-plan {name} to capture requirements
+  3. Run /session:plan-finalize {name} to create executable tasks
+  4. Run /session:plan-execute {name} to start implementation
+
+💡 Plans are global and accessible from any session.
+```
+
+---
+
+## Display Formatting
+
+### Plan Type Indicators
+- 📝 Conceptual (requirements captured, not yet broken down into tasks)
+- ⚙️ Implementation (executable tasks, ready to execute)
+
+### Status Icons
+- ✓ Completed
+- → In Progress
+- ○ Pending
+- ⊘ Blocked
+
+### Progress Bars
+For implementation plans with progress:
+```
+Progress: [████████████░░░░░░░░] 68% (15/22 tasks)
+```
+
+---
+
+## Error Handling
+
+- No plans directory: Show empty state message
+- Corrupted plan files: Skip and show warning for specific plan
+- CLI command failures: Show error message with suggestion to run /session:rebuild-index
+
+---
+
+## Notes
+
+- This command does NOT require an active session
+- Plans are stored globally in `.claude/plans/`
+- Both conceptual and implementation plans are shown
+- The list is sorted by last updated (most recent first)
+- Use /session:plan-status {name} for detailed information about a specific plan
diff --git a/commands/plan-save.md b/commands/plan-save.md
new file mode 100644
index 0000000..ce77bfe
--- /dev/null
+++ b/commands/plan-save.md
@@ -0,0 +1,285 @@
+You are executing the /session:save-plan command to create a conceptual plan from the current conversation.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`, `/session:plan-list`, `/session:save-plan`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`, `/plan-list`, `/save-plan`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+Parsed from user input:
+- `plan_name`: {name} (required)
+
+ARGUMENTS: {name}
+
+## Workflow
+
+### Step 1: Check for Active Session (Optional)
+
+Plans are now global and don't require a session. However, if there's an active session, we can extract requirements from the conversation.
+
+Check for active session:
+
+```bash
+[ -f .claude/sessions/.active-session ] && cat .claude/sessions/.active-session || echo "none"
+```
+
+- If result is "none": Skip Steps 2-3, create an empty plan with placeholder requirements (go to Step 4)
+- If there's an active session: Continue to Step 2 to extract from conversation
+
+### Step 2: Read Conversation Log (If Session Active)
+
+**Only execute this step if there's an active session.**
+
+Load the conversation log for the active session. The conversation log file is at:
+`.claude/sessions/{session_name}/conversation-log.jsonl`
+
+**Chunked Reading Strategy:**
+1. Check if file exists
+2. Count lines: `wc -l .claude/sessions/{session_name}/conversation-log.jsonl`
+3. If <= 2000 lines: Read entire file
+4. If > 2000 lines: Read in chunks of 2000 using Read tool's offset/limit parameters
+5. Concatenate all chunks into full conversation log
+
+If the file doesn't exist or is empty:
+- Show warning: "⚠️ No conversation log found. Creating empty plan template."
+- Continue to Step 4 with empty requirements
+
+### Step 3: Detect Work Type (Optional - If Session Active)
+
+Detect the work type from the conversation for metadata purposes:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js detect-work-type {session_name}
+```
+
+This returns a JSON object with:
+```json
+{
+  "type": "feature|bug|spike|refactor|unknown",
+  "confidence": 87,
+  "scores": {...},
+  "signals": {...}
+}
+```
+
+Show the detection result:
+```
+🔍 Analyzing conversation...
+✓ Detected work type: {TYPE} ({confidence}% confidence)
+```
+
+This is for metadata only - conceptual plans don't use templates.
+
+### Step 4: Extract Requirements from Conversation
+
+Use a subagent to analyze the conversation and extract requirements (not tasks!).
+
+Invoke the Task tool with:
+- subagent_type: "general-purpose"
+- model: "haiku"
+- prompt: Read the file at `${CLAUDE_PLUGIN_ROOT}/prompts/analyze-conversation.md`, replace placeholders with actual values, then execute those instructions
+
+The subagent will return extracted requirements:
+```json
+{
+  "goal": "High-level objective",
+  "requirements": [
+    {
+      "id": "req-1",
+      "description": "What the user wants (high-level)",
+      "notes": "Additional context",
+      "open_questions": ["Question 1", "Question 2"],
+      "priority": "high|medium|low",
+      "conversation_context": "Relevant conversation excerpts"
+    }
+  ],
+  "discussion_notes": "Free-form notes from planning discussion",
+  "conversation_summary": "Summary of conversation"
+}
+```
+
+**IMPORTANT:** Extract REQUIREMENTS (what user wants), NOT tasks (how to implement).
+- ✓ Good requirement: "Restrict products based on user permissions"
+- ✗ Bad (too detailed): "Add restriction_level column to public.new_product table"
+
+Requirements are exploratory and high-level during planning.
+
+### Step 4: Build Requirements Plan
+
+Create the requirements.json structure:
+
+```json
+{
+  "plan_name": "{plan_name}",
+  "plan_type": "conceptual",
+  "goal": "{extracted_goal_or_placeholder}",
+  "requirements": [...extracted_requirements_or_empty],
+  "discussion_notes": "{discussion_notes_or_empty}",
+  "conversation_summary": "{conversation_summary_or_empty}",
+  "created_at": "{ISO_8601_timestamp}",
+  "metadata": {
+    "work_type": "{detected_work_type_or_unknown}",
+    "estimated_complexity": "simple|moderate|complex",
+    "source_session": "{session_name_if_available_else_null}"
+  }
+}
+```
+
+**Note:** The `session_name` field is no longer required. Plans are global and can be created with or without session context.
+
+### Step 6: Show Requirements Preview
+
+Display a preview of the requirements to the user:
+
+```
+📋 Requirements Preview: {plan_name}
+
+Goal: {goal}
+Work Type: {type} ({confidence}% confidence)
+
+Requirements Captured:
+  1. {req-1-description}
+     Notes: {req-1-notes}
+     Open Questions: {req-1-questions}
+
+  2. {req-2-description}
+     Notes: {req-2-notes}
+
+  [Show first 3-4 requirements, summarize rest if more]
+
+💡 This is a conceptual plan. Use /session:plan-finalize to transform requirements into executable tasks.
+
+Options:
+  1. ✓ Save this plan (recommended)
+  2. Cancel
+```
+
+### Step 7: Get User Choice
+
+Use the AskUserQuestion tool to get the user's choice.
+
+Handle the response:
+- **Option 1 (Save)**: Continue to Step 8
+- **Option 2 (Cancel)**: Show "Plan creation cancelled" and STOP
+
+### Step 8: Validate Requirements
+
+Validate the requirements against requirements-schema.json:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js validate-requirements '{requirements_json}'
+```
+
+If validation fails, show the errors and STOP:
+```
+❌ Validation errors found:
+  1. {error_1}
+  2. {error_2}
+
+Cannot save invalid plan. Please review.
+```
+
+If validation succeeds, continue.
+
+### Step 9: Save Requirements File
+
+Create the requirements.json file using the global plans directory:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js save-requirements {plan_name} '{requirements_json}'
+```
+
+This creates:
+- `.claude/plans/{plan_name}/requirements.json`
+
+Also create a conversation context markdown file for reference (if conversation was analyzed):
+- Path: `.claude/plans/{plan_name}/conversation-context.md`
+- Content: Include conversation summary, key decisions, requirements discussion, etc.
+- If no session context, skip this file or create with placeholder noting manual plan creation
+
+### Step 10: Display Success
+
+Show success message with next steps:
+
+```
+✓ Global plan saved: {plan_name}
+
+📋 Plan Details:
+   • Type: Conceptual (requirements captured)
+   • Work type: {type} (detected with {confidence}% confidence)
+   • Requirements: {requirement_count}
+   • Location: .claude/plans/{plan_name}/requirements.json
+   • Scope: Global (accessible from any session)
+
+📝 Next Steps:
+
+   1. Review requirements:
+      /session:plan-status {plan_name}
+
+   2. List all plans:
+      /session:plan-list
+
+   3. Transform into executable plan:
+      /session:plan-finalize {plan_name}
+
+      This will use AI to break down requirements into concrete tasks
+      organized by implementation phases (Database, API, UI, etc.)
+
+   4. After finalization, execute:
+      /session:plan-execute {plan_name}
+
+💡 Plans are now global and accessible from any session. No need to be in a specific session to work with plans.
+```
+
+---
+
+## Error Handling
+
+At each step, handle errors gracefully:
+
+- **File not found**: Show clear message with suggestion on how to proceed
+- **Validation failed**: Show specific validation errors
+- **Detection low confidence**: Accept it (work type is just metadata)
+- **Parse errors**: Show error details and abort
+- **CLI command failures**: Check exit codes and show appropriate error messages
+
+Always provide actionable next steps when errors occur.
+
+---
+
+## Key Principles
+
+**Planning Phase = Requirements Capture**
+
+- Users discuss WHAT they want (requirements), not HOW to implement (tasks)
+- Keep it lightweight and exploratory
+- Allow for open questions and uncertainty
+- No forced structure - just capture what was discussed
+
+**Examples of Good Requirements:**
+- "Restrict products based on user permissions"
+- "Track who created each product"
+- "Add audit logging for product changes"
+- "Improve product search performance"
+
+**Examples of Bad (Too Detailed for Planning):**
+- "Add restriction_level column to public.new_product table" ← This is a task!
+- "Create POST /api/products validation" ← This is a task!
+- "Add checkbox in product form component" ← This is a task!
+
+The transformation from requirements → tasks happens in /session:plan-finalize.
+
+---
+
+## Notes
+
+- The ${CLAUDE_PLUGIN_ROOT} environment variable should point to the session plugin source directory
+- Plans are now global - stored in `.claude/plans/` regardless of active session
+- The {session_name} variable is optional - used only if there's an active session for conversation context
+- All CLI commands should use absolute paths
+- Error messages should be user-friendly and actionable
+- The workflow is designed to be interruptible - user can cancel at any point
+- Conceptual plans use requirements.json format (not orchestration.json)
+- Plans can be created with or without session context (conversation analysis is optional)
diff --git a/commands/plan-status.md b/commands/plan-status.md
new file mode 100644
index 0000000..6080eda
--- /dev/null
+++ b/commands/plan-status.md
@@ -0,0 +1,174 @@
+You are executing the /session:plan-status command to show plan execution status.
+
+**NOTE:** Plans are now global and independent of sessions.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`, `/session:plan-list`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`, `/plan-list`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+Parsed from user input:
+- `plan_name`: {name} (optional - if not provided, show status for all plans)
+
+ARGUMENTS: {name}
+
+## Workflow
+
+### Step 1: Get Plan Status
+
+Plans are stored globally in `.claude/plans/` and can be accessed without an active session.
+
+If plan_name is provided, get status for that specific plan:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-status {plan_name}
+```
+
+If no plan_name, list all global plans first:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-list
+```
+
+Then get status for each plan.
+
+### Step 2: Display Status
+
+**For a specific plan:**
+
+```
+📋 Plan Status: {plan_name}
+
+Goal: {goal}
+Work Type: {work_type}
+Scope: Global (accessible from any session)
+Created: {created_date}
+Last Updated: {updated_date}
+
+Overall Progress: {completed_tasks}/{total_tasks} tasks ({percentage}%)
+├─ Completed: {completed}
+├─ In progress: {in_progress}
+├─ Pending: {pending}
+└─ Blocked: {blocked}
+
+Phase Progress: {completed_phases}/{total_phases}
+
+Phases:
+  1. [✓] {phase_1_name} (completed)
+     ├─ Tasks: 5/5 (100%)
+     └─ Duration: {duration}
+
+  2. [→] {phase_2_name} (in-progress)
+     ├─ Tasks: 2/7 (29%)
+     ├─ Current Task: {current_task_id} - {description}
+     └─ Status: {task_status}
+
+  3. [ ] {phase_3_name} (pending)
+     ├─ Tasks: 0/6 (0%)
+     └─ Depends on: Phase 2
+
+  4. [ ] {phase_4_name} (pending)
+     ├─ Tasks: 0/4 (0%)
+     └─ Depends on: Phase 2, Phase 3
+
+Current Task: {current_task_id}
+  Description: {description}
+  Details: {details}
+  Status: {status}
+  Phase: {phase_name}
+
+Next Steps:
+  /session:plan-execute {plan_name} - Continue execution
+  /update-task-status {task_id} completed - Mark current task complete
+```
+
+**For all plans:**
+
+```
+📋 All Global Plans
+
+1. oauth-implementation (feature)
+   ├─ Progress: 15/22 tasks (68%)
+   ├─ Status: in-progress
+   ├─ Current: Phase 2 - OAuth Flow Implementation
+   └─ Last Updated: 2 hours ago
+
+2. database-migration (refactor)
+   ├─ Progress: 8/8 tasks (100%)
+   ├─ Status: completed
+   └─ Completed: 1 day ago
+
+3. api-redesign (spike)
+   ├─ Progress: 0/12 tasks (0%)
+   ├─ Status: pending
+   └─ Created: 3 days ago
+
+Use /session:plan-status {plan_name} for detailed status.
+Use /session:plan-execute {plan_name} to start/continue execution.
+Use /session:plan-list to see all available plans.
+```
+
+### Step 3: Show Recommendations
+
+Based on plan status, show context-aware recommendations:
+
+**If plan is in-progress:**
+```
+💡 Recommendations:
+- Current task is {status} - {recommendation}
+- {next_task_count} tasks remaining in current phase
+- Estimated completion: {estimate}
+```
+
+**If plan is blocked:**
+```
+⚠️ Blocked:
+- Task {task_id} is blocked
+- Blocker: {blocker_reason}
+- Action needed: {action}
+```
+
+**If plan is completed:**
+```
+✅ Plan Complete:
+- All {total_tasks} tasks completed
+- Duration: {total_duration}
+- Consider: Code review, testing, deployment
+```
+
+---
+
+## Display Formats
+
+### Progress Bar
+```
+Progress: [████████████░░░░░░░░] 68% (15/22 tasks)
+```
+
+### Status Icons
+- [✓] Completed
+- [→] In Progress
+- [ ] Pending
+- [✗] Failed
+- [⊘] Blocked
+- [~] Skipped
+
+### Phase Status
+```
+Phase 2: OAuth Flow Implementation (in-progress)
+  ├─ task-2-1: Create OAuth callback routes [✓]
+  ├─ task-2-2: Implement JWT generation [→]
+  ├─ task-2-3: Add session middleware [ ]
+  └─ task-2-4: Configure passport.js [ ]
+```
+
+---
+
+## Error Handling
+
+- Plan not found: Show available plans using /session:plan-list
+- Corrupted plan data: Show error with recovery steps
+- No plans exist: Show message about creating first plan with /session:save-plan {name}
diff --git a/commands/plan-update.md b/commands/plan-update.md
new file mode 100644
index 0000000..58f3785
--- /dev/null
+++ b/commands/plan-update.md
@@ -0,0 +1,344 @@
+You are executing the /session:plan-update command to modify an existing plan.
+
+**NOTE:** Plans are now global and independent of sessions.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:plan-execute`, `/session:plan-status`, `/session:plan-finalize`, `/session:plan-list`, `/session:plan-update`
+- ❌ Wrong: `/plan-execute`, `/plan-status`, `/plan-show`, `/plan-list`, `/plan-update`
+Use ONLY the exact command formats specified in this template.
+
+## Arguments
+
+Parsed from user input:
+- `plan_name`: {name} (required - the plan to update)
+- `update_request`: {request} (optional - natural language description of changes)
+
+ARGUMENTS: {name} {request}
+
+## Workflow
+
+### Step 1: Validate Plan Exists
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-exists {plan_name}
+```
+
+If the plan doesn't exist, show error and STOP:
+```
+❌ Error: Plan '{plan_name}' not found
+
+Use /session:plan-list to see available plans.
