@hienlh/ppm 0.8.86 → 0.8.87

package/docs/code-standards.md

@@ -86,6 +86,31 @@ async *streamMessages(input: string) {
 provider.sendMessage(input).then(...)
 ```
 
+### Long-Lived Streaming Sessions
+
+For chat-like features, maintain persistent streaming sessions instead of per-query execution:
+
+```typescript
+// Good: Session-scoped streaming (v0.8.55+)
+// Provider maintains AsyncGenerator per session
+const streaming = provider.sendMessage(sessionId, content);
+for await (const event of streaming) {
+  // Handle event (can receive multiple messages from same generator)
+  yield event;
+}
+
+// Avoid: Per-message query execution
+for (const message of messages) {
+  const result = await query(message); // Restarts SDK each time
+  yield result;
+}
+
+// Key Pattern: Decoupled streaming loop
+// Store streaming generator in session entry
+// Follow-up messages push into existing generator
+// FE disconnect ≠ abort (BE owns the connection)
+```
+
 ### Error Handling
 Use try-catch for async operations. Throw structured errors:
 
@@ -221,6 +246,63 @@ try {
 }
 ```
 
+### Provider Interface Pattern (Multi-Provider)
+
+PPM supports multiple AI providers through an extensible interface pattern:
+
+```typescript
+// Core interface defines required methods + optional capability methods
+interface AIProvider {
+  id: string;
+  name: string;
+
+  // Required: All providers must implement these
+  createSession(config: SessionConfig): Promise<Session>;
+  resumeSession(sessionId: string): Promise<Session>;
+  listSessions(): Promise<SessionInfo[]>;
+  deleteSession(sessionId: string): Promise<void>;
+  sendMessage(sessionId: string, message: string, opts?: SendMessageOpts): AsyncIterable<ChatEvent>;
+
+  // Optional: Providers implement what they support
+  abortQuery?(sessionId: string): void;
+  getMessages?(sessionId: string): Promise<ChatMessage[]>;
+  listSessionsByDir?(dir: string): Promise<SessionInfo[]>;
+  ensureProjectPath?(sessionId: string, path: string): void;
+  isAvailable?(): Promise<boolean>;
+}
+```
+
+**SDK-based Providers (Claude):**
+- Use Anthropic SDK for advanced features
+- Implement all optional methods
+- Streaming via SDK async generators
+
+**CLI-based Providers (Cursor, Codex, Gemini):**
+- Extend `CliProvider` abstract class
+- Implement: `buildArgs()`, `mapEvent()`, `extractSessionId()`, `isAvailable()`
+- Shared: spawn, parse NDJSON, abort, cleanup
+- Override `spawnProcess()` for provider-specific quirks (e.g., workspace trust)
+- Override `listSessions()` to read native provider history (e.g., SQLite)
+
+**Usage Pattern (Type-Safe Optional Calls):**
+```typescript
+// Good: Optional chaining for capabilities
+await provider.abortQuery?.(sessionId); // Silently skips if not implemented
+const messages = await provider.getMessages?.(sessionId) ?? [];
+
+// Avoid: Duck-typing checks (old pattern)
+if ("abortQuery" in provider) {
+  (provider as any).abortQuery(sessionId); // Not type-safe
+}
+```
+
+**Adding New Providers:**
+1. Create provider class implementing/extending AIProvider or CliProvider
+2. Implement `isAvailable()` for optional registration
+3. Register in `ProviderRegistry` via `bootstrapProviders()` if binary available
+4. Add event mapper if CLI-based (NDJSON → ChatEvent)
+5. ~100-150 lines per provider (extends CliProvider)
+
 ## API Conventions
 
