lair404 Multi-Session Orchestrator

VS Code extension for monitoring and controlling Claude Code multi-session workflows with real-time user action handling.

Features

📊 Active Tasks View

• Real-time monitoring of running tasks
• Task statistics (running, completed, failed, success rate)
• Cost and token usage tracking
• Clickable task output logs

📋 Task Queue Management

• View Queue: See all queued tasks with details (model, category, budget, timeout)
• Add Task: Create new tasks via interactive dialog
• Edit Task: Modify queued tasks (disabled for running tasks)
• Delete Task: Remove tasks from queue (disabled for running tasks)
• Queue Summary: See estimated costs and duration

⚠️ User Actions Required (v2.1.0 - NEW!)

• Real-time Choice Detection: Automatically detects when tasks need user input
• Interactive Choice Selection: View tasks waiting for choices/approvals in tree view
• Multi-step Prompts: Expand tasks to see all available options
• Status Bar Integration: See count of waiting tasks at a glance (click to view)
• Notifications: Optional popups when tasks need input (configurable)
• Dashboard API: Responds to choice APIs on dashboard server
• Offline Fallback: Automatically falls back to bash CLI when dashboard unavailable
• Force Proceed: Continue tasks without choosing if needed

How it works:

1. FileWatcherManager monitors .output files for choice patterns
2. When detected, task appears in "⚠️ User Actions Required" view
3. Expand task to see all options (Option 1/2/3, Choice A/B/C, etc.)
4. Click on choice or use context menu to respond
5. Status bar shows count of waiting tasks (yellow badge when > 0)

🎮 Sprint Control

• Start Sprint: Launch multi-session orchestrator from VS Code
• Pause Sprint: Move queue to paused state (orchestrator stops picking up tasks)
• Resume Sprint: Restore paused queue
• Cancel Sprint: Terminate running sprint immediately

📜 Session Logs

• Browse Logs: View session logs grouped by sprint
• Open Log: Click to open session files in editor
• Real-time Streaming: Stream logs to Output channel as tasks run
• Auto-refresh: Logs update automatically when new sessions complete

🎯 Parallel Task Selector (v1.8.0 - NEW!)

• View Parallel Tasks: See interference-free task sets from enriched-todos MCP
• Reload Tasks: Click refresh icon to fetch latest parallel tasks
• Postpone Task: Mark tasks as low priority (right-click → Postpone)
• Assign to Queue: Add tasks to spawn-queue.json (right-click → Assign to Queue)
• Resource Tracking: See CPU/IO/Network/Memory usage across parallel tasks
• Interference Detection: View excluded tasks with reasons (scope conflicts, resource limits)
• Task Details: Expand tasks to see description, type, scope, resource class, estimate, affected paths
• Visual Indicators: Color-coded icons (green=ready, yellow=warning, red=excluded)

💬 Pipeline Chat (Phase 5)

• Full Chat Interface: Claude-style chat directly in VS Code
• Pipeline Templates: 5 pre-configured pipeline types (Quick Fix, Feature Build, Test Suite, Refactor, Deploy Check)
• Interactive Prompting: Type your prompt and send to queue
• Message History: See all your prompts and responses
• Export Chat: Save conversation history as JSON
• Clear Chat: Reset conversation anytime

🔍 Web Search (v1.70.0 - UPGRADED!)

• Multi-Provider Search: Query multiple search engines simultaneously (Tavily, Brave, Google, Bing, Serper)
• AI Synthesis: Optional Groq-powered answer generation with citations
• Markdown Rendering: Beautiful formatted AI answers with headings, lists, code blocks (using marked.js)
• Interactive Citations: Clickable [1] references with hover previews showing title, URL, and snippet
• Quality Badges: Visual indicators (High/Medium/Low) based on consensus percentage and source count
• Keyboard Shortcuts:
  • Cmd+K - Focus search input
  • Ctrl+Enter - Submit search
  • Ctrl+L - Clear chat
• Search History: Automatic persistence across sessions (max 20 searches)
• Provider Selection: Choose specific providers or search all at once
• Result Fusion: CombMNZ algorithm ranks results by consensus
• Chat Interface: Conversation-style search history with state persistence
• Performance Metrics: See search time, consensus stats, and cache hits
• Export: Save search history to Markdown

Installation

Quick Install (if you have .vsix package)

code --install-extension multi-session-orchestrator-1.8.0.vsix

Development Install (from source)

cd extensions/multi-session-orchestrator
npm install
npm run compile

Then press Cmd+Shift+P → "Debug: Start Debugging" (or Cmd+Shift+D → click ▶ Run Extension)

See INSTALL.md for detailed installation instructions.