+Use /session:plan-save {name} to create a new plan.
+```
+
+### Step 2: Load Current Plan State
+
+Get the current plan state for preview generation:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-status {plan_name}
+```
+
+This returns JSON with plan metadata, phases, tasks, and execution state.
+
+### Step 3: Parse Update Request
+
+If `update_request` is provided, parse it into structured operations.
+
+**Common Update Patterns:**
+
+| Natural Language | Operation |
+|-----------------|-----------|
+| "add a new task to phase X" | `{ type: 'add', target: 'task', data: {...} }` |
+| "remove task-2-3" | `{ type: 'delete', target: 'task', data: { id: 'task-2-3' } }` |
+| "rename phase 1 to 'Setup'" | `{ type: 'update', target: 'phase', data: { id: 'phase-1', name: 'Setup' } }` |
+| "mark task-1-1 as completed" | `{ type: 'update', target: 'task', data: { id: 'task-1-1', status: 'completed' } }` |
+| "add a new phase called Testing" | `{ type: 'add', target: 'phase', data: { name: 'Testing' } }` |
+
+If no `update_request` is provided, prompt the user:
+```
+What changes would you like to make to plan '{plan_name}'?
+
+Examples:
+  - "add a new task 'Write unit tests' to phase 2"
+  - "remove the documentation phase"
+  - "rename task-3-1 to 'API integration'"
+  - "mark phase 1 as completed"
+
+Enter your update request:
+```
+
+### Step 4: Check Execution State
+
+Determine if the plan is being executed:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js get-execution-state {plan_name}
+```
+
+**If plan has started execution:**
+
+Show mode selection:
+```
+⚠️  Plan '{plan_name}' is currently being executed.
+
+Progress: {completed}/{total} tasks ({percentage}%)
+Completed phases: {completed_phases}
+Current phase: {current_phase}
+
+How would you like to proceed?
+
+1. [selective] Apply changes to pending tasks only (recommended)
+   - Completed work is preserved
+   - Only pending tasks/phases can be modified
+   - May need --force for some operations
+
+2. [rollback] Reset and replan (full update)
+   - All progress will be reset to pending
+   - Execution history is preserved in logs
+   - Start fresh with updated plan
+
+3. [cancel] Cancel update
+
+Choice [1/2/3]:
+```
+
+### Step 5: Generate Preview
+
+Show what changes will be made:
+
+```
+╔══════════════════════╗
+║  Plan Update Preview ║
+╚══════════════════════╝
+
+Plan: {plan_name}
+Mode: {selective|rollback}
+
+─── Summary ────────────────────────────────────────────
+  Operations: {count}
+    + Additions: {add_count}
+    ~ Updates: {update_count}
+    - Deletions: {delete_count}
+
+─── Operations ─────────────────────────────────────────
+  + [1] ADD task: "Write unit tests"
+        Phase: phase-2-implementation
+        Description: Create comprehensive test suite
+
+  ~ [2] UPDATE task: task-3-1
+        description: "API endpoints" → "API integration"
+
+  - [3] DELETE phase: phase-4-documentation
+        ⚠ Contains 3 tasks that will also be deleted
+
+─── Warnings ───────────────────────────────────────────
+  ⚠  Deleting phase with completed tasks - use --force
+  ⚠  Some operations affect completed work
+
+─── Safety Notices ─────────────────────────────────────
+  ⚠ DESTRUCTIVE: This update includes 1 deletion(s)
+```
+
+### Step 6: Confirm with User
+
+```
+Apply these changes? [+1 ~1 -1]
+  [y] Yes, apply changes
+  [n] No, cancel
+  [e] Edit operations
+  [?] Show help
+
+Choice:
+```
+
+**If user chooses 'e' (edit):**
+```
+--- Edit Mode ---
+Operations:
+  [1] ADD task: "Write unit tests"
+  [2] UPDATE task: task-3-1
+  [3] DELETE phase: phase-4-documentation
+
+Commands:
+  remove <n>  - Remove operation #n
+  done        - Finish editing and confirm
+  cancel      - Cancel all changes
+
+Edit>
+```
+
+### Step 7: Execute Update
+
+Based on mode selection:
+
+**For selective mode:**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-update-selective {plan_name} --operations '{operations_json}'
+```
+
+**For rollback mode:**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js plan-update-rollback {plan_name} --operations '{operations_json}'
+```
+
+### Step 8: Report Results
+
+**Success:**
+```
+✅ Plan Updated Successfully
+
+Plan: {plan_name}
+Mode: {mode}
+
+Changes Applied:
+  + Added 1 task
+  ~ Updated 1 task
+  - Deleted 1 phase (3 tasks removed)
+
+Updated Progress: {new_completed}/{new_total} tasks
+
+Next Steps:
+  /session:plan-status {plan_name} - View updated plan
+  /session:plan-execute {plan_name} - Continue execution
+```
+
+**Partial Success (selective mode):**
+```
+⚠️ Plan Partially Updated
+
+Applied: 2 operations
+Skipped: 1 operation (affects completed work)
+
+Skipped Operations:
+  - DELETE phase: phase-1-setup (completed, requires --force)
+
+To apply skipped operations, use:
+  /session:plan-update {plan_name} --force
+```
+
+**Failure:**
+```
+❌ Update Failed
+
+Error: {error_message}
+
+Details:
+  - Operation 3 failed: {reason}
+  - Validation error: {details}
+
+Recovery:
+  - Backup created at: .claude/plans/{plan_name}/.backups/backup-{timestamp}
+  - To restore: /session:plan-restore {plan_name} {backup_name}
+```
+
+---
+
+## Quick Update Examples
+
+### Add a Task
+```
+/session:plan-update my-plan add a new task "Implement caching" to phase-2
+```
+
+### Remove a Task
+```
+/session:plan-update my-plan remove task-3-2
+```
+
+### Rename a Phase
+```
+/session:plan-update my-plan rename phase-1 to "Infrastructure Setup"
+```
+
+### Mark Task Complete
+```
+/session:plan-update my-plan mark task-2-1 as completed
+```
+
+### Add a New Phase
+```
+/session:plan-update my-plan add a new phase called "Performance Testing" after phase-3
+```
+
+### Force Update (completed items)
+```
+/session:plan-update my-plan remove task-1-1 --force
+```
+
+---
+
+## Command Options
+
+| Option | Description |
+|--------|-------------|
+| `--force` | Allow modifications to completed items |
+| `--yes` | Skip confirmation prompt |
+| `--dry-run` | Preview changes without applying |
+| `--mode selective` | Only update pending items (default for executing plans) |
+| `--mode rollback` | Reset progress and apply all changes |
+| `--json` | Output results as JSON |
+
+---
+
+## Error Handling
+
+### Plan Not Found
+```
+❌ Error: Plan 'unknown-plan' not found
+
+Available plans:
+  - my-feature (in-progress, 45%)
+  - refactor-api (pending, 0%)
+
+Use /session:plan-list for full details.
+```
+
+### Invalid Operation
+```
+❌ Error: Could not parse update request
+
+Input: "do something to the plan"
+
+Could not determine:
+  - Operation type (add, remove, update?)
+  - Target (phase, task?)
+
+Try being more specific:
+  - "add a task called 'X' to phase Y"
+  - "remove task-1-2"
+  - "rename phase-1 to 'New Name'"
+```
+
+### Blocked by Execution State
+```
+❌ Error: Cannot modify in-progress task
+
+Task 'task-2-3' is currently in progress.
+Wait for completion or use rollback mode.
+
+Options:
+  1. Wait for task to complete
+  2. Use /session:plan-update {plan} --mode rollback
+```
+
+### Validation Error
+```
+❌ Error: Invalid operation
+
+Cannot add task with duplicate ID 'task-1-1'
+A task with this ID already exists in phase-1.
+
+Suggestion: Let the system generate a unique ID automatically.
+```
+
+---
+
+## Related Commands
+
+- `/session:plan-status {name}` - View current plan status
+- `/session:plan-execute {name}` - Execute plan tasks
+- `/session:plan-list` - List all plans
+- `/session:plan-save {name}` - Create new plan
+- `/session:plan-finalize {name}` - Finalize conceptual plan
diff --git a/commands/project-maps-generate.md b/commands/project-maps-generate.md
new file mode 100644
index 0000000..f2fd46c
--- /dev/null
+++ b/commands/project-maps-generate.md
@@ -0,0 +1,212 @@
+# Project Maps Generate Command
+
+Generate project context maps for the current project.
+
+## Usage
+
+```
+/project-maps-generate [--path <project-path>]
+```
+
+## Options
+
+- `--path <path>`: Project path to scan (defaults to current directory)
+- No options: Generate maps for current working directory
+
+## Implementation
+
+### Step 1: Determine Project Path
+
+Parse command arguments to determine project path:
+
+```javascript
+const args = process.argv.slice(2);
+const pathIndex = args.indexOf('--path');
+const projectPath = pathIndex >= 0 && args[pathIndex + 1]
+  ? args[pathIndex + 1]
+  : process.cwd();
+```
+
+Default to current working directory if no path specified.
+
+### Step 2: Check if Maps Already Exist
+
+Check if maps have already been generated for this project:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js {project_path} --staleness-only 2>&1
+```
+
+If maps exist (exit code 0), ask user if they want to regenerate:
+
+```
+⚠️  Maps already exist for this project
+
+Existing maps:
+  Location: {maps_dir}
+  Last generated: {timestamp}
+  Staleness: {score}/100
+
+Regenerate maps? This will overwrite existing maps.
+  • Yes, regenerate (will take ~10-30 seconds)
+  • No, use refresh instead (faster, preserves existing data)
+```
+
+If user chooses "No", suggest using refresh instead:
+```
+Tip: Use /project-maps-refresh for faster updates
+```
+
+### Step 3: Initialize and Scan Project
+
+Show progress and start generation:
+
+```
+Generating project context maps...
+
+Project: {project_path}
+Project hash: {hash}
+```
+
+Run the map generator:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-generator.js {project_path}
+```
+
+The map generator will:
+1. Initialize configuration
+2. Scan file system
+3. Generate 11 map files (Phases 1-3)
+4. Apply compression
+5. Save to `~/.claude/project-maps/{project-hash}/`
+
+### Step 4: Show Progress
+
+The map generator outputs progress:
+
+```
+Scanning project...
+Scanned 145 files in 127ms
+
+Generating maps...
+  ✓ summary.json (2.1 KB → 907 B, 56% compression)
+  ✓ tree.json (3.2 KB → 1.4 KB, 56% compression)
+  ✓ metadata.json (12.5 KB → 5.2 KB, 58% compression)
+  ✓ content-summaries.json (1.1 KB → 465 B, 58% compression)
+  ✓ indices.json (11.8 KB → 5.0 KB, 58% compression)
+  ✓ existence-proofs.json (4.0 KB → 1.7 KB, 58% compression)
+  ✓ quick-queries.json (6.4 KB → 2.7 KB, 58% compression)
+  ✓ dependencies-forward.json (1.1 KB → 477 B, 57% compression)
+  ✓ dependencies-reverse.json (1.1 KB → 445 B, 60% compression)
+  ✓ relationships.json (1.5 KB → 630 B, 58% compression)
+  ✓ issues.json (1.3 KB → 557 B, 57% compression)
+```
+
+### Step 5: Display Summary
+
+After generation completes, show summary:
+
+```
+✓ Maps generated successfully!
+
+Summary:
+  Files scanned: {file_count}
+  Maps created: 11
+  Storage location: {maps_dir}
+
+  Total size:
+    • Original: {original_size} KB
+    • Compressed: {compressed_size} KB
+    • Compression: {compression_ratio}%
+
+  Maps by tier:
+    • Tier 1 (Always loaded): summary.json, quick-queries.json
+    • Tier 2 (On demand): tree.json, existence-proofs.json
+    • Tier 3 (When needed): metadata.json, content-summaries.json, indices.json
+    • Tier 4 (Deep analysis): dependencies-*.json, relationships.json, issues.json
+
+Next steps:
+  • Load maps: /project-maps-load
+  • Quick queries: /project-maps-query
+  • Refresh maps: /project-maps-refresh
+```
+
+### Step 6: Verify Generation
+
+Verify all expected files were created:
+
+```bash
+ls -lh ~/.claude/project-maps/{project_hash}/*.json | wc -l
+```
+
+Expected: 11 files
+
+If count != 11, show warning:
+```
+⚠️  Warning: Expected 11 map files but found {actual_count}
+Some maps may have failed to generate.
+```
+
+## Error Handling
+
+**Project path doesn't exist:**
+```
+❌ Error: Project path not found
+
+Path: {project_path}
+Please check the path and try again.
+```
+
+**Permission denied:**
+```
+❌ Error: Permission denied
+
+Cannot write to: ~/.claude/project-maps/
+Check directory permissions.
+```
+
+**Generation failed:**
+```
+❌ Map generation failed: {error_message}
+
+Possible causes:
+  1. Insufficient disk space
+  2. Invalid project structure
+  3. File access permissions
+
+Try:
+  • Check disk space
+  • Verify project directory is accessible
+  • Run with --verbose for more details
+```
+
+**No files found:**
+```
+⚠️  No files found in project
+
+The project appears to be empty or all files are excluded.
+Check your .gitignore or .projectmaprc configuration.
+```
+
+## Performance Notes
+
+- Small projects (<100 files): ~5-10 seconds
+- Medium projects (100-1000 files): ~10-30 seconds
+- Large projects (1000+ files): ~30-60 seconds
+- Generation happens once, subsequent updates use incremental refresh (much faster)
+
+## Examples
+
+```bash
+# Generate maps for current directory
+/project-maps-generate
+
+# Generate maps for specific project
+/project-maps-generate --path /path/to/project
+
+# Generate with verbose output
+/project-maps-generate --verbose
+```
diff --git a/commands/project-maps-list.md b/commands/project-maps-list.md
new file mode 100644
index 0000000..5579fc9
--- /dev/null
+++ b/commands/project-maps-list.md
@@ -0,0 +1,120 @@
+# Project Maps List Command
+
+List all projects with generated context maps.
+
+## Usage
+
+```
+/project-maps-list
+```
+
+## Implementation
+
+### Step 1: Scan Maps Directory
+
+List all subdirectories in the maps storage:
+
+```bash
+ls -d ~/.claude/project-maps/*/ 2>/dev/null
+```
+
+Each subdirectory represents a project hash.