 ### Response Envelope
@@ -252,18 +334,35 @@ GET    /api/project/:name/files/...   # Files (project-scoped)
 Structure WebSocket messages as typed JSON objects:
 
 ```typescript
-// Client -> Server (chat)
-{ type: "message"; content: string }
+// Client -> Server (chat, v0.8.55+)
+{ type: "message"; content: string; priority?: "now"|"next"|"later"; images?: {id: string; data: string}[] }
 { type: "cancel" }
-{ type: "approval_response"; requestId: string; approved: boolean }
+{ type: "approval_response"; requestId: string; approved: boolean; reason?: string; data?: unknown }
+{ type: "ready" }  // FE handshake after WS open
 
 // Server -> Client (chat)
-{ type: "text"; content: string }
-{ type: "tool_use"; tool: string; input: unknown }
+{ type: "text"; content: string; parentToolUseId?: string }
+{ type: "thinking"; content: string; parentToolUseId?: string }
+{ type: "tool_use"; tool: string; input: unknown; toolUseId?: string; parentToolUseId?: string }
+{ type: "tool_result"; output: string; isError?: boolean; toolUseId?: string; parentToolUseId?: string }
 { type: "approval_request"; requestId: string; tool: string; input: unknown }
-{ type: "done"; sessionId: string }
+{ type: "done"; sessionId: string; contextWindowPct?: number; resultSubtype?: string }
 { type: "error"; message: string }
-```
+{ type: "account_info"; accountId: string; accountLabel: string }
+{ type: "phase_changed"; phase: SessionPhase; elapsed?: number }
+{ type: "session_state"; sessionId: string; phase: SessionPhase; pendingApproval: {...}|null; sessionTitle: string|null }
+{ type: "turn_events"; events: unknown[] }  // Buffered events on reconnect
+{ type: "title_updated"; title: string }
+{ type: "ping" }  // Server keepalive
+```
+
+**New Fields (v0.8.55+):**
+- `priority` — Message priority for queue ordering (future: "now" interrupts, "next" queues first, "later" queues at end)
+- `images` — Image attachments sent with message
+- `phase_changed` — Phase transitions (initializing → connecting → thinking/streaming → idle)
+- `session_state` — Session state snapshot on WS open/ready (includes current phase and pending approval)
+- `turn_events` — Buffered events sent on reconnection for sync
+- `parentToolUseId` — Hierarchical tool call support (nested tool results)
 
 ### Status Codes
 Use standard HTTP status codes:
@@ -646,3 +745,129 @@ if (options.share) {
 
 This keeps startup fast when features aren't used.
 
+---
+
+## Tab ID & URL Conventions (v0.8.77+)
+
+### Deterministic Tab IDs
+
+Tab IDs are derived from type + metadata, not random:
+
+```typescript
+// Format: "{type}:{identifier}" for most tabs, "{type}" for singletons
+editor:src/index.ts           // Editor tab with file path
+chat:claude/abc123            // Chat with provider/session ID
+terminal:1                    // Terminal tab index
+database:conn-1/users         // Database: connection ID / table
+git-graph                      // Singleton: no identifier
+settings                       // Singleton: no identifier
+
+// Derivation in panel-utils.ts
+export function deriveTabId(type: TabType, metadata?: Record<string, unknown>): string {
+  switch (type) {
+    case "editor":
+      return `editor:${metadata?.filePath ?? "untitled"}`;
+    case "chat": {
+      const provider = metadata?.providerId ?? "default";
+      return `chat:${provider}/${metadata?.sessionId ?? randomId()}`;
+    }
+    case "terminal":
+      return `terminal:${metadata?.terminalIndex ?? 1}`;
+    case "git-graph":
+      return "git-graph"; // Singleton
+    // ... other cases
+  }
+}
+```
+
+### URL Format
+
+URLs are built from project name + deterministic tab ID:
+
+```typescript
+// Format: /project/{projectName}/{type}/{identifier}
+/project/ppm                          // Project root (no active tab)
+/project/ppm/editor/src/index.ts      // Open editor
+/project/ppm/chat/claude/abc123       // Open chat
+/project/ppm/terminal/1               // Open terminal
+/project/ppm/database/conn-1/users    // Open database
+/project/ppm/git-graph                // Git history (singleton)
+
+// URL building in use-url-sync.ts
+export function buildUrl(projectName: string, tabId: string | null): string {
+  const colonIdx = tabId?.indexOf(":");
+  if (colonIdx === -1) {
+    // Singleton
+    url += `/${tabId}`;
+  } else {
+    const [type, identifier] = tabId.split(":", 1);
+    url += `/${type}/${identifier}`;
+  }
+  return url;
+}
+```
+
+### Deep Linking
+
+When user navigates to a URL, parse it and auto-create tabs:
+
+```typescript
+// In use-url-sync hook
+export function parseUrlState(): UrlState {
+  const match = path.match(/^\/project\/([^/]+)(?:\/([^/]+)(\/.*)?)?/);
+  const [, projectName, tabType, tabIdentifierPath] = match;
+
+  // Build tab metadata from URL
+  const metadata = buildMetadataFromUrl(tabType, tabIdentifier, projectName);
