Maestro/CLAUDE.md

CLAUDE.md

Essential guidance for working with this codebase. For detailed architecture, see ARCHITECTURE.md. For development setup and processes, see CONTRIBUTING.md.

Standardized Vernacular

Use these terms consistently in code, comments, and documentation:

UI Components

• Left Bar - Left sidebar with session list and groups (SessionList.tsx)
• Right Bar - Right sidebar with Files, History, Auto Run tabs (RightPanel.tsx)
• Main Window - Center workspace (MainPanel.tsx)
  • AI Terminal - Main window in AI mode (interacting with AI agents)
  • Command Terminal - Main window in terminal/shell mode
  • System Log Viewer - Special view for system logs (LogViewer.tsx)

Session States (color-coded)

• Green - Ready/idle
• Yellow - Agent thinking/busy
• Red - No connection/error
• Pulsing Orange - Connecting

Project Overview

Maestro is an Electron desktop app for managing multiple AI coding assistants (Claude Code, OpenAI Codex, Gemini CLI, Qwen3 Coder) simultaneously with a keyboard-first interface.

Quick Commands

npm run dev        # Development with hot reload
npm run dev:web    # Web interface development
npm run build      # Full production build
npm run clean      # Clean build artifacts
npm run lint       # TypeScript type checking (all configs)
npm run lint:eslint # ESLint code quality checks
npm run package    # Package for all platforms
npm run test       # Run test suite
npm run test:watch # Run tests in watch mode

Architecture at a Glance

src/
├── main/                    # Electron main process (Node.js)
│   ├── index.ts            # Entry point, IPC handlers
│   ├── process-manager.ts  # Process spawning (PTY + child_process)
│   ├── preload.ts          # Secure IPC bridge
│   ├── agent-detector.ts   # Agent detection and configuration
│   ├── agent-capabilities.ts # Agent capability definitions
│   ├── agent-session-storage.ts # Session storage interface
│   ├── parsers/            # Agent output parsers
│   │   ├── agent-output-parser.ts  # Parser interface
│   │   ├── claude-output-parser.ts # Claude Code parser
│   │   ├── opencode-output-parser.ts # OpenCode parser
│   │   └── error-patterns.ts # Error detection patterns
│   ├── storage/            # Session storage implementations
│   │   ├── claude-session-storage.ts
│   │   └── opencode-session-storage.ts
│   ├── tunnel-manager.ts   # Cloudflare tunnel support
│   ├── web-server.ts       # Fastify server for web/mobile interface
│   └── utils/execFile.ts   # Safe command execution
│
├── renderer/               # React frontend (desktop)
│   ├── App.tsx            # Main coordinator
│   ├── components/        # UI components
│   ├── hooks/             # Custom React hooks
│   ├── services/          # IPC wrappers (git.ts, process.ts)
│   ├── constants/         # Themes, shortcuts, priorities
│   └── contexts/          # Layer stack context
│
├── web/                    # Web/mobile interface
│   ├── mobile/            # Mobile-optimized React app
│   ├── components/        # Shared web components
│   └── hooks/             # Web-specific hooks
│
├── cli/                    # CLI tooling for batch automation
│   ├── commands/          # CLI command implementations
│   ├── services/          # Playbook and batch processing
│   └── index.ts           # CLI entry point
│
├── prompts/                # System prompts (editable .md files)
│   ├── wizard-*.md        # Wizard conversation prompts
│   ├── autorun-*.md       # Auto Run default prompts
│   └── index.ts           # Central exports
│
└── shared/                 # Shared types and utilities
    ├── types.ts           # Common type definitions
    └── templateVariables.ts # Template variable processing