+
+### Step 2: Load Summary for Each Project
+
+For each project hash, load the summary.json to get project details:
+
+```javascript
+const fs = require('fs').promises;
+const path = require('path');
+const MapLoader = require('./map-loader');
+
+const mapsBaseDir = path.join(process.env.HOME, '.claude/project-maps');
+const projectDirs = await fs.readdir(mapsBaseDir);
+
+for (const hash of projectDirs) {
+  if (hash === 'maps' || hash === 'schemas' || hash === 'temp') continue;
+
+  try {
+    const summaryPath = path.join(mapsBaseDir, hash, 'summary.json');
+    const summary = await loadSummary(summaryPath);
+    // Display project info
+  } catch (error) {
+    // Skip invalid projects
+  }
+}
+```
+
+### Step 3: Display Projects List
+
+Show all projects with their metadata:
+
+```
+Project Maps
+
+Found {count} projects with generated maps:
+
+1. {project_name}
+   Path: {project_path}
+   Hash: {project_hash}
+   Files: {file_count}
+   Generated: {timestamp}
+   Staleness: {score}/100 ({level})
+   Size: {total_size}
+
+2. {project_name}
+   Path: {project_path}
+   Hash: {project_hash}
+   Files: {file_count}
+   Generated: {timestamp}
+   Staleness: {score}/100 ({level})
+   Size: {total_size}
+
+Total storage: {total_storage_size}
+
+Commands:
+  • Load project: /project-maps-load
+  • Refresh: /project-maps-refresh
+  • Generate new: /project-maps-generate
+```
+
+### Step 4: Highlight Stale Projects
+
+Mark projects that need refresh:
+
+```
+⚠️  The following projects may need refresh:
+  • {project_name} - Staleness: {score}/100
+  • {project_name} - Staleness: {score}/100
+
+Run: /project-maps-refresh to update
+```
+
+## Error Handling
+
+**No maps found:**
+```
+No project maps found
+
+Storage location: ~/.claude/project-maps/
+
+Generate maps for current project:
+  /project-maps-generate
+```
+
+**Corrupt project data:**
+```
+⚠️  Found {count} projects, but {error_count} have corrupt data:
+  • {hash} - Cannot read summary
+  • {hash} - Missing metadata
+
+Consider removing corrupt entries:
+  rm -rf ~/.claude/project-maps/{hash}
+```
+
+## Examples
+
+```bash
+# List all projects with maps
+/project-maps-list
+```
diff --git a/commands/project-maps-load.md b/commands/project-maps-load.md
new file mode 100644
index 0000000..5f38b20
--- /dev/null
+++ b/commands/project-maps-load.md
@@ -0,0 +1,301 @@
+# Project Maps Load Command
+
+Load and display project context maps with tiered loading.
+
+## Usage
+
+```
+/project-maps-load [--tier <1|2|3|4>] [--map <map-name>]
+```
+
+## Options
+
+- `--tier <N>`: Load all maps from specific tier (1, 2, 3, or 4)
+- `--map <name>`: Load specific map by name
+- No options: Load Tier 1 (summary + quick-queries)
+
+## Tier Levels
+
+**Tier 1** (Always loaded, ~3KB):
+- `summary.json` - Project overview and statistics
+- `quick-queries.json` - Pre-computed answers to common questions
+
+**Tier 2** (On demand, ~10KB):
+- `tree.json` - Directory structure
+- `existence-proofs.json` - File manifests and negative space
+
+**Tier 3** (When needed, ~60KB):
+- `metadata.json` - Comprehensive file metadata
+- `content-summaries.json` - Exports, imports, entities
+- `indices.json` - Navigation indices by type/role/size/recency
+
+**Tier 4** (Deep analysis, ~10KB):
+- `dependencies-forward.json` - Import graphs
+- `dependencies-reverse.json` - Who imports what
+- `relationships.json` - Dependency chains
+- `issues.json` - Broken imports, circular deps, unused files
+
+## Implementation
+
+### Step 1: Check if Maps Exist
+
+First, verify maps exist for current project:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js . --staleness-only
+```
+
+If no maps exist (exit code 1), show error:
+```
+❌ No project maps found
+
+Generate maps first:
+  /project-maps-generate
+```
+
+### Step 2: Parse Arguments and Determine Load Mode
+
+Parse command arguments:
+
+```javascript
+const args = process.argv.slice(2);
+let loadMode = 'tier';
+let tierLevel = 1;
+let mapName = null;
+
+if (args.includes('--tier')) {
+  const tierIndex = args.indexOf('--tier');
+  tierLevel = parseInt(args[tierIndex + 1]) || 1;
+  loadMode = 'tier';
+}
+
+if (args.includes('--map')) {
+  const mapIndex = args.indexOf('--map');
+  mapName = args[mapIndex + 1];
+  loadMode = 'single';
+}
+```
+
+### Step 3: Load Maps
+
+**For Tier Loading:**
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js . --tier {tier_level}
+```
+
+**For Single Map Loading:**
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js . {map_name}
+```
+
+The loader will:
+1. Decompress maps automatically
+2. Check staleness and show warnings
+3. Return the map data
+
+### Step 4: Display Results
+
+**Tier 1 Display (Summary):**
+
+```
+✓ Tier 1 maps loaded
+
+Project Overview:
+  Name: {project_name}
+  Path: {project_path}
+  Files: {total_files}
+  Size: {total_size}
+  Primary languages: {languages}
+
+Framework: {framework_name} ({framework_type})
+
+Quick Answers:
+  • Entry points: {entry_points}
+  • Test location: {test_location}
+  • Largest files: {largest_files}
+  • Recently modified: {recent_files}
+
+Staleness: {staleness_score}/100 ({staleness_level})
+
+Next steps:
+  • Load more detail: /project-maps-load --tier 2
+  • Query specific info: /project-maps-query
+  • Refresh maps: /project-maps-refresh
+```
+
+**Tier 2 Display (Structure):**
+
+```
+✓ Tier 2 maps loaded
+
+Directory Structure:
+{tree_visualization}
+
+Top-level directories:
+  • src/ ({file_count} files)
+  • tests/ ({file_count} files)
+  • config/ ({file_count} files)
+
+File Existence:
+  ✓ package.json
+  ✓ tsconfig.json
+  ✓ .gitignore
+  ✗ jest.config.js (missing)
+  ✗ .env.example (missing)
+```
+
+**Tier 3 Display (Detailed Metadata):**
+
+```
+✓ Tier 3 maps loaded
+
+Detailed File Metadata: {total_files} files
+
+By Type:
+  • TypeScript: {ts_count} files ({ts_size})
+  • JavaScript: {js_count} files ({js_size})
+  • JSON: {json_count} files ({json_size})
+
+By Role:
+  • Source: {source_count} files
+  • Tests: {test_count} files
+  • Config: {config_count} files
+
+Content Summaries:
+  Top exports:
+    • {file}: {exports}
+    • {file}: {exports}
+
+Indices loaded:
+  • By type: {type_count} types
+  • By role: {role_count} roles
+  • By size: {size_buckets} buckets
+  • By recency: Last {recent_days} days
+```
+
+**Tier 4 Display (Dependencies):**
+
+```
+✓ Tier 4 maps loaded
+
+Dependency Analysis:
+
+Forward Dependencies:
+  • {file} imports from {import_count} sources
+  • {file} imports from {import_count} sources
+
+Reverse Dependencies (Most imported):
+  1. {file} - imported by {count} files
+  2. {file} - imported by {count} files
+  3. {file} - imported by {count} files
+
+Issues Detected:
+  ⚠️  Circular dependencies: {circular_count}
+  ⚠️  Broken imports: {broken_count}
+  ⚠️  Unused files: {unused_count}
+
+Relationship Metrics:
+  • Max dependency depth: {max_depth}
+  • Average dependencies per file: {avg_deps}
+  • Tightly coupled modules: {coupled_count}
+```
+
+**Single Map Display:**
+
+Show the specific map data in formatted JSON:
+
+```
+✓ Map loaded: {map_name}
+
+{formatted_json_output}
+
+Staleness: {score}/100
+Size: {compressed_size} compressed ({original_size} original, {ratio}% compression)
+```
+
+### Step 5: Staleness Warning
+
+If staleness score >= 30, show warning (automatically done by loader):
+
+```
+⚠️  Maps may be outdated
+   Staleness: {score}/100 - {recommendation}
+   Consider running: /project-maps-refresh --incremental
+```
+
+## Error Handling
+
+**Maps not found:**
+```
+❌ No project maps found
+
+Generate maps first:
+  /project-maps-generate
+```
+
+**Invalid tier:**
+```
+❌ Invalid tier level: {tier}
+
+Valid tiers: 1, 2, 3, or 4
+```
+
+**Map not found:**
+```
+❌ Map not found: {map_name}
+
+Available maps:
+  • summary
+  • tree
+  • metadata
+  • content-summaries
+  • indices
+  • existence-proofs
+  • quick-queries
+  • dependencies-forward
+  • dependencies-reverse
+  • relationships
+  • issues
+```
+
+**Load failed:**
+```
+❌ Failed to load maps: {error_message}
+
+Try:
+  1. Regenerate maps: /project-maps-generate
+  2. Check file permissions
+  3. Verify maps directory exists
+```
+
+## Performance Notes
+
+- Tier 1: Instant load (~3KB)
+- Tier 2: Very fast (~10KB)
+- Tier 3: Fast (~60KB)
+- Tier 4: Fast (~10KB)
+- Single map: Depends on map size
+- All decompression happens automatically
+
+## Examples
+
+```bash
+# Load Tier 1 (default)
+/project-maps-load
+
+# Load Tier 2 (directory structure)
+/project-maps-load --tier 2
+
+# Load Tier 3 (detailed metadata)
+/project-maps-load --tier 3
+
+# Load Tier 4 (dependencies)
+/project-maps-load --tier 4
+
+# Load specific map
+/project-maps-load --map summary
+/project-maps-load --map dependencies-forward
+```
diff --git a/commands/project-maps-query.md b/commands/project-maps-query.md
new file mode 100644
index 0000000..549020c
--- /dev/null
+++ b/commands/project-maps-query.md
@@ -0,0 +1,297 @@
+# Project Maps Query Command
+
+Query pre-computed answers from project context maps.
+
+## Usage
+
+```
+/project-maps-query <query-type>
+```
+
+## Query Types
+
+- `entry-points` - What are the entry points?
+- `framework` - Which framework is used?
+- `tests` - Where are the tests?
+- `largest` - What are the largest files?
+- `recent` - What changed recently?
+- `structure` - What's the project structure?
+- `languages` - What languages are used?
+- `all` - Show all available queries
+
+## Implementation
+
+### Step 1: Check if Maps Exist
+
+Verify maps exist for current project:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js . --staleness-only
+```
+
+If no maps exist, show error and stop.
+
+### Step 2: Parse Query Type
+
+Extract query type from arguments:
+
+```javascript
+const args = process.argv.slice(2);
+const queryType = args[0] || 'all';
+```
+
+### Step 3: Load Quick Queries Map
+
+Load the pre-computed quick-queries map:
+
+```javascript
+const MapLoader = require('./map-loader');
+const loader = new MapLoader(process.cwd());
+const quickQueries = await loader.load('quick-queries');
+```
+
+### Step 4: Execute Query and Display Results
+
+Based on query type, extract and display the relevant information:
+
+**Entry Points Query:**
+
+```
+Query: What are the entry points?
+
+Entry Points:
+  1. src/index.ts - Main application entry
+  2. src/server.ts - Server entry
+  3. src/cli.ts - CLI entry
+
+Description:
+  These files serve as the main entry points to the application.
+  They are typically referenced in package.json or imported first.
+
+Related files:
+  • package.json: "main": "src/index.ts"
+  • package.json: "bin": "src/cli.ts"
+```
+
+**Framework Query:**
+
+```
+Query: Which framework is used?
+
+Framework: React
+Type: Frontend (Single Page Application)
+Version: Detected from package.json
+
+Key indicators:
+  • package.json dependencies: react, react-dom
+  • File patterns: *.jsx, *.tsx files in src/
+  • Common React patterns detected
+
+Related:
+  • Build tool: Vite
+  • Testing: Jest + React Testing Library
+  • State management: Redux
+```
+
+**Tests Query:**
+
+```
+Query: Where are the tests?
+
+Test Files: {test_count} files
+
+Locations:
+  • tests/ - {count} files
+  • src/**/*.test.ts - {count} files
+  • src/**/*.spec.ts - {count} files
+
+Test Frameworks:
+  • Jest - Main test runner
+  • React Testing Library - Component tests
+
+Coverage:
+  • Source files: {source_count}
+  • Test files: {test_count}
+  • Coverage ratio: {ratio}%
+```
+
+**Largest Files Query:**
+
+```
+Query: What are the largest files?
+
+Largest Files (by size):
+  1. dist/bundle.js - 2.4 MB (generated)
+  2. src/components/Dashboard.tsx - 842 KB
+  3. src/utils/helpers.ts - 567 KB
+  4. node_modules/.bin/webpack - 234 KB (dependency)
+  5. src/data/constants.json - 189 KB
+
+Recommendations:
+  ⚠️  Dashboard.tsx is unusually large (842 KB)
+     Consider splitting into smaller components
+
+  ⚠️  helpers.ts contains many utilities (567 KB)
+     Consider organizing into separate modules
+```
+
+**Recent Changes Query:**
+
+```
+Query: What changed recently?
+
+Recently Modified Files (last 7 days):
+  1. src/components/Login.tsx - 2 hours ago
+  2. src/api/auth.ts - 5 hours ago
+  3. tests/auth.test.ts - 5 hours ago
+  4. src/types/user.ts - 1 day ago
+  5. README.md - 3 days ago
+
+Activity Summary:
+  • Most active directory: src/components/ ({count} changes)
+  • Most active author: {author} ({count} commits)
+  • Files modified: {count}
+```
+
+**Structure Query:**
+
+```
+Query: What's the project structure?
+
+Project Structure:
+
+├── src/                    ({file_count} files)
+│   ├── components/         ({file_count} files)
+│   ├── utils/              ({file_count} files)
+│   ├── api/                ({file_count} files)
+│   └── types/              ({file_count} files)
+├── tests/                  ({file_count} files)
+├── config/                 ({file_count} files)
+└── dist/                   ({file_count} files)
+
+Conventions:
+  • Source code: src/
+  • Tests: tests/ and *.test.ts
+  • Configuration: config/
+  • Build output: dist/
+```
+
+**Languages Query:**
+
+```
+Query: What languages are used?
+
+Primary Languages:
+  1. TypeScript - {count} files ({percentage}%)
+  2. JavaScript - {count} files ({percentage}%)
+  3. JSON - {count} files ({percentage}%)
+  4. Markdown - {count} files ({percentage}%)
+
+Language Distribution:
+  [▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░] TypeScript (75%)
+  [▓▓▓░░░░░░░░░░░░░░░] JavaScript (15%)
+  [▓░░░░░░░░░░░░░░░░░] JSON (5%)
+  [░░░░░░░░░░░░░░░░░░] Other (5%)
+```
+
+**All Queries:**
+
+```
+Available Queries:
+
+1. entry-points - What are the entry points?
+2. framework - Which framework is used?
+3. tests - Where are the tests?
+4. largest - What are the largest files?
+5. recent - What changed recently?
+6. structure - What's the project structure?
+7. languages - What languages are used?
+
+Usage:
+  /project-maps-query <query-type>
+
+Example:
+  /project-maps-query entry-points
+  /project-maps-query framework
+```
+
+### Step 5: Show Staleness If Relevant
+
+If staleness score >= 30, append warning:
+
+```
+⚠️  Note: Maps are {staleness_score}/100 stale
+   Query results may be outdated. Consider refreshing:
+   /project-maps-refresh
+```
+
+## Error Handling
+
+**Maps not found:**
+```
+❌ No project maps found
+
+Generate maps first:
+  /project-maps-generate
+```
+
+**Invalid query type:**
+```
+❌ Unknown query type: {query_type}
+
+Available query types:
+  • entry-points
+  • framework
+  • tests
+  • largest
+  • recent
+  • structure
+  • languages
+  • all
+
+Usage:
+  /project-maps-query <query-type>
+```
+
+**No data available:**
+```
+⚠️  No data available for query: {query_type}
+
+This might happen if:
+  1. The project doesn't have this information
+  2. Maps need to be regenerated
+  3. The data wasn't captured during generation
+
+Try:
+  /project-maps-refresh
+```
+
+## Performance Notes
+
+- Queries are instant (pre-computed during generation)
+- No file system scanning required
+- Answers come from quick-queries.json (~3KB)
+- Perfect for getting quick insights without deep analysis
+
+## Examples
+
+```bash
+# What are the entry points?
+/project-maps-query entry-points
+
+# Which framework is used?
+/project-maps-query framework
+
+# Where are the tests?
+/project-maps-query tests
+
+# What are the largest files?
+/project-maps-query largest
+
+# What changed recently?