+
+  // Auto-open tab if it doesn't exist
+  if (metadata) {
+    panelStore.openTab({
+      type: tabType,
+      title: getTabTitle(tabType, metadata),
+      metadata,
+    });
+  }
+}
+```
+
+### Server-Side Workspace Persistence
+
+Tab layouts are persisted in the `workspace_state` SQLite table:
+
+```typescript
+// workspace.ts routes
+GET  /api/project/:name/workspace  → { layout: PanelLayout, updatedAt: string }
+PUT  /api/project/:name/workspace  → { updatedAt: string }
+
+// PanelLayout structure
+interface PanelLayout {
+  panels: Record<string, Panel>;    // panelId → { tabs, activeTabId }
+  grid: string[][];                 // Row/column grid of panel IDs
+  focusedPanelId: string;
+}
+
+// Sync is debounced (1.5s) client-side after layout changes
+// Latest-wins: server timestamp compared with client localStorage
+```
+
+### Migration from Random IDs
+
+Old tab IDs (tab-xxxx) are automatically migrated to deterministic format:
+
+```typescript
+// In panel-utils.ts
+export function migrateTabIdToDeterministic(tab: Tab): Tab {
+  if (tab.id.startsWith("tab-")) {
+    // Convert to deterministic ID
+    return {
+      ...tab,
+      id: deriveTabId(tab.type, tab.metadata),
+    };
+  }
+  return tab;
+}
+```
+
+This happens on first load per project; old URLs redirect to project root.
+

package/docs/codebase-summary.md

@@ -104,7 +104,7 @@ ppm/
 │       │   ├── use-health-check.ts  # Detect server crashes/restarts via health endpoint
 │       │   ├── use-usage.ts         # Fetch token usage from backend
 │       │   └── use-push-notification.ts # Web push notifications via Service Worker
-│       ├── lib/                     # Utilities (11 files)
+│       ├── lib/                     # Utilities (12 files)
 │       │   ├── api-client.ts        # Fetch wrapper with auth token, envelope unwrapping
 │       │   ├── api-settings.ts      # AI settings API client (GET/PUT /api/settings/ai)
 │       │   ├── ws-client.ts         # WebSocket with exponential backoff + Cloudflare handshake
@@ -113,6 +113,7 @@ ppm/
 │       │   ├── project-palette.ts   # 12-color palette for project avatars
 │       │   ├── use-monaco-theme.ts  # Sync Monaco Editor theme with app theme
 │       │   ├── color-utils.ts       # WCAG color contrast helper
+│       │   ├── csv-parser.ts        # CSV state-machine parser/serializer
 │       │   └── utils.ts             # Helpers (cn, randomId, basename, etc.)
 │       ├── styles/
 │       │   └── globals.css          # Tailwind directives, custom CSS
@@ -131,9 +132,12 @@ ppm/
 │           │   ├── usage-badge.tsx  # Token usage display
 │           │   ├── attachment-chips.tsx # Display attached files
 │           │   └── chat-placeholder.tsx # Empty state
-│           ├── editor/              # Code editor (650+ LOC, 3 files)
+│           ├── editor/              # Code editor (800+ LOC, 6 files)
 │           │   ├── code-editor.tsx  # Monaco Editor integration (@monaco-editor/react, v2.0+)
 │           │   ├── diff-viewer.tsx  # Monaco diff viewer for git diffs (v2.0+)
+│           │   ├── editor-breadcrumb.tsx # VSCode-style breadcrumb with nested dropdown
+│           │   ├── editor-toolbar.tsx # File-type contextual toolbar
+│           │   ├── csv-preview.tsx  # CSV table viewer with @tanstack/react-table
 │           │   └── editor-placeholder.tsx
 │           ├── explorer/            # File tree (489 LOC, 2 files)
 │           │   ├── file-tree.tsx    # Directory tree view
@@ -346,11 +350,13 @@ UI updates staged/unstaged lists
 | @monaco-editor/react | Code editor | 4.7.0 |
 | xterm | Terminal emulator | 6.0 |
 | zustand | State management | 5.0.11 |
-| @anthropic-ai/claude-agent-sdk | AI provider | 0.2.76 |
+| @anthropic-ai/claude-agent-sdk | AI provider | 0.2.81 |
 | vite | Frontend bundler | 8.0 |
 | tailwindcss | Utility CSS | 4.2 |
 | radix-ui | Accessible components | 1.4.3 |
 | next-themes | Theme switcher | 0.4.6 |
+| @tanstack/react-table | Table library | 8.21.3 |
+| @tanstack/react-virtual | Virtual scrolling | 3.13.23 |
 
 ## Build Output
 

package/docs/design-guidelines.md

@@ -393,6 +393,27 @@ screens: {
 // - Word wrap toggle (Alt+Z)
 // - Monaco diff viewer for git diffs
 // - Theme sync with app dark/light mode
+// - VSCode-style breadcrumb navigation (EditorBreadcrumb)
+// - File-type contextual toolbar (EditorToolbar)
+```
+
+### CSV Preview
+
+```tsx
+// src/web/components/editor/csv-preview.tsx
+// Table viewer for CSV files with sorting and editing
+
+<CsvPreview
+  content={csvString}
+  onContentChange={setCsvString}
+/>
+
+// Features:
+// - State-machine CSV parser (handles quoted fields, embedded commas/newlines)
+// - Virtual scrolling (@tanstack/react-virtual) for large files
+// - Column sorting via @tanstack/react-table
+// - Inline cell editing with live serialization
+// - Mobile-friendly table layout
 ```
 