Key Files for Common Tasks

Task	Primary Files
Add IPC handler	src/main/index.ts, src/main/preload.ts
Add UI component	src/renderer/components/
Add web/mobile component	src/web/components/, src/web/mobile/
Add keyboard shortcut	src/renderer/constants/shortcuts.ts, App.tsx
Add theme	src/renderer/constants/themes.ts
Add modal	Component + src/renderer/constants/modalPriorities.ts
Add setting	src/renderer/hooks/useSettings.ts, src/main/index.ts
Add template variable	src/shared/templateVariables.ts, src/renderer/utils/templateVariables.ts
Modify system prompts	src/prompts/*.md (wizard, Auto Run, etc.)
Add CLI command	src/cli/commands/, src/cli/index.ts
Configure agent	src/main/agent-detector.ts, src/main/agent-capabilities.ts
Add agent output parser	src/main/parsers/, src/main/parsers/index.ts
Add agent session storage	src/main/storage/, src/main/agent-session-storage.ts
Add agent error patterns	src/main/parsers/error-patterns.ts
Add playbook feature	src/cli/services/playbooks.ts
Modify wizard flow	src/renderer/components/Wizard/ (see Onboarding Wizard section)
Add tour step	src/renderer/components/Wizard/tour/tourSteps.ts
Modify file linking	src/renderer/utils/remarkFileLinks.ts (remark plugin for [[wiki]] and path links)

Core Patterns

1. Process Management

Each session runs two processes simultaneously:

• AI agent process (Claude Code, etc.) - spawned with -ai suffix
• Terminal process (PTY shell) - spawned with -terminal suffix

// Session stores both PIDs
session.aiPid       // AI agent process
session.terminalPid // Terminal process

2. Security Requirements

Always use execFileNoThrow for external commands:

import { execFileNoThrow } from './utils/execFile';
const result = await execFileNoThrow('git', ['status'], cwd);
// Returns: { stdout, stderr, exitCode } - never throws

Never use shell-based command execution - it creates injection vulnerabilities. The execFileNoThrow utility is the safe alternative.

3. Settings Persistence

Add new settings in useSettings.ts:

// 1. Add state
const [mySetting, setMySettingState] = useState(defaultValue);

// 2. Add wrapper that persists
const setMySetting = (value) => {
  setMySettingState(value);
  window.maestro.settings.set('mySetting', value);
};

// 3. Load in useEffect
const saved = await window.maestro.settings.get('mySetting');
if (saved !== undefined) setMySettingState(saved);

4. Adding Modals

1. Create component in src/renderer/components/
2. Add priority in src/renderer/constants/modalPriorities.ts
3. Register with layer stack:

import { useLayerStack } from '../contexts/LayerStackContext';
import { MODAL_PRIORITIES } from '../constants/modalPriorities';

const { registerLayer, unregisterLayer } = useLayerStack();
const onCloseRef = useRef(onClose);
onCloseRef.current = onClose;

useEffect(() => {
  if (isOpen) {
    const id = registerLayer({
      type: 'modal',
      priority: MODAL_PRIORITIES.YOUR_MODAL,
      onEscape: () => onCloseRef.current(),
    });
    return () => unregisterLayer(id);
  }
}, [isOpen, registerLayer, unregisterLayer]);

5. Theme Colors

Themes have 13 required colors. Use inline styles for theme colors:

style={{ color: theme.colors.textMain }}  // Correct
className="text-gray-500"                  // Wrong for themed text

6. Multi-Tab Sessions

Sessions support multiple AI conversation tabs:

// Each session has an array of tabs
session.aiTabs: AITab[]
session.activeTabId: string

// Each tab maintains its own conversation
interface AITab {
  id: string;
  name: string;
  logs: LogEntry[];           // Tab-specific history
  agentSessionId?: string;    // Agent session continuity
}

// Tab operations
const activeTab = session.aiTabs.find(t => t.id === session.activeTabId);

7. Execution Queue

Messages are queued when the AI is busy:

// Queue items for sequential execution
interface QueuedItem {
  type: 'message' | 'slashCommand';
  content: string;
  timestamp: number;
}

// Add to queue instead of sending directly when busy
session.executionQueue.push({ type: 'message', content, timestamp: Date.now() });

8. Auto Run

File-based document automation system:

// Auto Run state on session
session.autoRunFolderPath?: string;    // Document folder path
session.autoRunSelectedFile?: string;  // Currently selected document
session.autoRunMode?: 'edit' | 'preview';

// API for Auto Run operations
window.maestro.autorun.listDocuments(folderPath);
window.maestro.autorun.readDocument(folderPath, filename);
window.maestro.autorun.saveDocument(folderPath, filename, content);

Worktree Support: Auto Run can operate in a git worktree, allowing users to continue interactive editing in the main repo while Auto Run processes tasks in the background. When batchRunState.worktreeActive is true, read-only mode is disabled and a git branch icon appears in the UI. See useBatchProcessor.ts for worktree setup logic.

Code Conventions

TypeScript

• Strict mode enabled
• Interface definitions for all data structures
• Types exported via preload.ts for renderer

React Components

• Functional components with hooks
• Tailwind for layout, inline styles for theme colors
• tabIndex={-1} + outline-none for programmatic focus

Commit Messages

feat: new feature
fix: bug fix
docs: documentation
refactor: code refactoring

Session Interface

Key fields on the Session object (abbreviated - see src/renderer/types/index.ts for full definition):

interface Session {
  // Identity
  id: string;
  name: string;
  groupId?: string;             // Session grouping
  toolType: ToolType;           // 'claude-code' | 'aider' | 'terminal' | etc.
  state: SessionState;          // 'idle' | 'busy' | 'error' | 'connecting'
  inputMode: 'ai' | 'terminal'; // Which process receives input
  bookmarked?: boolean;         // Pinned to top

  // Paths
  cwd: string;                  // Current working directory (can change via cd)
  projectRoot: string;          // Initial directory (never changes, used for Claude session storage)
  fullPath: string;             // Full resolved path

  // Processes
  aiPid: number;                // AI process ID
  port: number;                 // Web server communication port

  // Multi-Tab Support (NEW)
  aiTabs: AITab[];              // Multiple Claude Code conversation tabs
  activeTabId: string;          // Currently active tab
  closedTabHistory: ClosedTab[]; // Undo stack for closed tabs

  // Logs (per-tab)
  shellLogs: LogEntry[];        // Terminal output history

  // Execution Queue (replaces messageQueue)
  executionQueue: QueuedItem[]; // Sequential execution queue

  // Usage & Stats
  usageStats?: UsageStats;      // Token usage and cost
  contextUsage: number;         // Context window usage percentage
  workLog: WorkLogItem[];       // Work tracking

  // Git Integration
  isGitRepo: boolean;           // Git features enabled
  changedFiles: FileArtifact[]; // Git change tracking
  gitBranches?: string[];       // Branch cache for completion
  gitTags?: string[];           // Tag cache for completion

  // File Explorer
  fileTree: any[];              // File tree structure
  fileExplorerExpanded: string[]; // Expanded folder paths
  fileExplorerScrollPos: number; // Scroll position

  // Web/Live Sessions (NEW)
  isLive: boolean;              // Accessible via web interface
  liveUrl?: string;             // Live session URL

  // Auto Run (NEW)
  autoRunFolderPath?: string;   // Auto Run document folder
  autoRunSelectedFile?: string; // Selected document
  autoRunMode?: 'edit' | 'preview'; // Current mode

  // Command History
  aiCommandHistory?: string[];  // AI input history
  shellCommandHistory?: string[]; // Terminal input history

  // Error Handling (NEW)
  agentError?: AgentError;        // Current agent error (auth, tokens, rate limit, etc.)
  agentErrorPaused?: boolean;     // Input blocked while error modal shown
}

interface AITab {
  id: string;
  name: string;
  logs: LogEntry[];             // Tab-specific conversation history
  agentSessionId?: string;      // Agent session for this tab
  scrollTop?: number;
  draftInput?: string;
}

IPC API Surface

The window.maestro API exposes:

Core APIs

• settings - Get/set app settings
• sessions / groups - Persistence
• process - Spawn, write, kill, resize
• fs - readDir, readFile
• dialog - Folder selection
• shells - Detect available shells
• logger - System logging