+/project-maps-query recent
+
+# Show all available queries
+/project-maps-query all
+```
diff --git a/commands/project-maps-refresh.md b/commands/project-maps-refresh.md
new file mode 100644
index 0000000..9d7cdd9
--- /dev/null
+++ b/commands/project-maps-refresh.md
@@ -0,0 +1,176 @@
+# Project Maps Refresh Command
+
+Refresh project context maps (full or incremental).
+
+## Usage
+
+```
+/project-maps-refresh [--full|--incremental]
+```
+
+## Options
+
+- `--full`: Perform full rescan and regenerate all maps (slower, complete)
+- `--incremental`: Only update changed files (faster, recommended for small changes)
+- No option: Auto-detect based on changes (incremental if <30% files changed, otherwise full)
+
+## Implementation
+
+### Step 1: Check if Maps Exist
+
+First, check if maps have been generated for this project:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-loader.js . --staleness-only
+```
+
+If no maps exist (exit code 1), show error and suggest running generation first:
+```
+❌ No project maps found
+
+Generate maps first:
+  /project-maps-generate
+
+Or manually:
+  node session/cli/lib/map-generator.js generate
+```
+
+### Step 2: Determine Refresh Mode
+
+Parse command arguments to determine refresh mode:
+- If user specified `--full`: Use full refresh
+- If user specified `--incremental`: Use incremental refresh
+- If no option: Auto-detect
+
+For auto-detect:
+1. Load summary map to get project hash and last refresh info
+2. Count changed files using git diff
+3. If changed files > 30% of total: Recommend full refresh
+4. Otherwise: Use incremental refresh
+
+### Step 3: Perform Refresh
+
+**For Full Refresh:**
+
+Run the map generator to regenerate all maps:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/map-generator-cli.js refresh --full --project .
+```
+
+This will:
+- Rescan entire project
+- Regenerate all 11 maps
+- Update staleness metadata
+- Show completion statistics
+
+**For Incremental Refresh:**
+
+Run the incremental updater:
+
+```bash
+cd {working_directory}
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/incremental-updater-cli.js --project . --hash {project_hash}
+```
+
+This will:
+- Use git diff to find changed files
+- Rescan only changed files
+- Update affected map entries
+- Regenerate dependency chains
+- Much faster than full refresh
+
+### Step 4: Show Results
+
+After refresh completes, display results:
+
+**Full Refresh:**
+```
+✓ Full refresh completed
+
+Maps regenerated:
+  • summary.json (Level 1)
+  • tree.json (Level 2)
+  • metadata.json (Level 3)
+  • content-summaries.json (Level 3)
+  • indices.json (Level 3)
+  • existence-proofs.json (Level 2)
+  • quick-queries.json (Level 1)
+  • dependencies-forward.json (Level 4)
+  • dependencies-reverse.json (Level 4)
+  • relationships.json (Level 4)
+  • issues.json (Level 4)
+
+Statistics:
+  Files scanned: {total_files}
+  Time taken: {time_ms}ms
+  Maps location: {maps_dir}
+
+Staleness: 0/100 (fresh)
+```
+
+**Incremental Refresh:**
+```
+✓ Incremental refresh completed
+
+Files updated:
+  • Modified: {modified_count}
+  • Added: {added_count}
+  • Deleted: {deleted_count}
+
+Maps updated:
+  • metadata.json
+  • content-summaries.json
+  • dependencies-forward.json
+  • dependencies-reverse.json
+
+Statistics:
+  Files scanned: {scanned_count}
+  Time taken: {time_ms}ms
+  Time saved: ~{time_saved_percent}% vs full rescan
+
+Staleness: 0/100 (fresh)
+```
+
+### Step 5: Verification
+
+After refresh, automatically verify staleness:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/lib/staleness-checker.js . {maps_dir}/summary.json
+```
+
+Expected result: Score should be 0/100 (fresh)
+
+## Error Handling
+
+**Git not available:**
+```
+⚠️  Warning: Git not available, falling back to full refresh
+```
+
+**Too many changes for incremental:**
+```
+⚠️  Too many files changed ({change_percent}%)
+
+Incremental refresh may miss dependencies.
+Recommend: /project-maps-refresh --full
+```
+
+**Refresh failed:**
+```
+❌ Refresh failed: {error_message}
+
+Try:
+  1. Check git status
+  2. Ensure no file permission issues
+  3. Run full refresh: /project-maps-refresh --full
+```
+
+## Notes
+
+- Incremental refresh is faster (target: <20% of full scan time)
+- Full refresh guarantees consistency
+- Refresh automatically updates staleness metadata
+- Maps are compressed automatically during refresh
diff --git a/commands/project-maps-stats.md b/commands/project-maps-stats.md
new file mode 100644
index 0000000..376e9cf
--- /dev/null
+++ b/commands/project-maps-stats.md
@@ -0,0 +1,139 @@
+# Project Maps Stats Command
+
+Show compression metrics and system statistics for project maps.
+
+## Usage
+
+```
+/project-maps-stats
+```
+
+## Implementation
+
+### Step 1: Scan All Project Maps
+
+Scan the maps directory and collect statistics:
+
+```bash
+du -sh ~/.claude/project-maps/
+ls -lR ~/.claude/project-maps/ | wc -l
+```
+
+### Step 2: Calculate Compression Metrics
+
+For each project, read compression metadata:
+
+```javascript
+const allStats = {
+  totalProjects: 0,
+  totalFiles: 0,
+  totalMaps: 0,
+  originalSize: 0,
+  compressedSize: 0,
+  totalStorage: 0
+};
+
+// Aggregate from all summary.json files
+for (const projectHash of projectHashes) {
+  const summary = await loadSummary(projectHash);
+  allStats.totalFiles += summary.statistics.totalFiles;
+  // ... aggregate other stats
+}
+```
+
+### Step 3: Display Statistics
+
+Show comprehensive system stats:
+
+```
+Project Maps Statistics
+
+System Overview:
+  Projects mapped: {project_count}
+  Total files scanned: {total_files}
+  Total maps generated: {total_maps}
+  Storage location: ~/.claude/project-maps/
+
+Storage Metrics:
+  Total storage used: {total_storage} MB
+  Original data size: {original_size} MB
+  Compressed size: {compressed_size} MB
+  Space saved: {space_saved} MB ({savings_percent}%)
+
+Compression Efficiency:
+  Average compression ratio: {avg_ratio}%
+  Best compression: {best_ratio}% ({project_name})
+  Worst compression: {worst_ratio}% ({project_name})
+
+  Compression breakdown:
+    • Level 1 (minification): {level1_ratio}%
+    • Level 2 (key abbreviation): {level2_ratio}%
+    • Level 3 (deduplication): {level3_ratio}%
+
+Map Distribution:
+  • Per project: 11 maps (fixed)
+  • Total maps: {total_maps}
+
+Performance:
+  • Average generation time: {avg_gen_time}ms
+  • Average map size: {avg_map_size} KB
+  • Staleness distribution:
+      Fresh (0-30): {fresh_count} projects
+      Moderate (30-60): {moderate_count} projects
+      Critical (60+): {critical_count} projects
+
+Recommendations:
+  ⚠️  {stale_count} projects need refresh
+  ✓  Storage efficiency: {efficiency}%
+  💡 Consider cleaning old maps: /project-maps-clean
+```
+
+### Step 4: Per-Project Breakdown
+
+Optionally show detailed per-project stats:
+
+```
+Per-Project Breakdown:
+
+1. {project_name}
+   Maps: 11
+   Original: {original_size} KB
+   Compressed: {compressed_size} KB
+   Ratio: {ratio}%
+   Staleness: {score}/100
+
+2. {project_name}
+   Maps: 11
+   Original: {original_size} KB
+   Compressed: {compressed_size} KB
+   Ratio: {ratio}%
+   Staleness: {score}/100
+```
+
+## Error Handling
+
+**No maps found:**
+```
+No project maps found
+
+Generate maps first:
+  /project-maps-generate
+```
+
+**Cannot calculate stats:**
+```
+⚠️  Unable to calculate complete statistics
+
+Some map files may be corrupt or inaccessible.
+Try regenerating maps for affected projects.
+```
+
+## Examples
+
+```bash
+# Show system statistics
+/project-maps-stats
+
+# Show detailed per-project stats
+/project-maps-stats --detailed
+```
diff --git a/commands/save.md b/commands/save.md
new file mode 100644
index 0000000..4253501
--- /dev/null
+++ b/commands/save.md
@@ -0,0 +1,215 @@
+You are managing a session memory system. The user wants to manually save a snapshot of the current session state.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:list`, `/session:start`, `/session:continue`, `/session:status`
+- ❌ Wrong: `/session list`, `/session start`, `/session continue`, `/session status`
+Use ONLY the exact command formats specified in this template.
+
+## Task: Save Session Snapshot
+
+Capture the current context and save it as a timestamped snapshot.
+
+**CRITICAL OPTIMIZATION:** This command now uses CLI for snapshot writes, enabling plan mode support.
+
+### Step 1: Validate Active Session
+
+1. Check if `.claude/sessions/.active-session` file exists
+2. If NOT exists, show error:
+   ```
+   ❌ Error: No active session
+   💡 Use /session:start [name] to create a new session
+   💡 Or use /session:continue [name] to resume an existing session
+   ```
+   Then STOP.
+3. Read the active session name from `.active-session`
+4. Verify the session directory `.claude/sessions/{active_session}/` exists
+5. If NOT exists, show error:
+   ```
+   ❌ Error: Active session directory not found
+   The session may have been deleted. Please start a new session.
+   ```
+   Then STOP.
+
+### Step 2: Capture Conversation Context
+
+Analyze the recent conversation (last 10-20 exchanges) and create a summary:
+- What topics were discussed
+- What problems were solved
+- What decisions were made
+- What tasks were attempted or completed
+
+Keep this summary concise but informative (3-5 sentences).
+
+### Step 3: Capture Completed Todos
+
+Look for any todos that were marked as completed in this conversation:
+- Check for completed checkboxes in messages
+- Check for explicit "completed" or "done" mentions
+- List them in checkbox format: `- [x] Task description`
+
+If no todos were completed, write: `- No todos completed in this segment`
+
+### Step 4: Capture Files Modified
+
+Identify all files that were modified in this conversation:
+- Files written with Write tool
+- Files edited with Edit tool
+- Files created or modified
+
+For each file, capture:
+- File path
+- Brief description of changes made (1-2 lines)
+
+Format:
+```markdown
+### path/to/file.ts
+- Added authentication middleware
+- Updated error handling
+```
+
+If no files were modified, write: `No files modified in this segment`
+
+### Step 5: Capture Current State
+
+Describe the current state of the work:
+- What's working now
+- What's in progress
+- What's blocked
+- What's next
+
+Format as bullet points (3-5 items).
+
+### Step 6: Capture Key Code Snippets (Optional)
+
+If any particularly important code was written or discussed, include a representative snippet:
+- Limit to 20-30 lines maximum
+- Include language identifier for syntax highlighting
+- Add brief explanation
+
+Only include if genuinely important. Skip if no critical code.
+
+### Step 7: Generate Snapshot Content
+
+Create the snapshot content in this structure (but DON'T write it yet):
+
+```markdown
+# Snapshot: {session_name}
+**Timestamp**: {full_timestamp_YYYY-MM-DD_HH:MM:SS}
+**Format Version**: 2.0
+
+## Topics Discussed
+1. **[Category]**: {topic_1_brief_description}
+2. **[Category]**: {topic_2_brief_description}
+{continue_for_all_topics}
+
+## Suggestions & Recommendations
+1. **[Category]**: {suggestion} - {rationale}
+{if_any_suggestions_given}
+
+## Decisions Made
+1. **[Decision]**: {rationale}
+{if_any_decisions_made}
+
+## Tasks Completed
+1. {completed_task_1}
+2. {completed_task_2}
+{or_write_"No_tasks_completed"}
+
+## Files Modified
+1. `{file_path}`: {what_changed_and_why}
+{or_write_"No_files_modified"}
+
+## Current Status
+- **Progress**: {where_things_stand}
+- **Next Steps**: {what_should_be_done_next}
+- **Blockers**: {issues_or_"None"}
+
+## Notes
+{any_additional_important_observations}
+```
+
+### Step 8: Write Snapshot via CLI (CRITICAL - Plan Mode Support)
+
+**Instead of using the Write tool**, use the CLI with stdin to write the snapshot.
+This enables snapshot saves even in plan mode where Write tool is blocked.
+
+Run this command, piping the snapshot content via stdin:
+
+```bash
+echo "{snapshot_content}" | node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js write-snapshot {session_name} --stdin --type manual
+```
+
+**IMPORTANT:** Escape any special characters in the content appropriately for the shell.
+
+Alternative approach if echo has escaping issues:
+```bash
+cat <<'EOF' | node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js write-snapshot {session_name} --stdin --type manual
+{snapshot_content}
+EOF
+```
+
+The CLI will:
+- Write the snapshot file with timestamp
+- Update the index automatically
+- Return success with filename
+
+### Step 9: Update context.md
+
+1. Read current `.claude/sessions/{active_session}/context.md`
+2. If any key decisions were made in this conversation, add them to "Key Decisions" section:
+   ```markdown
+   ### {current_timestamp}
+   - Decision: {what_was_decided}
+   - Rationale: {why_this_approach}
+   - Impact: {what_this_affects}
+   ```
+3. If any important discoveries were made, add to "Important Discoveries" section
+4. If any blockers were encountered or resolved, update "Blockers & Resolutions" section
+5. Update the "Summary" section with a running summary including this snapshot
+6. Write the updated context.md back (use Edit tool for this - it's allowed in plan mode for session files)
+
+### Step 10: Update session.md Timestamp
+
+1. Read `.claude/sessions/{active_session}/session.md`
+2. Update the "Last Updated" field to current timestamp
+3. If any new files were involved, add them to "Files Involved" section
+4. Write back the updated session.md (use Edit tool)
+
+### Step 11: Display Success Message
+
+Show confirmation:
+
+```
+✓ Session snapshot saved
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+📁 .claude/sessions/{session_name}/{timestamp}.md
+📝 Captured:
+   - Conversation summary
+   - {X} completed todos
+   - {Y} file changes
+   - Current state and next steps
+   {if_decisions: "- {Z} key decisions"}
+
+💾 Context updated in context.md
+⏰ Session timestamp updated
+```
+
+---
+
+**CRITICAL BENEFITS - PLAN MODE SUPPORT:**
+- **Problem:** In plan mode, Write tool is blocked (prevents accidental code changes)
+- **Impact:** Snapshots couldn't be saved, causing data loss on `/clear`
+- **Solution:** CLI delegation via Bash bypasses Write tool restrictions
+- **Result:** Zero data loss in plan mode, seamless user experience
+
+**PERFORMANCE BENEFITS:**
+- **Before:** 15-25K tokens for snapshot analysis and writes
+- **After:** 8-15K tokens, CLI handles write operations
+- **Improvement:** ~40-50% token reduction, instant index update
+
+**IMPORTANT NOTES:**
+- Use heredoc (cat <<'EOF') for reliable content piping, avoids escaping issues
+- CLI write-snapshot automatically updates index, no separate step needed
+- Edit tool can still be used for session.md and context.md (they're documentation, not code)
+- This approach works in BOTH normal mode and plan mode
diff --git a/commands/setup.md b/commands/setup.md
new file mode 100644
index 0000000..28b8a7d
--- /dev/null
+++ b/commands/setup.md
@@ -0,0 +1,369 @@
+You are helping the user configure session plugin hooks in their project's `.claude/settings.json` file.
+
+## Task: Setup Session Plugin Hooks
+
+This command manages the automatic configuration of session plugin hooks that enable auto-tracking of sessions, user interactions, and file modifications.