 ### Terminal Component

package/docs/project-changelog.md

@@ -2,7 +2,121 @@
 
 All notable changes to PPM are documented here. Format follows [Keep a Changelog](https://keepachangelog.com/).
 
-**Current Version:** v0.8.53
+**Current Version:** v0.8.76
+
+---
+
+## [0.8.77] — 2026-04-01 (Planned)
+
+### Added
+- **Deterministic Tab URLs + Backend Workspace Sync**
+  - Tab IDs now deterministic ({type}:{identifier}) instead of random (tab-xxxx)
+  - URL format changed: `/project/{name}/{tabType}/{identifier}` (e.g., `/project/ppm/editor/src/index.ts`)
+  - New `workspace_state` database table (schema v10) persists tab layout per project
+  - GET/PUT `/api/project/:name/workspace` endpoints for server-side persistence
+  - Frontend syncs workspace layout to server with 1.5s debounce
+  - Deep linking from URL auto-creates tabs if missing
+  - Latest-wins conflict resolution: server timestamp > client localStorage
+  - Tab ID patterns standardized: `editor:path`, `chat:provider/sessionId`, `terminal:index`, etc.
+  - Migration from random IDs to deterministic IDs on first load
+
+### Technical Details
+- **Files Created:**
+  - `src/server/routes/workspace.ts` — Workspace GET/PUT endpoints
+  - `src/web/hooks/use-workspace-sync.ts` — Server sync orchestration with debounce
+- **Files Modified:**
+  - `src/services/db.service.ts` — Added workspace_state table (schema v10), helper functions
+  - `src/web/stores/panel-utils.ts` — Deterministic tab ID derivation logic
+  - `src/web/hooks/use-url-sync.ts` — URL parsing/building with new format
+  - `src/web/stores/panel-store.ts` — Load workspace from server on project switch
+  - `src/web/stores/tab-store.ts` — Tab interface updated with metadata
+- **Breaking Changes:** URLs changed; old `/project/{name}/tab/{id}` URLs redirected or ignored (fallback to project root)
+- **Migration:** Client-side: old random tab IDs migrated to deterministic format on first load
+
+### Benefits
+- **Shareable deep links** — URLs point to specific files/chats (e.g., `/project/ppm/editor/src/index.ts`)
+- **Cross-device persistence** — Workspace layout saved on server, restored on any device
+- **URL-driven navigation** — Paste URL to recreate workspace state
+- **Conflict-free sync** — Latest timestamp wins; no manual merge dialogs
+
+---
+
+## [0.8.61] — 2026-03-26 (Beta)
+
+### Added
+- **Multi-Provider Architecture** — Generic AI provider system supporting Claude (SDK-based) and CLI-spawning providers
+  - `AIProvider` interface with optional capability methods (`abortQuery?`, `getMessages?`, `listSessionsByDir?`)
+  - `CliProvider` abstract base class for CLI-spawning providers (Cursor, Codex, Gemini)
+  - `CursorCliProvider` implementation — spawns `cursor-agent` with NDJSON streaming
+  - NDJSON line parser utility for TCP packet boundary handling
+  - Cursor event mapper — normalizes Cursor NDJSON → standard ChatEvent union
+  - Cursor history reader — loads sessions from `~/.cursor/chats/` SQLite DAG
+  - Provider selector UI component — users can choose provider when creating chat
+  - Async provider bootstrap — checks binary availability, registers only if available
+  - Workspace trust auto-retry — detects trust prompts, retries with `--trust` flag
+  - Process lifecycle management — SIGTERM → SIGKILL escalation, orphan cleanup
+
+### Technical Details
+- **Files Created:**
+  - `src/utils/ndjson-line-parser.ts` — NDJSON streaming parser
+  - `src/providers/cli-provider-base.ts` — Abstract CliProvider base class
+  - `src/providers/cursor-cli/cursor-provider.ts` — CursorCliProvider