Agent & Agent Sessions

• agents - Detect, get, config, refresh, custom paths, getCapabilities
• agentSessions - Generic agent session storage API (list, read, search, delete)
• agentError - Agent error handling (clearError, retryAfterError)
• claude - (Deprecated) Claude Code sessions - use agentSessions instead

Git Integration

• git - Status, diff, isRepo, numstat, branches, tags, info
• git - Worktree support: worktreeInfo, getRepoRoot, worktreeSetup, worktreeCheckout
• git - PR creation: createPR, checkGhCli, getDefaultBranch

Web & Live Sessions

• web - Broadcast user input, Auto Run state, tab changes to web clients
• live - Toggle live sessions, get status, dashboard URL, connected clients
• webserver - Get URL, connected client count
• tunnel - Cloudflare tunnel: isCloudflaredInstalled, start, stop, getStatus

Automation

• autorun - Document and image management for Auto Run
• playbooks - Batch run configuration management
• history - Per-session execution history (see History API below)
• cli - CLI activity detection for playbook runs
• tempfile - Temporary file management for batch processing

History API

Per-session history storage with 5,000 entries per session (up from 1,000 global). Each session's history is stored as a JSON file in ~/Library/Application Support/Maestro/history/{sessionId}.json.

window.maestro.history = {
  getAll: (projectPath?, sessionId?) => Promise<HistoryEntry[]>,
  add: (entry) => Promise<boolean>,
  clear: (projectPath?, sessionId?) => Promise<boolean>,
  delete: (entryId, sessionId?) => Promise<boolean>,
  update: (entryId, updates, sessionId?) => Promise<boolean>,
  // For AI context integration:
  getFilePath: (sessionId) => Promise<string | null>,
  listSessions: () => Promise<string[]>,
  // External change detection:
  onExternalChange: (handler) => () => void,
  reload: () => Promise<boolean>,
};

AI Context Integration: Use getFilePath(sessionId) to get the path to a session's history file. This file can be passed directly to AI agents as context, giving them visibility into past completed tasks, decisions, and work patterns.

Utilities

• fonts - Font detection
• notification - Desktop notifications, text-to-speech
• devtools - Developer tools: open, close, toggle
• attachments - Image attachment management

Available Agents

ID	Name	Status	Notes
claude-code	Claude Code	Active	Primary agent, uses --print --verbose --output-format stream-json
opencode	OpenCode	Stub	Output parser implemented, session storage stub ready
terminal	Terminal	Internal	Hidden from UI, used for shell sessions
openai-codex	OpenAI Codex	Planned	Coming soon
gemini-cli	Gemini CLI	Planned	Coming soon
qwen3-coder	Qwen3 Coder	Planned	Coming soon

Additional ToolType values (aider, claude) are defined in types but not yet implemented in agent-detector.ts.

Agent Capabilities

Each agent declares capabilities that control UI feature availability. See src/main/agent-capabilities.ts for the full interface.

Capability	Description	UI Feature Controlled
supportsResume	Can resume previous sessions	Resume button
supportsReadOnlyMode	Has plan/read-only mode	Read-only toggle
supportsJsonOutput	Emits structured JSON	Output parsing
supportsSessionId	Emits session ID	Session ID pill
supportsImageInput	Accepts image attachments	Attach image button
supportsSlashCommands	Has discoverable commands	Slash autocomplete
supportsSessionStorage	Persists browsable sessions	Sessions browser
supportsCostTracking	Reports token costs	Cost widget
supportsUsageStats	Reports token counts	Context window widget
supportsBatchMode	Runs per-message	Batch processing
supportsStreaming	Streams output	Real-time display
supportsResultMessages	Distinguishes final result	Message classification

For detailed agent integration guide, see AGENT_SUPPORT.md.

Onboarding Wizard

The wizard (src/renderer/components/Wizard/) guides new users through first-run setup, creating AI sessions with Auto Run documents.