+
+**IMPORTANT:** This command uses the session CLI tool to perform hook management operations safely with atomic writes and backups.
+
+### Parse Command Arguments
+
+Extract the operation from the command arguments. Format: `/session:setup [options]`
+
+**Supported options:**
+- No arguments: Install/configure hooks (default)
+- `--permissions`: Install hooks + permission bypasses (eliminates ALL prompts)
+- `--remove`: Remove hooks from settings.json
+- `--status`: Show current hook and permission configuration
+- `--force-cleanup`: Force cleanup of orphaned hooks
+- `--dry-run`: Preview changes without applying them
+
+### Step 1: Determine Operation Mode
+
+1. Check arguments to determine which operation to perform:
+   - **Install mode** (default/no args): Add hooks to settings.json
+   - **Remove mode** (`--remove`): Remove hooks from settings.json
+   - **Status mode** (`--status`): Show current configuration
+   - **Cleanup mode** (`--force-cleanup`): Clean up orphaned hooks
+   - **Dry run** (`--dry-run`): Can be combined with other modes to preview
+
+### Step 2: Execute CLI Command
+
+Run the appropriate CLI command based on the operation mode:
+
+**For Install (hooks only):**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks
+```
+
+**For Install (hooks + permissions):**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks --permissions
+```
+
+**For Remove:**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks --remove
+```
+
+**For Status:**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks --status
+```
+
+**For Force Cleanup:**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks --force-cleanup
+```
+
+**For Dry Run (example with install):**
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js setup-hooks --dry-run
+```
+
+The CLI will return JSON with the operation result.
+
+### Step 3: Display Results
+
+Parse the JSON response and display user-friendly output based on the action:
+
+#### Install Success (action: "installed")
+
+**If permissions were added (permissionsRequested: true, permissionsCount > 0):**
+```
+✅ Session plugin configured successfully!
+
+📋 Hooks added to .claude/settings.json:
+  ✓ SessionStart - Auto-clears session on /clear
+  ✓ SessionEnd - Cleanup on Claude Code exit
+  ✓ UserPromptSubmit - Tracks user interactions
+  ✓ PostToolUse (Write, Edit, NotebookEdit) - Tracks file changes
+  ✓ Stop - Captures Claude's responses for self-contained logs
+
+🔓 Permissions added ({permissionsCount} entries):
+  ✓ Read(.claude/sessions/**)
+  ✓ Bash(git log --oneline:*)
+  ✓ Bash(git status --porcelain:*)
+  ✓ Bash(git diff --stat:*)
+  ✓ Bash(git branch -vv:*)
+  ✓ Bash(git rev-parse --abbrev-ref:*)
+
+⚡ Result: ZERO permission prompts during:
+  • /session:continue (no Read prompts, no git prompts)
+  • /session:save (no Read prompts)
+  • /session:status (no Read prompts)
+  • All session operations run silently
+
+💾 Backup saved: {backupPath}
+📁 Settings: {settingsPath}
+
+⚠️  IMPORTANT: Restart Claude Code for changes to take effect!
+   Hooks and permissions are loaded at startup.
+
+💡 To disable temporarily: Set "disableAllHooks": true in settings.json
+💡 To remove these hooks: /session:setup --remove
+```
+
+**If permissions were NOT added (permissionsRequested: false):**
+```
+✅ Session plugin hooks configured successfully!
+
+📋 Hooks added to .claude/settings.json:
+  ✓ SessionStart - Auto-clears session on /clear
+  ✓ SessionEnd - Cleanup on Claude Code exit
+  ✓ UserPromptSubmit - Tracks user interactions
+  ✓ PostToolUse (Write, Edit, NotebookEdit) - Tracks file changes
+  ✓ Stop - Captures Claude's responses for self-contained logs
+
+🎯 Sessions now automatically track:
+  • Session state changes
+  • User prompts and interactions
+  • Claude's full responses
+  • File modifications
+
+💡 TIP: Eliminate ALL permission prompts during session operations!
+   Run: /session:setup --permissions
+
+   This adds safe read-only permissions for:
+   • .claude/sessions/** files (Read access)
+   • git commands (log, status, diff, branch)
+
+   Result: Zero prompts during /session:continue ⚡
+
+💾 Backup saved: {backupPath}
+📁 Settings: {settingsPath}
+
+⚠️  IMPORTANT: Restart Claude Code for hooks to take effect!
+   Hooks are loaded at startup and won't activate until you restart.
+
+💡 To disable temporarily: Set "disableAllHooks": true in settings.json
+💡 To remove these hooks: /session:setup --remove
+```
+
+#### Already Configured (action: "already_configured")
+```
+ℹ️  Session plugin hooks are already configured!
+
+✓ All session plugin hooks are properly configured in settings.json
+
+Current hooks:
+  ✓ SessionStart - Active
+  ✓ UserPromptSubmit - Active
+  ✓ PostToolUse - Active
+  ✓ Stop - Active
+
+💡 To reinstall: /session:setup --remove, then /session:setup
+💡 To view details: /session:setup --status
+```
+
+#### Remove Success (action: "removed")
+```
+✅ Session plugin hooks removed!
+
+📋 Removed from .claude/settings.json:
+  ✓ SessionStart
+  ✓ UserPromptSubmit
+  ✓ PostToolUse
+  ✓ Stop
+
+💾 Backup saved: {backupPath}
+
+⚠️  IMPORTANT: Restart Claude Code to stop hooks!
+   Active hooks will continue running until you restart.
+
+💡 Sessions still work, but automatic tracking is disabled
+💡 To re-enable: /session:setup
+```
+
+#### Not Configured (action: "not_configured")
+```
+ℹ️  No session plugin hooks found
+
+Session plugin hooks are not currently configured in settings.json
+
+💡 To install hooks: /session:setup
+💡 This will enable automatic session tracking
+```
+
+#### Status Display (action: "status")
+
+**When properly configured:**
+```
+✅ Session Plugin Status
+
+📋 Hooks configured in .claude/settings.json:
+
+  ✓ SessionStart
+    → node ${CLAUDE_PLUGIN_ROOT}/hooks/session-start.js
+
+  ✓ SessionEnd
+    → node ${CLAUDE_PLUGIN_ROOT}/hooks/session-end.js
+
+  ✓ UserPromptSubmit
+    → node ${CLAUDE_PLUGIN_ROOT}/hooks/user-prompt-submit.js
+
+  ✓ PostToolUse (3 matchers)
+    → Write: node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js
+    → Edit: node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js
+    → NotebookEdit: node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js
+
+  ✓ Stop
+    → node ${CLAUDE_PLUGIN_ROOT}/hooks/stop.js
+
+📊 Hooks: {configuredHookTypes}/{totalHookTypes} hook types configured
+✅ All hooks pointing to valid scripts
+
+🔓 Permissions configured: {permissionsCount}/{totalPermissions}
+
+{if permissionsConfigured === true}
+  ✅ All session permissions configured
+  • Read(.claude/sessions/**)
+  • Bash(git log --oneline:*)
+  • Bash(git status --porcelain:*)
+  • Bash(git diff --stat:*)
+  • Bash(git branch -vv:*)
+  • Bash(git rev-parse --abbrev-ref:*)
+
+  ⚡ Result: Zero prompts during session operations
+{else if permissionsCount > 0 && permissionsCount < totalPermissions}
+  ⚠️  Partial permissions configured ({permissionsCount}/{totalPermissions})
+
+  💡 To add missing permissions: /session:setup --permissions
+{else}
+  ⚠️  No session permissions configured
+
+  💡 To eliminate permission prompts: /session:setup --permissions
+     This adds safe read-only permissions for session files and git commands
+{end if}
+
+🎯 Plugin Status: Installed & Configured
+
+💡 To remove: /session:setup --remove
+```
+
+**When orphaned (plugin uninstalled but hooks remain):**
+```
+⚠️  Session Plugin Hooks Status
+
+📋 Found in .claude/settings.json:
+
+  ⚠️  SessionStart
+    → node ${CLAUDE_PLUGIN_ROOT}/hooks/session-start.js
+    ❌ Script not found (plugin may be uninstalled)
+
+  ⚠️  UserPromptSubmit
+    ❌ Script not found
+
+  ⚠️  PostToolUse (3 matchers)
+    ❌ All scripts not found
+
+  ⚠️  Stop
+    ❌ Script not found
+
+🔍 Plugin Status: Not Found
+❌ Hooks are orphaned (pointing to missing files)
+
+Found {orphanedHooks.length} orphaned hook entries
+
+🧹 Cleanup Options:
+  1. Auto-cleanup: /session:setup --force-cleanup
+  2. Manual cleanup: Edit .claude/settings.json
+  3. Reinstall plugin to restore functionality
+```
+
+#### Cleanup Success (action: "cleaned")
+```
+✅ Orphaned hooks cleaned up!
+
+🧹 Removed {removedCount} orphaned hook entries:
+{List each orphaned hook that was removed}
+
+💾 Backup saved: {backupPath}
+
+⚠️  IMPORTANT: Restart Claude Code to apply cleanup!
+   Settings changes take effect on next restart.
+
+💡 Settings.json has been cleaned
+💡 To reinstall hooks: /session:setup
+```
+
+#### No Orphans (action: "no_orphans")
+```
+✅ No orphaned hooks found
+
+All hooks in settings.json are valid
+
+💡 Everything looks good!
+```
+
+#### Error Cases
+
+**On error (success: false):**
+```
+❌ Error: {error message}
+
+{Provide helpful context based on the error:}
+
+Common issues:
+- **Malformed JSON**: settings.json has syntax errors
+  → Fix: Restore from backup or manually repair JSON
+
+- **Permission denied**: No write access to .claude/settings.json
+  → Fix: Check file permissions
+
+- **Plugin not found**: Session plugin may not be installed
+  → Fix: Reinstall the session plugin
+
+💡 To restore from backup: cp .claude/settings.json.backup .claude/settings.json
+💡 Need help? Check the README or /session:setup --status
+```
+
+### Important Notes
+
+- **Atomic Operations**: All writes use atomic operations with backups
+- **Idempotent**: Safe to run multiple times, won't create duplicates
+- **Portable Paths**: Uses `${CLAUDE_PLUGIN_ROOT}` variable for cross-platform compatibility
+- **Merge Strategy**: Preserves existing hooks when installing
+- **Safety**: Creates `.backup` file before any modifications
+
+### Usage Examples
+
+```bash
+# Install hooks only (basic setup)
+/session:setup
+
+# Install hooks + permissions (RECOMMENDED - eliminates ALL prompts)
+/session:setup --permissions
+
+# Check current status (shows hooks and permissions)
+/session:setup --status
+
+# Preview what would be installed (dry run)
+/session:setup --dry-run
+
+# Preview with permissions
+/session:setup --permissions --dry-run
+
+# Remove hooks before uninstalling plugin
+/session:setup --remove
+
+# Clean up orphaned hooks after plugin uninstall
+/session:setup --force-cleanup
+```
+
+**Recommended:** Always use `--permissions` flag for best experience:
+- Zero permission prompts during `/session:continue`
+- Zero permission prompts during `/session:save`
+- Zero permission prompts during `/session:status`
+- Completely silent session operations
+
+---
+
+**CRITICAL for Plugin Uninstallation:**
+
+Before uninstalling the session plugin, users should run:
+```bash
+/session:setup --remove
+```
+
+This ensures clean removal and prevents orphaned hook entries in settings.json.
diff --git a/commands/start.md b/commands/start.md
new file mode 100644
index 0000000..ad40675
--- /dev/null
+++ b/commands/start.md
@@ -0,0 +1,293 @@
+You are managing a session memory system. The user wants to start a new named session.
+
+**CRITICAL - Command Format:**
+All session plugin commands use the `/session:` prefix. DO NOT suggest commands without this prefix.
+- ✅ Correct: `/session:list`, `/session:start`, `/session:continue`, `/session:status`
+- ❌ Wrong: `/session list`, `/session start`, `/session continue`, `/session status`
+Use ONLY the exact command formats specified in this template.
+
+## Task: Start New Session
+
+Parse the session name from the command arguments. The command format is: `/session:start [name]`
+
+**OPTIMIZATION:** This command now updates the metadata index after creating the session.
+
+### Step 1: Validate Session Name
+
+1. Extract the session name from the arguments
+2. Validate the name matches regex: `^[a-zA-Z0-9_-]+$`
+3. If invalid, show error:
+   ```
+   ❌ Error: Invalid session name '{name}'
+   Session names can only contain:
+   - Letters (a-z, A-Z)
+   - Numbers (0-9)
+   - Hyphens (-)
+   - Underscores (_)
+
+   Example: my-session or my_session_123
+   ```
+   Then STOP.
+
+### Step 2: Check for Duplicate Sessions
+
+**CRITICAL:** Check both directory names AND session.md file contents to prevent duplicates.
+
+1. **Check directory collision:**
+   - Check if `.claude/sessions/{name}/` directory already exists
+   - If exists, show error:
+     ```
+     ❌ Error: Session directory '{name}' already exists
+     💡 Use /session:continue {name} to resume
+     💡 Or use /session:list to see all sessions
+     ```
+     Then STOP.
+
+2. **Check session name collision (NEW - prevents duplicate names):**
+   - Use the CLI to get all existing sessions:
+     ```bash
+     node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js list
+     ```
+   - Parse the JSON response and check if any existing session has `name: "{name}"`
+   - If a session with this name exists (even if directory name is different), show error:
+     ```
+     ❌ Error: A session named '{name}' already exists
+     💡 Use /session:list to see all sessions
+     💡 Choose a different name for this session
+     ```
+     Then STOP.
+
+This dual-check prevents both directory collisions and duplicate session names.
+
+### Step 3: Check for Active Session and Transition (NEW - Session Cleanup)
+
+Before creating the new session, check if there's already an active session:
+
+1. Check if `.claude/sessions/.active-session` exists
+2. If it exists:
+   - Read the current active session name
+   - Show: "📋 Closing previous session '{previous_session_name}'..."
+   - Close the previous session status:
+     ```bash
+     node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-status "{previous_session_name}" "closed"
+     ```
+   - This marks the previous session as closed in `.auto-capture-state`
+3. Continue to next step (the new session activation will overwrite `.active-session`)
+
+**Note:** The SessionEnd hook will handle final cleanup on Claude Code termination.
+
+### Step 4: Create Session Folder
+
+1. Create directory: `.claude/sessions/{name}/`
+2. Verify creation was successful
+
+### Step 5: Initialize session.md
+
+**CRITICAL:** The session name in session.md MUST match the directory name to prevent duplicates.
+
+Create `.claude/sessions/{name}/session.md` with this content:
+
+```markdown
+# Session: {name}
+
+**Started**: {current_timestamp_YYYY-MM-DD_HH:MM}
+**Last Updated**: {current_timestamp_YYYY-MM-DD_HH:MM}
+
+## Goal
+[Ask user to describe the goal for this session]
+
+## Overview
+Working on: {name}
+
+## Key Milestones
+- [ ] [To be defined during session]
+
+## Files Involved
+- [Will be tracked as work progresses]
+
+## Configuration
+- Auto-capture: enabled
+
+## Notes
+Session started. Use /session:save to capture important milestones.
+```
+
+### Step 6: Initialize context.md
+
+Create `.claude/sessions/{name}/context.md` with this content:
+
+```markdown
+# Session Context: {name}
+
+## Key Decisions
+[Decisions will be captured during session]
+
+## Important Discoveries
+[Discoveries will be documented as they occur]
+
+## Blockers & Resolutions
+[Issues and resolutions will be tracked here]
+
+## Technical Context
+[Architecture decisions and technical details]
+
+## Summary
+Session started on {current_timestamp}. Ready to capture context.