+  - `src/providers/cursor-cli/cursor-event-mapper.ts` — Event mapping
+  - `src/providers/cursor-cli/cursor-history.ts` — SQLite history reader
+  - `src/web/components/chat/provider-selector.tsx` — Provider selection UI
+  - `tests/unit/ndjson-line-parser.test.ts` — Parser tests
+  - `tests/unit/cursor-event-mapper.test.ts` — Mapper tests
+  - `tests/integration/cursor-provider.test.ts` — Integration tests
+  - `tests/integration/chat-service-multi-provider.test.ts` — Service tests
+- **Files Modified:**
+  - `src/types/chat.ts` — Added optional capability methods to AIProvider, added `system` event type
+  - `src/types/config.ts` — Added `"cli"` type, `cli_command` field to AIProviderConfig
+  - `src/providers/registry.ts` — Added async `bootstrapProviders()` for conditional registration
+  - `src/server/ws/chat.ts` — Removed `as any` casts, use optional chaining for capabilities
+  - `src/services/chat.service.ts` — Use optional methods instead of duck-typing
+  - `src/web/components/chat/session-picker.tsx` — Integrated provider selector
+- **Breaking Changes:** None (backward compatible, all tests passing)
+- **Architecture:** All phases complete (6/6), 479 tests passing
+
+### Benefits
+- Extensible foundation for Codex, Gemini, and future providers (~100-150 lines each)
+- No more `as any` casts for provider methods — type-safe optional capability pattern
+- CLI providers can override session history reading (e.g., Cursor SQLite DAG)
+- Graceful degradation — missing CLI binary doesn't crash, logs info, skips provider
+
+---
+
+## [0.8.55] — 2026-03-26
+
+### Added
+- **Streaming Input Migration** — Persistent AsyncGenerator session model for chat
+  - Provider maintains long-lived streaming input per session (not per message)
+  - Follow-up messages push into existing generator instead of abort-and-replace
+  - Single streaming loop decoupled from WebSocket message handler
+  - Message priority support (`now`/`next`/`later`) for intelligent message ordering
+  - Image attachment support in message sending
+  - Session state persistence across FE disconnections (5-minute cleanup timeout)
+  - Event buffering on reconnect: clients receive buffered turn events after reconnection
+  - Phase transitions: `idle` → `initializing` → `connecting` → `thinking`/`streaming` → `idle`
+
+### Technical Details
+- **Provider Layer** (`src/providers/claude-agent-sdk.ts`): Maintains streaming session per sessionId
+- **WebSocket Handler** (`src/server/ws/chat.ts`):
+  - `runStreamLoop()` executes independently (detached from WS scope)
+  - `activeSessions` map persists session state across FE disconnections
+  - `startSessionConsumer()` pattern replaces per-message `runStreamLoop()` calls
+  - Event buffering (turnEvents array, max 10k events) for reconnect sync
+  - Per-client ping intervals (15s keepalive)
+- **Types** (`src/types/api.ts`, `src/types/chat.ts`):
+  - `ChatWsClientMessage` extended with `priority?: string` and optional `images`
+  - `SessionPhase` = "initializing" | "connecting" | "thinking" | "streaming" | "idle"
+  - `SessionEntry` tracks clients, abort handle, phase, pending approvals, buffered events
+- **Frontend** (`src/web/components/chat/message-input.tsx`): PriorityToggle component (visible during streaming)
+- **Benefits**:
+  - No SDK subprocess restarts between messages (faster context preservation)
+  - Tool approvals integrated into streaming (no query restart)
+  - Message buffering prevents loss on FE reconnection
+  - Cleaner architecture: BE owns Claude connection, FE disconnect ≠ abort
 