Usage

Adding a Task

1. Open Multi-Session sidebar
2. Click "Task Queue" section
3. Click "+" icon or run command "Add Task to Queue"
4. Fill in task details

Using Pipeline Chat

1. Open Multi-Session sidebar
2. Click "Pipeline Chat" section
3. Select a Template (left sidebar):
   • 🚀 Quick Fix - Fast bug fix (haiku, $3)
   • ✨ Feature Build - End-to-end feature (sonnet, $10)
   • 🧪 Test Suite - Comprehensive tests (sonnet, $8)
   • 🔧 Refactor Code - Improve quality (sonnet, $7)
   • ✅ Deploy Check - Pre-deployment (haiku, $5)
4. Type Your Prompt in the text area (Ctrl+Enter to send)
5. Send - Task gets added to queue automatically
6. View Messages - See your conversation history
7. Export - Save chat as JSON file
8. Clear - Reset conversation

Managing Sprints

Start Sprint: Click ▶️ play icon in Task Queue toolbar

Pause Sprint: Click ⏸ pause icon - Queue moves to paused state

Resume Sprint: Click ▶️ continue icon - Paused queue restores

Cancel Sprint: Click ⏹ stop icon - Terminates entire sprint

Viewing Logs

Browse Session Logs: Expand sprint groups in Session Logs view

Stream Logs in Real-time: Right-click log → "Stream Log (Real-time)"

Stop All Streams: Click 🛑 stop icon in Session Logs toolbar

Using Parallel Task Selector

1. Open Multi-Session sidebar
2. Click "🎯 Parallel Task Selector" section (after Task Queue)
3. Click Refresh Icon to load parallel tasks from enriched-todos.json
4. View Parallel Tasks:
   • See count of ready tasks
   • Expand to see details (description, type, scope, resource, estimate, paths)
   • Hover for full tooltip
5. Right-click Task for actions:
   • Assign to Queue - Adds task to spawn-queue.json for execution
   • Postpone - Marks task as low priority
6. View Resource Usage:
   • See CPU/IO/Network/Memory task distribution
7. View Excluded Tasks:
   • See why tasks were excluded (scope conflicts, resource limits, max tasks)

Requirements:

• .claude/enriched-todos.json must exist (created by enriched-todos MCP)
• Run bulk_enrich via enriched-todos MCP to populate data

Using Web Search

1. Open Multi-Session sidebar
2. Click "Web Search" section
3. Select Providers (left top):
   • 🌐 All - Search all providers
   • Individual providers: Tavily, Brave, Google, Bing, Serper
4. Configure Options:
   • ☑️ AI Synthesis - Toggle Groq-powered answer generation
   • Max Results: 5, 10, or 20 results per provider
5. Enter Query in text area (Cmd+K to focus, Ctrl+Enter to search)
6. View Results:
   • 🤖 AI-synthesized answer with markdown formatting and interactive citations [1] (if enabled)
   • 🏅 Quality badge (High/Medium/Low consensus)
   • 📑 Ranked search results with links, snippets, and scores
   • 📊 Fusion statistics (consensus, total results)
   • ⏱️ Performance metrics (timing, cache status)
   • 💡 Hover over [1] to see citation preview (title, URL, snippet)
   • 🔗 Click [1] to open source in browser
7. Export - Save search history to Markdown file
8. Clear - Reset search history

Requirements:

• Multi-web search server must be running at http://localhost:3000
• Start server: cd tools/multi-web && npm start

Version History

v1.8.0 (2026-01-10) 🎯 Parallel Task Selector & Multi-Web Search v1.70.0

Major Features:

• 🎯 Parallel Task Selector: View interference-free task sets from enriched-todos MCP
  • Reload, postpone, assign actions
  • Resource usage tracking (CPU, IO, Network, Memory)
  • Interference detection with exclusion reasons
  • Tree view with expandable task details
  • Color-coded visual indicators
• 🔍 Multi-Web Search v1.70.0 Upgrade: Full parity with multi-dashboard UX
  • ✨ Markdown Rendering: AI answers beautifully formatted with headings, lists, code blocks (using marked.js v11.1.1)
  • 🔗 Interactive Citations: Click [1] to open source, hover for preview with title/URL/snippet
  • 🏅 Quality Badges: Visual indicators (High/Medium/Low) based on consensus % and source count
  • ⌨️ Keyboard Shortcuts: Cmd+K focus search, Ctrl+Enter submit
  • 💾 Search History: Automatic persistence across sessions (max 20 searches)
  • 🎨 AI Answer Highlight: Blue gradient border for synthesized answers
• 🔧 MCP Client Integration: Abstracted MCP server communication layer
  • MCPClient.ts utility class for enriched-todos MCP
  • JSON-RPC request/response handling
  • Simplified interference detection (MVP: scope-based blocking)
  • Error handling and fallbacks

Technical Improvements:

• Created ParallelTaskSelectorProvider (489 lines)
• Created MCPClient utility (150 lines)
• Updated WebSearchPanelProvider with v1.70.0 features
• Added marked.js CDN integration (CSP-compliant)
• State persistence for search history
• TypeScript compilation: 0 errors, 0 warnings

Known Limitations:

• MCP integration MVP: Reads enriched-todos.json directly (future: full stdio MCP)
• Simple interference detection: Same-scope blocking only (future: 5x5 matrix)
• No real-time updates: Manual refresh required (future: file watcher)
• Markdown rendering: Requires internet for CDN (fallback: plain text)

Related:

• See CHANGELOG.md for complete v1.8.0 release notes
• See PARALLEL-TASK-SELECTOR-IMPLEMENTATION.md for feature guide

v1.7.0 (2026-01-10) 🔍 NEW - Multi-Web Search Integration

Phase 6: Web Search & UX Improvements

• 🔍 Web Search Panel: Integrated multi-provider search with AI synthesis
• 🌐 Multi-Provider Support: Query Tavily, Brave, Google, Bing, Serper simultaneously
• 🤖 AI Synthesis: Optional Groq-powered answer generation with citations
• 📊 Result Fusion: CombMNZ algorithm ranks results by consensus across providers
• 💬 Chat Interface: Conversation-style search history with state persistence
• 📈 Performance Metrics: Search time, consensus stats, cache hits displayed
• 📤 Export to Markdown: One-click export of search history with results and stats
• ⌨️ Keyboard Shortcuts: Ctrl+Enter to search, integrated with existing accessibility features
• 🎨 UX Improvements: Session Logs moved to bottom of sidebar for better workflow

Technical Improvements:

• WebSearchPanelProvider with full webview UI
• Integration with multi-web search API (http://localhost:3000)
• Provider selection buttons (all providers or specific ones)
• Synthesize toggle and max results configuration
• State persistence across webview reloads
• Real-time result display with fusion statistics
• Markdown export with formatted results
• Accessibility: ARIA labels, keyboard navigation, screen reader support

Related:

• goal-multi-session-014: Multi-Web Search Integration
• tools/multi-web: Multi-web search backend server
• WebSearchPanelProvider.ts: Main implementation

v1.6.0 (2026-01-10) 🚀 Production Excellence

Phase 3: Robustness & State Management

• 💾 State Persistence: Chat history and active pipeline survive webview reloads
• 🔄 Retry Logic: Failed API calls retry up to 3 times with exponential backoff (1s → 2s → 4s)
• ⏱️ Request Timeouts: 10-second timeout for all API calls with clear error messages
• 🛡️ Error Boundaries: Comprehensive error handling with descriptive messages
• 📊 Network Resilience: Graceful degradation when dashboard temporarily unavailable

Phase 4: Accessibility & Polish

• ♿ ARIA Labels: Full screen reader support for all interactive elements
• ⌨️ Keyboard Shortcuts:
  • Ctrl+Enter - Send message
  • Ctrl+L - Clear chat
  • Ctrl+E - Export chat
  • Ctrl+/ - Focus input
  • Arrow keys - Navigate pipeline templates
  • Enter/Space - Select pipeline
  • Escape - Blur input
• 🎯 Focus Management: Clear focus indicators following VS Code design system
• 🎨 High Contrast Support: Proper focus borders and theme-aware styling
• 📋 Keyboard Navigation: Full functionality without mouse

Technical Improvements:

• Retry with exponential backoff for network resilience
• AbortController for request timeout management
• In-memory state persistence across webview reloads
• Comprehensive ARIA roles (navigation, main, log, button)
• Tab order optimization with tabindex attributes
• Screen-reader-only hints for keyboard shortcuts

v1.5.0 (2026-01-10) ✨ Backend Integration & Security

• 🔒 Security Hardening: Content Security Policy (CSP) with nonce-based script validation
• 🔌 Backend Integration: HTTP + WebSocket connection to multi-dashboard (port 3333)
• 🔄 Real-time Updates: WebSocket notifications for task completions and sprint events
• 📋 Dynamic Templates: Load pipeline templates from dashboard API (40+ endpoints)
• 🎯 Smart Sprint Detection: Automatically detects active sprint file (no more sprint999 hardcoding)
• 🔀 Graceful Fallback: Works offline with file-based mode when dashboard unavailable
• ✅ Auto-reconnect: WebSocket automatically reconnects on connection loss (5 attempts)
• 📊 Extended API Access: Templates, learnings, statistics, sprint status from dashboard
• 🎨 Enhanced Chat: Pipeline templates loaded from backend, fallback to local defaults

v1.4.0 (2026-01-10)

• ✨ Phase 5: Pipeline Chat with full webview interface
• 💬 Claude-style chat directly in VS Code
• 📋 5 pre-configured pipeline templates
• 💾 Export conversation history
• 🎨 Beautiful UI matching VS Code theme

v1.3.0 (2026-01-10)

• ✨ Phase 4: Session Logs & Real-time Streaming
• ✨ Phase 3C: Queue Management Actions (Edit, Delete, Pause, Resume, Cancel)

v1.2.0 (2026-01-10)

• ✨ Phase 3B: Add Task to Queue

v1.1.0 (2026-01-10)

• ✨ Phase 3A: Queue Visualization

v1.0.0 (2026-01-09)

• ✨ Initial release with Active Tasks monitoring