+```
+
+### Step 7: Update Active Session Tracker
+
+1. Write the session name to `.claude/sessions/.active-session` (single line, just the name)
+2. This marks it as the currently active session
+
+### Step 7.5: Initialize .auto-capture-state with Session Status
+
+Create `.claude/sessions/{name}/.auto-capture-state` with initial session status:
+
+```bash
+echo '{
+  "interaction_count": 0,
+  "modified_files": [],
+  "last_snapshot_timestamp": null,
+  "session_status": "active",
+  "session_started": "'$(date -u +%Y-%m-%dT%H:%M:%SZ)'",
+  "session_closed": null
+}' > .claude/sessions/{name}/.auto-capture-state
+```
+
+This initializes the operational state tracking for the session, including the session status which is now tracked in JSON for fast, atomic updates.
+
+### Step 8: Update Index (NEW - CRITICAL)
+
+Run the CLI command to add the new session to the metadata index:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js update-index --session {name}
+```
+
+This ensures the new session appears immediately in `/session:list` without requiring a full rebuild.
+
+### Step 9: Display Success Message
+
+Show this confirmation to the user:
+
+```
+✓ Session '{name}' started
+📁 Location: .claude/sessions/{name}/
+💡 Use /session:save to capture important milestones
+📊 Check /session:status to monitor tokens
+
+Ready to work! What would you like to accomplish in this session?
+```
+
+### Step 10: Ask for Session Goal
+
+After showing the success message, ask the user:
+"What is the main goal or purpose of this session? (This will be saved to session.md)"
+
+Wait for their response, then update the "Goal" section in session.md with their answer.
+
+---
+
+**PERFORMANCE BENEFITS:**
+- **Before:** No index integration, sessions invisible until manual list command scans
+- **After:** Instant index update, session immediately visible in list
+- **Improvement:** Better UX, no stale index issues
+
+**IMPORTANT:**
+- Execute all steps in order
+- Use the Read, Write, and Bash tools to perform file operations
+- The CLI update-index command is fast (< 10ms) and ensures consistency
+- Do not skip any steps
+
+---
+
+## CRITICAL: Check for Unconsolidated Logs (Inline Analysis)
+
+**IMPORTANT:** New session start should NOT have unconsolidated logs, but check anyway for robustness.
+
+**MUST CHECK BEFORE DISPLAYING SUCCESS MESSAGE:**
+
+After starting the session, check if previous session left unconsolidated logs:
+
+1. Check if `.claude/sessions/{name}/conversation-log.jsonl` exists
+2. If the file exists:
+   - Show brief message: "📊 Analyzing previous session... (this may take 1-3 seconds)"
+   - Read the conversation log file
+   - Parse interactions from JSONL format
+   - **Capture git history (if available):**
+     - Run: `node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js capture-git "{name}"`
+     - This creates `.claude/sessions/{name}/git-history.json` (~2-3KB compressed)
+     - Contains: last 50 commits, uncommitted changes, branch status, hotspots
+     - Performance: ~60-90ms (acceptable at session boundary)
+     - If no git repo, command returns success: false (silent skip, no error)
+   - Analyze the conversation with Claude inline:
+     - Extract conversation summary (2-3 paragraphs covering what happened)
+     - Identify key decisions made with rationale
+     - List completed todos/tasks
+     - Document files modified with context about what changed and why
+     - Assess current state, what's next, and any blockers
+   - Create consolidated snapshot via CLI:
+     ```bash
+     echo "# Consolidated Snapshot: {session_name}
+**Timestamp**: {ISO timestamp}
+**Method**: Claude Inline Analysis (Free)
+**Status**: Consolidated from conversation log
+**Format Version**: 2.0
+
+## Topics Discussed
+
+1. **[Category]**: [Brief description of what was discussed]
+2. **[Category]**: [Brief description of what was discussed]
+[Continue for ALL topics in chronological order]
+
+## Suggestions & Recommendations
+
+1. **[Category]**: [Specific suggestion] - [Rationale]
+2. **[Category]**: [Specific suggestion] - [Rationale]
+[Continue for ALL suggestions]
+
+## Decisions Made
+
+1. **[Decision]**: [Rationale and context]
+2. **[Decision]**: [Rationale and context]
+[Continue for ALL decisions]
+
+## Tasks Completed
+
+1. [Action completed in past tense]
+2. [Action completed in past tense]
+[Continue for ALL tasks]
+
+## Files Modified
+
+1. `[file_path]`: [What changed and why]
+2. `[file_path]`: [What changed and why]
+[Continue for ALL files]
+
+## Current Status
+
+- **Progress**: [Where things stand - what's been accomplished]
+- **Next Steps**: [What should be done next]
+- **Blockers**: [Issues or write \"None\"]
+
+## Notes
+Consolidated via Claude inline analysis at session boundary. Zero cost, highest quality." | node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js write-snapshot "{name}" --stdin --type auto
+     ```
+   - Delete conversation-log.jsonl after successful snapshot creation
+   - Update `.auto-capture-state` to reset counters and set last_snapshot_timestamp
+3. If no log exists:
+   - **Still capture git history** for repository context:
+     - Run: `node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js capture-git "{name}"`
+     - This provides git context even for new sessions
+     - Silent skip if no git repo (no error)
+   - Initialize `.auto-capture-state` if needed
+
+**PERFORMANCE:**
+- Log check: <5ms
+- Claude analysis: 1-3s (acceptable at session boundaries - users expect loading)
+- Snapshot write: <50ms
+- Log deletion: <5ms
+- **Total: ~1-3 seconds** (users expect loading at session start)
+
+**NOTE:** This is the v3.5.1 architecture where conversation logging is incremental (<2ms per interaction) and consolidation happens inline at session boundaries using FREE Claude analysis for highest quality.
diff --git a/commands/status.md b/commands/status.md
new file mode 100644
index 0000000..7676e3d
--- /dev/null
+++ b/commands/status.md
@@ -0,0 +1,153 @@
+You are managing a session memory system. The user wants to check the current session status.
+
+## Task: Display Session Status (Ultra-Minimal)
+
+Show essential session information quickly and efficiently.
+
+**OPTIMIZATION:** Ultra-minimal output. Uses CLI for instant stats (<50ms, <50 tokens).
+
+### Step 1: Get Active Session from CLI
+
+Run the CLI command to get session list and find active session:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js list
+```
+
+Parse the JSON response to get `activeSession` field.
+
+If `activeSession` is null, show error:
+```
+❌ No active session
+
+💡 /session:start [name] to create new session
+💡 /session:continue [name] to resume existing session
+```
+Then STOP.
+
+### Step 2: Get Session Statistics
+
+Run the CLI stats command for the active session:
+
+```bash
+node ${CLAUDE_PLUGIN_ROOT}/cli/session-cli.js stats {activeSession}
+```
+
+This returns JSON with:
+- Session metadata (status, started, lastUpdated, goal)
+- Snapshot counts (snapshotCount)
+- File counts (filesInvolvedCount)
+
+### Step 3: Calculate Session Duration
+
+Parse the `started` timestamp from stats (format: "2025-11-13 14:30").
+
+Calculate duration from start time to now:
+1. Parse started date/time
+2. Calculate difference in minutes
+3. Format as:
+   - If < 60 min: "Xm" (e.g., "45m")
+   - If < 24 hours: "Xh Ym" (e.g., "2h 15m" or "5h 0m")
+   - If >= 24 hours: "Xd Yh" (e.g., "2d 3h")
+
+### Step 4: Calculate Time Since Last Activity
+
+Parse `lastUpdated` timestamp (ISO format: "2025-11-14T10:27:04.345Z").
+
+Calculate time since last update:
+1. Parse ISO timestamp
+2. Calculate difference from now
+3. Format as relative time:
+   - "just now" (< 1 min)
+   - "Xm ago" (< 60 min)
+   - "Xh ago" (< 24 hours)
+   - "Xd ago" (>= 24 hours)
+
+### Step 5: Display Ultra-Minimal Status
+
+Show clean, essential information only:
+
+```
+✓ Session: {sessionName} ({status})
+  Working for: {duration} (started {started_short})
+  Snapshots: {snapshotCount} total
+  Files: {filesInvolvedCount} tracked
+  Last activity: {time_since_last_update}
+
+💡 /session:save to capture milestones
+💡 /session:close to finalize
+```
+
+**Format notes:**
+- {status}: "active" or "closed"
+- {started_short}: "Nov 14, 14:30" format
+- All numbers: plain integers (no formatting)
+- No emojis except status indicator (✓)
+- No Unicode art, no progress bars
+- 5 lines max
+
+### Example Output
+
+**Active session:**
+```
+✓ Session: feature-auth (active)
+  Working for: 2h 15m (started Nov 14, 14:30)
+  Snapshots: 12 total
+  Files: 7 tracked
+  Last activity: 30m ago
+
+💡 /session:save to capture milestones
+💡 /session:close to finalize
+```
+
+**Recently started:**
+```
+✓ Session: bug-fix-login (active)
+  Working for: 15m (started Nov 14, 18:45)
+  Snapshots: 1 total
+  Files: 2 tracked
+  Last activity: just now
+
+💡 /session:save to capture milestones
+💡 /session:close to finalize
+```
+
+**Closed session:**
+```
+✓ Session: refactor-api (closed)
+  Worked for: 5h 30m (started Nov 13, 09:00)
+  Snapshots: 18 total
+  Files: 12 tracked
+  Last activity: 2d ago
+
+💡 /session:continue {sessionName} to resume
+```
+
+---
+
+**PERFORMANCE:**
+- **Tokens:** ~50 tokens (vs 150 before) - **66% reduction**
+- **Speed:** <50ms (CLI delegation)
+- **Lines:** 5-7 lines (vs 15-20 before) - **70% reduction**
+
+**WHAT WAS REMOVED:**
+- ❌ Token usage tracking (unreliable, verbose)
+- ❌ Progress bars (unnecessary decoration)
+- ❌ Warning messages (not reliable)
+- ❌ Unicode art (bloat)
+- ❌ Total size display (rarely useful)
+- ❌ Auto vs manual snapshot breakdown (not essential)
+
+**WHAT WAS KEPT:**
+- ✅ Session name and status
+- ✅ Duration (how long working)
+- ✅ Snapshot count (progress indicator)
+- ✅ File count (scope indicator)
+- ✅ Last activity (freshness indicator)
+- ✅ Quick action tips
+
+**FOR TOKEN USAGE:** Users should run `/context` directly for accurate real-time token information.
+
+**ERROR HANDLING:**
+- If CLI command fails, show error and suggest rebuilding index
+- If timestamps can't be parsed, show raw values
diff --git a/hooks/hooks.json b/hooks/hooks.json
new file mode 100644
index 0000000..7c9e42b
--- /dev/null
+++ b/hooks/hooks.json
@@ -0,0 +1,71 @@
+{
+  "SessionStart": [
+    {
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/session-start.js"
+        }
+      ]
+    }
+  ],
+  "UserPromptSubmit": [
+    {
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/user-prompt-submit.js"
+        }
+      ]
+    }
+  ],
+  "PostToolUse": [
+    {
+      "matcher": "Write",
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js"
+        }
+      ]
+    },
+    {
+      "matcher": "Edit",
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js"
+        }
+      ]
+    },
+    {
+      "matcher": "NotebookEdit",
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/post-tool-use.js"
+        }
+      ]
+    }
+  ],
+  "Stop": [
+    {
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/stop.js"
+        }
+      ]
+    }
+  ],
+  "SessionEnd": [
+    {
+      "hooks": [
+        {
+          "type": "command",
+          "command": "node ${CLAUDE_PLUGIN_ROOT}/hooks/session-end.js"
+        }
+      ]
+    }
+  ]
+}
diff --git a/hooks/post-tool-use.js b/hooks/post-tool-use.js
new file mode 100755
index 0000000..3bc900f
--- /dev/null
+++ b/hooks/post-tool-use.js
@@ -0,0 +1,188 @@
+#!/usr/bin/env node
+// Intelligent Post-Tool-Use Hook - Tracks file modifications for smart analysis
+// Works with async analysis system for intelligent snapshots
+//
+// SAFETY: Includes graceful failure handling to avoid blocking Claude Code
+// if plugin is uninstalled or dependencies are missing.
+
+const fs = require('fs');
+const path = require('path');
+
+// Graceful failure wrapper - protect against plugin uninstallation
+try {
+  // Check if critical dependencies exist (indicates plugin is installed)
+  const cliLibPath = path.join(__dirname, '../cli/lib');
+  if (!fs.existsSync(cliLibPath)) {
+    // Plugin likely uninstalled, exit silently
+    process.exit(0);
+  }
+
+  const LockManager = require('../cli/lib/lock-manager');
+
+  // Read tool operation data from stdin (Claude Code provides this as JSON)
+  // NOTE: Do NOT use process.env.CLAUDE_TOOL_NAME - it's always "unknown" due to Claude Code bug
+  let toolName = 'unknown';
+  let filePath = null;
+
+  try {
+  const input = fs.readFileSync(0, 'utf8').trim();
+  if (input) {
+    const eventData = JSON.parse(input);
+
+    // Get tool name from stdin data (NOT environment variable)
+    toolName = eventData.tool_name || 'unknown';
+
+    // Extract file path from tool input
+    // Different tools use different field names for file paths
+    if (eventData.tool_input) {
+      if (eventData.tool_input.file_path) {
+        filePath = eventData.tool_input.file_path;
+      } else if (eventData.tool_input.notebook_path) {
+        // NotebookEdit uses notebook_path instead of file_path
+        filePath = eventData.tool_input.notebook_path;
+      }
+    }
+  }
+} catch (err) {
+  // If we can't read stdin or parse it, exit early
+  process.exit(0);
+}
+
+// Only track file modifications (Write, Edit, and NotebookEdit tools)
+if (toolName !== 'Write' && toolName !== 'Edit' && toolName !== 'NotebookEdit') {
+  process.exit(0);
+}
+
+// Configuration
+const SESSIONS_DIR = '.claude/sessions';
+const ACTIVE_SESSION_FILE = path.join(SESSIONS_DIR, '.active-session');
+const lockManager = new LockManager(SESSIONS_DIR);
+
+// Exit early if no active session
+if (!fs.existsSync(ACTIVE_SESSION_FILE)) {
+  process.exit(0);
+}
+
+// Read active session name
+let activeSession;
+try {
+  activeSession = fs.readFileSync(ACTIVE_SESSION_FILE, 'utf8').trim();
+} catch (err) {
+  process.exit(0);
+}
+
+if (!activeSession) {
+  process.exit(0);
+}
+
+const sessionDir = path.join(SESSIONS_DIR, activeSession);
+if (!fs.existsSync(sessionDir)) {
+  process.exit(0);
+}
+
+// Check if auto-capture is enabled
+const sessionMd = path.join(sessionDir, 'session.md');
+if (fs.existsSync(sessionMd)) {
+  try {
+    const content = fs.readFileSync(sessionMd, 'utf8');
+    if (content.includes('Auto-capture: disabled')) {
+      process.exit(0);
+    }
+  } catch (err) {
+    // Continue if we can't read the file
+  }
+}
+
+// State file
+const stateFile = path.join(sessionDir, '.auto-capture-state');
+
+// Use lock to prevent race conditions with user-prompt-submit.js hook
+const lock = lockManager.acquireLock(`auto-capture-${activeSession}`, {
+  timeout: 1000,
+  wait: true
+});
+
+if (!lock.acquired) {
+  // Could not acquire lock - skip this update to avoid blocking
+  // The file will be captured on next tool use
+  process.exit(0);
+}
+
+try {
+  // Initialize state if doesn't exist
+  let state = {
+    file_count: 0,
+    interaction_count: 0,
+    interactions_since_last_analysis: 0,
+    last_snapshot: '',
+    last_reason: '',
+    last_analysis_timestamp: '',
+    modified_files: []  // Track actual file paths
+  };
+
+  if (fs.existsSync(stateFile)) {
+    try {
+      state = JSON.parse(fs.readFileSync(stateFile, 'utf8'));
+    } catch (err) {
+      // Use default state if parse fails
+    }
+  }
+
+  // Increment file count
+  state.file_count++;
+
+  // Track the modified file path if we captured it
+  if (filePath) {
+    // Initialize modified_files array if it doesn't exist (backward compatibility)
+    if (!state.modified_files) {
+      state.modified_files = [];
+    }
+
+    // Add file to list if not already tracked
+    const fileEntry = {
+      path: filePath,
+      operation: toolName.toLowerCase(),
+      timestamp: new Date().toISOString()
+    };
+
+    // Avoid duplicates - check if file already in list
+    const existingIndex = state.modified_files.findIndex(f => f.path === filePath);
+    if (existingIndex >= 0) {
+      // Update existing entry with latest operation
+      state.modified_files[existingIndex] = fileEntry;
+    } else {
+      // Add new file
+      state.modified_files.push(fileEntry);
+    }
+  }
+
+  // Update state atomically
+  // Note: We no longer immediately create snapshot markers here
+  // Instead, the file_count contributes to analysis queue logic in user-prompt-submit.js
+  const tempPath = `${stateFile}.tmp.${Date.now()}`;
+  try {
+    fs.writeFileSync(tempPath, JSON.stringify(state, null, 2));
+    fs.renameSync(tempPath, stateFile);
+  } catch (writeError) {
+    // Clean up temp file
+    if (fs.existsSync(tempPath)) {
+      try {
+        fs.unlinkSync(tempPath);
+      } catch (cleanupError) {
+        // Ignore
+      }
+    }
+    throw writeError;
+  }
+  } finally {
+    // Always release lock
+    lock.release();
+  }
+
+  process.exit(0);
+
+} catch (error) {
+  // Outer catch: Handle plugin missing/uninstalled
+  // Exit silently to avoid blocking Claude Code
+  process.exit(0);
+}
diff --git a/hooks/session-end.js b/hooks/session-end.js
new file mode 100755
index 0000000..05ee15a
--- /dev/null
+++ b/hooks/session-end.js
@@ -0,0 +1,111 @@
+#!/usr/bin/env node
+
+/**
+ * SessionEnd Hook
+ *
+ * Handles cleanup when Claude Code session terminates.