 ---
 

package/docs/project-roadmap.md

@@ -1,6 +1,6 @@
 # PPM Project Roadmap
 
-**Last Updated:** March 22, 2026
+**Last Updated:** March 27, 2026
 
 ## Vision
 
@@ -56,19 +56,26 @@ PPM is the **lightest path from phone to code** — a self-hosted, BYOK, multi-d
 
 ### v0.9.0 — "Open Platform" (Q2–Q3 2026)
 
-**Theme:** Multi-provider AI + extension system. Expand user base beyond Claude-only developers.
+**Theme:** Multi-provider AI (Claude + Cursor) + extension system. Ship a focused release, expand providers later.
 
-| Feature | Priority | Description |
-|---------|----------|-------------|
-| **Multi-provider AI** | Critical | Refactor `ProviderInterface` for clean provider abstraction. Tiered support: Tier 1 (full agentic) = Claude Agent SDK; Tier 2 (chat + tools) = Gemini CLI, OpenAI Codex; Tier 3 (chat-only) = any OpenAI-compatible API. Clean base code for future Chinese providers (DeepSeek, Qwen). |
-| **Extension architecture** | High | Dynamic extension loading system. Extensions = npm packages exporting skills + optional UI panels. First extension: extract DB viewer from core. Extension API: register routes, UI panels, sidebar tabs, skills. Config: `"extensions": ["@ppm/ext-database", "@ppm/ext-docker"]`. |
-| **MCP Management** | Medium | UI to add/remove/configure MCP servers. Test connection. Per-project MCP overrides. Store in SQLite. Pass to Agent SDK via `mcpServers`. |
+**Overall progress: ~40%** (1/3 features complete, merge all at 100%)
 