Wizard Architecture

src/renderer/components/Wizard/
├── MaestroWizard.tsx           # Main orchestrator, screen transitions
├── WizardContext.tsx           # State management (useReducer pattern)
├── WizardResumeModal.tsx       # Resume incomplete wizard dialog
├── WizardExitConfirmModal.tsx  # Exit confirmation dialog
├── ScreenReaderAnnouncement.tsx # Accessibility announcements
├── screens/                    # Individual wizard steps
│   ├── AgentSelectionScreen.tsx    # Step 1: Choose AI agent
│   ├── DirectorySelectionScreen.tsx # Step 2: Select project folder
│   ├── ConversationScreen.tsx      # Step 3: AI project discovery
│   └── PhaseReviewScreen.tsx       # Step 4: Review generated plan
├── services/                   # Business logic
│   ├── wizardPrompts.ts           # System prompts, response parser
│   ├── conversationManager.ts     # AI conversation handling
│   └── phaseGenerator.ts          # Document generation
└── tour/                       # Post-setup walkthrough
    ├── TourOverlay.tsx            # Spotlight overlay
    ├── TourStep.tsx               # Step tooltip
    ├── tourSteps.ts               # Step definitions
    └── useTour.tsx                # Tour state management

Wizard Flow

1. Agent Selection → Select available AI (Claude Code, etc.) and project name
2. Directory Selection → Choose project folder, validates Git repo status
3. Conversation → AI asks clarifying questions, builds confidence score (0-100)
4. Phase Review → View/edit generated Phase 1 document, choose to start tour

When confidence reaches 80+ and agent signals "ready", user proceeds to Phase Review where Auto Run documents are generated and saved to Auto Run Docs/.

Triggering the Wizard

// From anywhere with useWizard hook
const { openWizard } = useWizard();
openWizard();

// Keyboard shortcut (default)
Cmd+Shift+N  // Opens wizard

// Also available in:
// - Command K menu: "New Agent Wizard"
// - Hamburger menu: "New Agent Wizard"

State Persistence (Resume)

Wizard state persists to wizardResumeState in settings when user advances past step 1. On next app launch, if incomplete state exists, WizardResumeModal offers "Resume" or "Start Fresh".

// Check for saved state
const hasState = await hasResumeState();

// Load saved state
const savedState = await loadResumeState();

// Clear saved state
clearResumeState();

Tour System

The tour highlights UI elements with spotlight cutouts:

// Add data-tour attribute to spotlight elements
<div data-tour="autorun-panel">...</div>

// Tour steps defined in tourSteps.ts
{
  id: 'autorun-panel',
  title: 'Auto Run in Action',
  description: '...',
  selector: '[data-tour="autorun-panel"]',
  position: 'left',  // tooltip position
  uiActions: [       // UI state changes before spotlight
    { type: 'setRightTab', value: 'autorun' },
  ],
}

Customization Points

What	Where
Add wizard step	WizardContext.tsx (WIZARD_TOTAL_STEPS, WizardStep type, STEP_INDEX)
Modify wizard prompts	src/prompts/wizard-*.md (content), services/wizardPrompts.ts (logic)
Change confidence threshold	READY_CONFIDENCE_THRESHOLD in wizardPrompts.ts (default: 80)
Add tour step	tour/tourSteps.ts array
Modify Auto Run document format	src/prompts/wizard-document-generation.md
Change wizard keyboard shortcut	shortcuts.ts → openWizard

Related Settings

// In useSettings.ts
wizardCompleted: boolean    // First wizard completion
tourCompleted: boolean      // First tour completion
firstAutoRunCompleted: boolean  // Triggers celebration modal

Debugging

Focus Not Working

1. Add tabIndex={0} or tabIndex={-1}
2. Add outline-none class
3. Use ref={(el) => el?.focus()} for auto-focus

Settings Not Persisting

1. Check wrapper function calls window.maestro.settings.set()
2. Check loading code in useSettings.ts useEffect

Modal Escape Not Working

1. Register with layer stack (don't handle Escape locally)
2. Check priority is set correctly