+ * Ensures .active-session file is removed and index is updated.
+ *
+ * Hook receives:
+ * - session_id: string
+ * - transcript_path: string
+ * - cwd: string
+ * - permission_mode: string
+ * - hook_event_name: "SessionEnd"
+ * - reason: "exit" | "clear" | "logout" | "prompt_input_exit" | "other"
+ *
+ * SAFETY: Includes graceful failure handling to avoid blocking Claude Code shutdown
+ * if plugin is uninstalled or dependencies are missing.
+ */
+
+const fs = require('fs');
+const path = require('path');
+
+// Graceful failure wrapper - protect against plugin uninstallation
+try {
+  // Check if critical dependencies exist (indicates plugin is installed)
+  const cliLibPath = path.join(__dirname, '../cli/lib');
+  if (!fs.existsSync(cliLibPath)) {
+    // Plugin likely uninstalled, exit silently
+    process.exit(0);
+  }
+
+  const IndexManager = require('../cli/lib/index-manager');
+
+  // Constants
+  const SESSIONS_DIR = '.claude/sessions';
+  const ACTIVE_SESSION_FILE = path.join(SESSIONS_DIR, '.active-session');
+  const indexManager = new IndexManager(SESSIONS_DIR);
+
+  try {
+    // Read input from stdin (provided by Claude Code)
+    const input = fs.readFileSync(0, 'utf8').trim();
+
+    // Parse the JSON input
+    let eventData;
+    try {
+      eventData = JSON.parse(input);
+    } catch (parseError) {
+      // If parsing fails, exit silently (no input provided)
+      process.exit(0);
+    }
+
+    // Clean up active session marker
+    let sessionName = null;
+
+    if (fs.existsSync(ACTIVE_SESSION_FILE)) {
+      try {
+        sessionName = fs.readFileSync(ACTIVE_SESSION_FILE, 'utf8').trim();
+      } catch (readError) {
+        // Continue even if read fails
+      }
+
+      // Delete the .active-session file
+      try {
+        fs.unlinkSync(ACTIVE_SESSION_FILE);
+      } catch (unlinkError) {
+        // File may already be deleted, continue
+      }
+    }
+
+    // Mark session as closed in .auto-capture-state (fast, atomic)
+    if (sessionName) {
+      const sessionDir = path.join(SESSIONS_DIR, sessionName);
+      const statePath = path.join(sessionDir, '.auto-capture-state');
+
+      if (fs.existsSync(statePath)) {
+        try {
+          let state = JSON.parse(fs.readFileSync(statePath, 'utf8'));
+          state.session_status = 'closed';
+          state.session_closed = new Date().toISOString();
+          fs.writeFileSync(statePath, JSON.stringify(state, null, 2));
+        } catch (stateError) {
+          // Silent fail - hook must be safe
+          // Session status update failed but continue with other cleanup
+        }
+      }
+    }
+
+    // Update the index.json to clear activeSession using IndexManager
+    // This uses proper locking and atomic writes to prevent corruption
+    try {
+      const index = indexManager.read({ skipValidation: true });
+      index.activeSession = null;
+      indexManager.write(index);
+    } catch (indexError) {
+      // Continue even if index update fails
+      // This prevents blocking the hook if index is temporarily locked
+    }
+
+    // Exit successfully
+    process.exit(0);
+
+  } catch (error) {
+    // Exit silently on any errors to avoid blocking Claude Code shutdown
+    process.exit(0);
+  }
+} catch (error) {
+  // Outer catch: Handle plugin missing/uninstalled
+  // Exit silently to avoid blocking Claude Code
+  process.exit(0);
+}
diff --git a/hooks/session-start.js b/hooks/session-start.js
new file mode 100755
index 0000000..0bc9e91
--- /dev/null
+++ b/hooks/session-start.js
@@ -0,0 +1,110 @@
+#!/usr/bin/env node
+
+/**
+ * SessionStart Hook
+ *
+ * Handles session state cleanup when /clear is executed.
+ * Clears active session markers to prevent confusion when context is lost.
+ *
+ * Hook receives:
+ * - source: "startup" | "resume" | "clear" | "compact"
+ * - session_id: string
+ * - cwd: string
+ * - permission_mode: string
+ *
+ * SAFETY: Includes graceful failure handling to avoid blocking Claude Code
+ * if plugin is uninstalled or dependencies are missing.
+ */
+
+const fs = require('fs');
+const path = require('path');
+
+// Graceful failure wrapper - protect against plugin uninstallation
+try {
+  // Check if critical dependencies exist (indicates plugin is installed)
+  const cliLibPath = path.join(__dirname, '../cli/lib');
+  if (!fs.existsSync(cliLibPath)) {
+    // Plugin likely uninstalled, exit silently
+    process.exit(0);
+  }
+
+  // Import IndexManager for safe index updates
+  const IndexManager = require('../cli/lib/index-manager');
+
+  // Constants
+  const SESSIONS_DIR = '.claude/sessions';
+  const ACTIVE_SESSION_FILE = path.join(SESSIONS_DIR, '.active-session');
+  const indexManager = new IndexManager(SESSIONS_DIR);
+
+  try {
+  // Read input from stdin (provided by Claude Code)
+  const input = fs.readFileSync(0, 'utf8').trim();
+
+  // Parse the JSON input
+  let eventData;
+  try {
+    eventData = JSON.parse(input);
+  } catch (parseError) {
+    // If parsing fails, exit silently (no input provided)
+    process.exit(0);
+  }
+
+  const { source } = eventData;
+
+  // Only process when source is "clear"
+  // Other sources ("startup", "resume", "compact") should allow normal auto-resume
+  if (source === 'clear') {
+    let sessionName = null;
+
+    // Read the active session name before clearing (for the message)
+    if (fs.existsSync(ACTIVE_SESSION_FILE)) {
+      try {
+        sessionName = fs.readFileSync(ACTIVE_SESSION_FILE, 'utf8').trim();
+      } catch (readError) {
+        // Continue even if read fails
+      }
+
+      // Clear the .active-session file
+      try {
+        fs.unlinkSync(ACTIVE_SESSION_FILE);
+      } catch (unlinkError) {
+        // File may already be deleted, continue
+      }
+    }
+
+    // Update the index.json to clear activeSession using IndexManager
+    // This uses proper locking and atomic writes to prevent corruption
+    try {
+      const index = indexManager.read({ skipValidation: true });
+      index.activeSession = null;
+      indexManager.write(index);
+    } catch (indexError) {
+      // Continue even if index update fails
+      // This prevents blocking the hook if index is temporarily locked
+    }
+
+    // Inject helpful context message to Claude
+    const output = {
+      hookSpecificOutput: {
+        additionalContext: sessionName
+          ? `📋 Session '${sessionName}' was auto-closed due to /clear command. The conversation context has been cleared.\n\nTo resume your work on this session, use: /session:continue ${sessionName}\nTo view all sessions, use: /session:list`
+          : 'Previous session was auto-closed due to /clear command. Use /session:list to view available sessions.'
+      }
+    };
+
+    // Output the context injection
+    console.log(JSON.stringify(output));
+  }
+
+  // Exit successfully
+  process.exit(0);
+
+  } catch (error) {
+    // Exit silently on any errors to avoid blocking Claude Code startup
+    process.exit(0);
+  }
+} catch (error) {
+  // Outer catch: Handle plugin missing/uninstalled
+  // Exit silently to avoid blocking Claude Code
+  process.exit(0);
+}
diff --git a/hooks/stop.js b/hooks/stop.js
new file mode 100755
index 0000000..fe0f2c7
--- /dev/null
+++ b/hooks/stop.js
@@ -0,0 +1,210 @@
+#!/usr/bin/env node
+// Stop Hook - Captures Claude's complete response for self-contained conversation logs
+// This hook fires after Claude completes each response
+//
+// SAFETY: Includes graceful failure handling to avoid blocking Claude Code
+// if plugin is uninstalled or dependencies are missing.
+
+const fs = require('fs');
+const path = require('path');
+
+// Graceful failure wrapper - protect against plugin uninstallation
+try {
+  // Check if critical dependencies exist (indicates plugin is installed)
+  const cliLibPath = path.join(__dirname, '../cli/lib');
+  if (!fs.existsSync(cliLibPath)) {
+    // Plugin likely uninstalled, exit silently
+    process.exit(0);
+  }
+
+  const LockManager = require('../cli/lib/lock-manager');
+
+// Read stdin to get transcript path
+let stdinData;
+try {
+  const stdinInput = fs.readFileSync(0, 'utf8').trim();
+
+  if (!stdinInput) {
+    process.exit(0);
+  }
+  stdinData = JSON.parse(stdinInput);
+} catch (stdinErr) {
+  // Cannot parse stdin, exit silently
+  process.exit(0);
+}
+
+const transcriptPath = stdinData.transcript_path;
+
+if (!transcriptPath || !fs.existsSync(transcriptPath)) {
+  // No transcript path or file doesn't exist
+  process.exit(0);
+}
+
+// Read transcript file to get Claude's last response (with exponential backoff retry)
+// Uses smart retry strategy: fast success path (0-50ms), patient for edge cases (750ms max)
+const MAX_RETRIES = 5;
+const RETRY_DELAYS = [0, 50, 100, 200, 400]; // Exponential backoff in milliseconds
+
+function tryFindAssistantMessage() {
+  try {
+    const transcriptContent = fs.readFileSync(transcriptPath, 'utf8');
+    const lines = transcriptContent.trim().split('\n');
+
+    // Find last assistant message (search backwards for efficiency)
+    for (let i = lines.length - 1; i >= 0; i--) {
+      try {
+        const entry = JSON.parse(lines[i]);
+        // Claude Code transcript format: {type: 'assistant', message: {role, content}}
+        if (entry.type === 'assistant' && entry.message) {
+          return entry.message; // Return the message object which has role and content
+        }
+      } catch (parseErr) {
+        // Skip malformed lines
+        continue;
+      }
+    }
+    return null;
+  } catch (readErr) {
+    return null;
+  }
+}
+
+// Sleep function for retry delays
+function sleep(ms) {
+  const start = Date.now();
+  while (Date.now() - start < ms) {
+    // Busy wait (acceptable for short delays in hooks)
+  }
+}
+
+// Try to find assistant message with exponential backoff retries
+let lastAssistantMessage = null;
+
+for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
+  lastAssistantMessage = tryFindAssistantMessage();
+
+  if (lastAssistantMessage) {
+    break;
+  }
+
+  // If not found and not last attempt, wait with exponential backoff
+  if (attempt < MAX_RETRIES) {
+    const nextDelay = RETRY_DELAYS[attempt]; // Next delay for next attempt
+    sleep(nextDelay);
+  }
+}
+
+if (!lastAssistantMessage) {
+  // No assistant message found after all retries
+  process.exit(0);
+}
+
+// Configuration
+const SESSIONS_DIR = '.claude/sessions';
+const ACTIVE_SESSION_FILE = path.join(SESSIONS_DIR, '.active-session');
+
+// Exit early if no active session
+if (!fs.existsSync(ACTIVE_SESSION_FILE)) {
+  process.exit(0);
+}
+
+// Read active session name
+let activeSession;
+try {
+  activeSession = fs.readFileSync(ACTIVE_SESSION_FILE, 'utf8').trim();
+} catch (err) {
+  process.exit(0);
+}
+
+if (!activeSession) {
+  process.exit(0);
+}
+
+const sessionDir = path.join(SESSIONS_DIR, activeSession);
+
+if (!fs.existsSync(sessionDir)) {
+  process.exit(0);
+}
+
+// Check if auto-capture is enabled
+const sessionMd = path.join(sessionDir, 'session.md');
+
+if (fs.existsSync(sessionMd)) {
+  try {
+    const content = fs.readFileSync(sessionMd, 'utf8');
+    if (content.includes('Auto-capture: disabled')) {
+      process.exit(0);
+    }
+  } catch (err) {
+    // Continue if we can't read the file
+  }
+}
+
+// Use lock to prevent race conditions
+const lockManager = new LockManager(SESSIONS_DIR);
+const lock = lockManager.acquireLock(`auto-capture-${activeSession}`, {
+  timeout: 1000,
+  wait: true
+});
+
+if (!lock.acquired) {
+  // Could not acquire lock - skip this update to avoid blocking
+  process.exit(0);
+}
+
+try {
+  // Extract tool uses from response
+  const toolUses = [];
+  if (lastAssistantMessage.content && Array.isArray(lastAssistantMessage.content)) {
+    lastAssistantMessage.content.forEach(block => {
+      if (block.type === 'tool_use') {
+        toolUses.push({
+          tool: block.name,
+          input: block.input,
+          id: block.id || null
+        });
+      }
+    });
+  }
+
+  // Helper to extract text from content blocks
+  function extractTextContent(content) {
+    if (typeof content === 'string') {
+      return content;
+    }
+    if (Array.isArray(content)) {
+      return content
+        .filter(block => block.type === 'text')
+        .map(block => block.text)
+        .join('\n');
+    }
+    return '';
+  }
+
+  const responseText = extractTextContent(lastAssistantMessage.content);
+
+  // Log Claude's response
+  const ConversationLogger = require('../cli/lib/conversation-logger');
+  const logger = new ConversationLogger(sessionDir);
+
+  logger.logAssistantResponse({
+    timestamp: new Date().toISOString(),
+    response_text: responseText,
+    tools_used: toolUses,
+    message_id: lastAssistantMessage.id || null
+  });
+
+} catch (error) {
+  // Silent failure - don't block hook execution
+} finally {
+  // Always release lock
+  lock.release();
+}
+
+process.exit(0);
+
+} catch (error) {
+  // Outer catch: Handle plugin missing/uninstalled
+  // Exit silently to avoid blocking Claude Code
+  process.exit(0);
+}
diff --git a/hooks/user-prompt-submit.js b/hooks/user-prompt-submit.js
new file mode 100755
index 0000000..5259d0f
--- /dev/null
+++ b/hooks/user-prompt-submit.js
@@ -0,0 +1,226 @@
+#!/usr/bin/env node
+// Truly Automatic Auto-Capture Hook - Creates snapshots automatically
+// This hook creates snapshot files directly via CLI with NO manual intervention
+// Snapshots are created every N interactions or when file thresholds are met
+//
+// SAFETY: Includes graceful failure handling to avoid blocking Claude Code
+// if plugin is uninstalled or dependencies are missing.
+//
+// ORPHAN DETECTION: Periodically checks for orphaned hooks and auto-cleans them
+
+const fs = require('fs');
+const path = require('path');
+
+// Graceful failure wrapper - protect against plugin uninstallation
+try {
+  // Check if critical dependencies exist (indicates plugin is installed)
+  const cliLibPath = path.join(__dirname, '../cli/lib');
+  if (!fs.existsSync(cliLibPath)) {
+    // Plugin likely uninstalled, exit silently
+    process.exit(0);
+  }
+
+  const LockManager = require('../cli/lib/lock-manager');
+
+// Configuration
+const SESSIONS_DIR = '.claude/sessions';
+const ACTIVE_SESSION_FILE = path.join(SESSIONS_DIR, '.active-session');
+const lockManager = new LockManager(SESSIONS_DIR);
+
+// Living Context Configuration
+const CONTEXT_UPDATE_THRESHOLD = 2; // Update context every 2 interactions (lightweight)
+const SNAPSHOT_THRESHOLD = 5; // Full snapshot every 5 interactions (heavier)
+
+// Exit early if no active session
+if (!fs.existsSync(ACTIVE_SESSION_FILE)) {
+  process.exit(0);
+}
+
+// Read active session name
+let activeSession;
+try {
+  activeSession = fs.readFileSync(ACTIVE_SESSION_FILE, 'utf8').trim();
+} catch (err) {
+  process.exit(0);
+}
+
+if (!activeSession) {
+  process.exit(0);
+}
+
+const sessionDir = path.join(SESSIONS_DIR, activeSession);
+if (!fs.existsSync(sessionDir)) {
+  process.exit(0);
+}
+
+// Check if auto-capture is enabled
+const sessionMd = path.join(sessionDir, 'session.md');
+if (fs.existsSync(sessionMd)) {
+  try {
+    const content = fs.readFileSync(sessionMd, 'utf8');
+    if (content.includes('Auto-capture: disabled')) {
+      process.exit(0);
+    }
+  } catch (err) {
+    // Continue if we can't read the file
+  }
+}
+
+// File paths
+const stateFile = path.join(sessionDir, '.auto-capture-state');
+
+// Use lock to prevent race conditions during state read-modify-write
+const lock = lockManager.acquireLock(`auto-capture-${activeSession}`, {
+  timeout: 1000,
+  wait: true
+});
+
+if (!lock.acquired) {
+  // Could not acquire lock - skip this update to avoid blocking
+  // The next interaction will pick up the count
+  process.exit(0);
+}
+
+try {
+  // Initialize state if doesn't exist
+  let state = {
+    file_count: 0,
+    interaction_count: 0,
+    interactions_since_context_update: 0,
+    interactions_since_snapshot: 0,
+    last_context_update: '',
+    last_snapshot_timestamp: ''
+  };
+
+  if (fs.existsSync(stateFile)) {
+    try {
+      state = JSON.parse(fs.readFileSync(stateFile, 'utf8'));
+      // Ensure Living Context fields exist
+      state.interactions_since_context_update = state.interactions_since_context_update || 0;
+      state.interactions_since_snapshot = state.interactions_since_snapshot || 0;
+      state.last_context_update = state.last_context_update || '';
+      state.last_snapshot_timestamp = state.last_snapshot_timestamp || '';
+    } catch (err) {
+      // Use default state if parse fails
+    }
+  }
+
+  // Increment interaction count
+  state.interaction_count++;
+  state.interactions_since_context_update++;
+  state.interactions_since_snapshot++;
+
+  // INCREMENTAL LOGGING SYSTEM: Log conversation for later consolidation
+  // This replaces the blocking snapshot system with fast (~1-2ms) logging
+  const timestamp = new Date().toISOString();
+
+  // Read stdin to get transcript path and user prompt
+  let transcriptPath = null;
+  let userPrompt = null;
+  try {
+    const stdinData = fs.readFileSync(0, 'utf8').trim();
+    if (stdinData) {
+      const eventData = JSON.parse(stdinData);
+      transcriptPath = eventData.transcript_path || null;
+      userPrompt = eventData.prompt || null;
+    }
+  } catch (stdinErr) {
+    // If we can't read stdin, continue without transcript path
+    // This ensures hook doesn't fail if stdin format changes
+  }
+
+  // Log interaction incrementally (non-blocking, ~1-2ms)
+  try {
+    const ConversationLogger = require('../cli/lib/conversation-logger');
+    const logger = new ConversationLogger(sessionDir);
+
+    logger.logInteraction({
+      num: state.interaction_count,
+      timestamp: timestamp,
+      transcript_path: transcriptPath,
+      user_prompt: userPrompt,
+      state: state,
+      modified_files: state.modified_files || []
+    });
+  } catch (err) {
+    // Silent failure - don't block hook execution
+    // Consolidation will work with whatever data is available
+  }
+
+  // Note: Snapshot consolidation now happens at session start/continue
+  // via consolidate-worker.js running in background
+  // This eliminates the 10-15 second blocking issue
+
+  // Update state file atomically
+  const tempPath = `${stateFile}.tmp.${Date.now()}`;
+  try {
+    fs.writeFileSync(tempPath, JSON.stringify(state, null, 2));
+    fs.renameSync(tempPath, stateFile);
+  } catch (writeError) {
+    // Clean up temp file
+    if (fs.existsSync(tempPath)) {
+      try {
+        fs.unlinkSync(tempPath);
+      } catch (cleanupError) {
+        // Ignore
+      }
+    }
+    throw writeError;
+  }
+  } finally {
+    // Always release lock
+    lock.release();
+  }
+
+  // ORPHAN DETECTION: Check for orphaned hooks every 20 prompts
+  // This auto-cleans up hooks if plugin was uninstalled without cleanup
+  try {
+    const orphanCheckFile = path.join(SESSIONS_DIR, '.orphan-check-counter');
+    let checkCounter = 0;
+
+    if (fs.existsSync(orphanCheckFile)) {
+      try {
+        checkCounter = parseInt(fs.readFileSync(orphanCheckFile, 'utf8').trim(), 10) || 0;
+      } catch (e) {
+        // Use default
+      }
+    }
+
+    checkCounter++;
+
+    // Check every 20 prompts
+    if (checkCounter >= 20) {
+      const HooksManager = require('../cli/lib/hooks-manager');
+      const pluginRoot = path.dirname(__dirname);
+      const manager = new HooksManager(process.cwd());
+
+      // Detect orphaned hooks
+      const settings = manager.readSettings();
+      const orphaned = manager.detectOrphanedHooks(settings, pluginRoot);
+
+      if (orphaned.length > 0) {
+        // Auto-cleanup orphaned hooks
+        const cleaned = manager.removePluginHooks(settings, pluginRoot);
+        manager.createBackup();
+        manager.writeSettings(cleaned);
+
+        // NOTE: We cannot inject context here as this is UserPromptSubmit hook
+        // The cleanup happens silently in the background
+      }
+
+      checkCounter = 0; // Reset counter
+    }
+
+    // Write counter back
+    fs.writeFileSync(orphanCheckFile, checkCounter.toString());
+  } catch (orphanError) {
+    // Silent failure - don't block hook execution
+  }
+
+  process.exit(0);
+
+} catch (error) {
+  // Outer catch: Handle plugin missing/uninstalled
+  // Exit silently to avoid blocking Claude Code
+  process.exit(0);
+}
diff --git a/plugin.lock.json b/plugin.lock.json
new file mode 100644
index 0000000..b6e7795
--- /dev/null
+++ b/plugin.lock.json
@@ -0,0 +1,153 @@
+{
+  "$schema": "internal://schemas/plugin.lock.v1.json",
+  "pluginId": "gh:awudevelop/claude-plugins:session",
+  "normalized": {
+    "repo": null,
+    "ref": "refs/tags/v20251128.0",
+    "commit": "6ec7a588b3153f448ed95229402b71570157359c",
+    "treeHash": "064dd651da7c4c8e8214c1c6289c9072110d35aa2e92a597f7db105870faca37",
+    "generatedAt": "2025-11-28T10:14:05.352142Z",
+    "toolVersion": "publish_plugins.py@0.2.0"
+  },
+  "origin": {
+    "remote": "git@github.com:zhongweili/42plugin-data.git",
+    "branch": "master",
+    "commit": "aa1497ed0949fd50e99e70d6324a29c5b34f9390",
+    "repoRoot": "/Users/zhongweili/projects/openmind/42plugin-data"
+  },
+  "manifest": {
+    "name": "session",
+    "description": "v3.17.4 - Advanced session management for Claude Code with intelligent context tracking, automatic snapshots, and git history capture. Features zero-blocking conversation logging, AI-powered analysis, and optimized parallel subagent architecture. Never lose context!",
+    "version": "3.17.4"
+  },
+  "content": {
+    "files": [
+      {
+        "path": "README.md",
+        "sha256": "3c319320ab963e04cd6c699198b7f6db1b3d89804a940d8f6cf1c12c11abf37c"
+      },
+      {
+        "path": "hooks/stop.js",
+        "sha256": "37a676c6d29bff787dd6f3d18b86e4e14c92c0699c7ff32a352da3208c59d203"
+      },
+      {
+        "path": "hooks/post-tool-use.js",
+        "sha256": "4917273a37d71ecda0dc832cd1c452a3e8b7335a9984600301f644d5d9d486da"
+      },
+      {
+        "path": "hooks/session-start.js",
+        "sha256": "f62f244cb4129d1b67c819e5c6dff01b8b3e4b9b6afe5183132e70bc3f38d23b"
+      },
+      {
+        "path": "hooks/session-end.js",
+        "sha256": "20a3105b844442ba037bec12db191530e8fc40fb2d0a26336f188e55434a0e7a"
+      },
+      {
+        "path": "hooks/hooks.json",
+        "sha256": "0f1dbf53520fd253698253ff9486d5ae036d3ca51af5dc9c402e37084d8f50c2"
+      },
+      {
+        "path": "hooks/user-prompt-submit.js",
+        "sha256": "b1fb6c52127d2f3d9953efef04578f762dd89982f9323ca2b50de63d026edac4"
+      },
+      {
+        "path": ".claude-plugin/plugin.json",
+        "sha256": "3af37a81dae483514f556399eaba1620a25a8f6a6b9c5a527977adffe6cda128"
+      },
+      {
+        "path": "commands/plan-list.md",
+        "sha256": "693693eb8e5cd2006b9c4fa1f6b49ecf0633d9266bd2d08f665b293c66f99962"
+      },
+      {
+        "path": "commands/project-maps-stats.md",
+        "sha256": "a374903c057cba5fecca3d7c74da3f350d69c59ef79c356f54516b9b5d437b95"
+      },
+      {
+        "path": "commands/plan-status.md",
+        "sha256": "8d16bcb657e39a8511e4ae9aa8e20ddb3b4a33e208d1e19bf97370b3d4a365b7"
+      },
+      {
+        "path": "commands/plan-finalize.md",
+        "sha256": "454a4840bb50554491864231774d6260f896a1e39955ceda2d1814e770327641"
+      },
+      {
+        "path": "commands/project-maps-load.md",
+        "sha256": "d1f854feeda94633aa2a0846bb4a1c1cd4193765055f84fda564dd4a211c42b0"
+      },
+      {
+        "path": "commands/plan-save.md",
+        "sha256": "f3e2eddab439fccbfac377a18834262b0548e3d9ebd95e809bab7089f6035849"
+      },
+      {
+        "path": "commands/continue.md",
+        "sha256": "5980d87156eecdd37a4457d9e786047a1fe72e6d3fbe9454525897fc7a25d954"
+      },
+      {
+        "path": "commands/setup.md",
+        "sha256": "2d98544ae5c7de808eb768cca498ef85ffb698a35440d965384687c28345f234"
+      },
+      {
+        "path": "commands/status.md",
+        "sha256": "cdcfb2408bfced31d88cb0e630f51912a36ce74e90874bf76f06d2062d068381"
+      },
+      {
+        "path": "commands/plan-execute.md",
+        "sha256": "fbb322c14fdbc6c6f6be144a311e0f85487ffc9ed42902c2ad114d5bd969541a"
+      },
+      {
+        "path": "commands/continue.md.v3.6.4.backup",
+        "sha256": "2bf5ae7c4e32882a698f7551cad3551f59f6254fef465419ba9b25d1e3359857"
+      },
+      {
+        "path": "commands/project-maps-refresh.md",
+        "sha256": "e1a33fed3e0663826ec6d7d3ee3d4d9df84d95f6b205a8cc6bc93b02ab850a17"
+      },
+      {
+        "path": "commands/list.md",
+        "sha256": "1646d0b1670563cd520649efacdf09946a65f48c9d5d2ed49a9723b91d9d56f8"
+      },
+      {
+        "path": "commands/close.md",
+        "sha256": "888bd9d87e8f2b0a4360ee36f1661a2cfbc0a0f5f8a40aa880ff4b266d7a4c9f"
+      },
+      {
+        "path": "commands/project-maps-query.md",
+        "sha256": "6abcad95011a19f3f89de02efa5ca0d698abde6c7c7c2f4746c5bb2c16d448b1"
+      },
+      {
+        "path": "commands/git-decompress.md",
+        "sha256": "eb3585f1fa09c96785d8d6b0ccfcbf0c6e7078b52969b0914a338233ab84b4bd"
+      },
+      {
+        "path": "commands/delete.md",
+        "sha256": "10aa0161ee9a2dd3b29b02d6b7061f591d64ee9c6810c337ec7c4a7593137561"
+      },
+      {
+        "path": "commands/save.md",
+        "sha256": "5fc88d41d7c9dd80cfee82cf01f886452bf461557533286841bd8f7bf5f54a10"
+      },
+      {
+        "path": "commands/project-maps-generate.md",
+        "sha256": "7d1183f203b4fa3f51705c999d66ac4963af275b661129d1f4f61adc9bfe6b76"
+      },
+      {
+        "path": "commands/start.md",
+        "sha256": "ef5fb5dde77fdaecd887c384c0926c1987822c42f885b40cbe5794a9893439ec"
+      },
+      {
+        "path": "commands/project-maps-list.md",
+        "sha256": "d01bb7d92e97dae0e09742997dd15c42672ca45e735502f199d6beb899106313"
+      },
+      {
+        "path": "commands/plan-update.md",
+        "sha256": "9c0bb605c3d31f36999a128cfee4a677c2ec7f2d61c9da178ac3c6cf9fe254a9"
+      }
+    ],
+    "dirSha256": "064dd651da7c4c8e8214c1c6289c9072110d35aa2e92a597f7db105870faca37"
+  },
+  "security": {
+    "scannedAt": null,
+    "scannerVersion": null,
+    "flags": []
+  }
+}
\ No newline at end of file