-**Multi-provider — tiered approach:**
+| Feature | Priority | Status | Description |
+|---------|----------|--------|-------------|
+| **Multi-provider AI** | Critical | ✅ Done | ProviderInterface, registry, Cursor CLI, CLI provider base, UI provider/model selector, permission mode selector, system prompt customization, comprehensive tests — all on beta branch. |
+| **Extension architecture** | High | 🔴 0% | Dynamic extension loading. Extensions = npm packages. First extension: extract DB viewer from core. Extension API: register routes, UI panels, sidebar tabs, skills. Config: `"extensions": ["@ppm/ext-database", "@ppm/ext-docker"]`. |
+| **MCP Management** | Medium | 🔴 0% | UI to add/remove/configure MCP servers. Test connection. Per-project MCP overrides. Store in SQLite. Pass to Agent SDK via `mcpServers`. |
+
+**Multi-provider — v0.9 scope (reduced):**
 - Tier 1 (full agentic): Claude Agent SDK — file edit, terminal, git, full autonomy
-- Tier 2 (chat + tools): Provider-specific CLIs (Gemini CLI, Codex) — agentic via their own tool system
-- Tier 3 (chat-only): Any OpenAI-compatible API — conversation only, no tools
-- Provider interface refactor is foundation work — do it clean now, avoid painful refactor later
+- Tier 2 (agentic CLI): Cursor — agentic via its own tool system
+- Provider interface is clean enough to add more providers later without refactor
+
+**Deferred to v0.9.5+:**
+- Gemini CLI (Tier 2)
+- OpenAI Codex (Tier 2)
+- Tier 3 (chat-only): Any OpenAI-compatible API
+- Chinese providers (DeepSeek, Qwen) — v1.0+
 
 **Extension architecture — design principles:**
 - Extensions are npm packages: `ppm ext install @ppm/ext-database`
@@ -79,15 +86,27 @@ PPM is the **lightest path from phone to code** — a self-hosted, BYOK, multi-d
 
 ---
 
-### v0.10.0 — "Intelligence" (Q3 2026)
+### v0.10.0 — "Enhanced Workflow" (Q3 2026)
 
-**Theme:** PPM's own AI layer. Built-in bot + programmable skills.
+**Theme:** Chat UX upgrade + git workflow. High-impact, independent features that ship fast.
 
 | Feature | Priority | Description |
 |---------|----------|-------------|
+| **Chat history graph** | High | Visual branching tree of chat sessions. Fork conversations, navigate history graph. Game-changer for AI chat UX. |
+| **Worktree management** | Medium | UI to create/switch/delete git worktrees. Use different providers on different branches. Integrated with project switcher. |
+
+---
+
+### v0.11.0 — "Intelligence" (Q3–Q4 2026)
+
+**Theme:** Event system + PPM's own AI layer. Hooks → Skills API → Clawbot dependency chain.
+
+| Feature | Priority | Description |
+|---------|----------|-------------|
+| **Hooks system** | High | Event hooks for PPM lifecycle (file save, git commit, chat message, etc.). Foundation for Skills API and deeper extension integration. |
 | **PPM Skills API** | High | Stable internal API for AI to control PPM: file.read/write/search, terminal.run, git.status/commit/diff, db.query, editor.open/goto, project.switch. Skills are the bridge between AI and PPM features. |
 | **Built-in Clawbot** | High | Lightweight AI agent built into PPM using Anthropic Messages API (not Agent SDK). Uses Skills API + MCP tools. Instant response, no external CLI deps. For quick tasks: file search, code explanation, simple refactors. |
-| **Inline SQL** | Medium | Select text in Monaco → run as SQL. Connection picker in editor context menu. Results panel below editor. Leverages existing DB service. |
+| **More providers** | Medium | Gemini CLI (Tier 2), OpenAI Codex (Tier 2), Tier 3 chat-only (any OpenAI-compatible API). Provider interface already clean from v0.9. |
 
 **Built-in Clawbot — why it matters:**
 - Claude Agent SDK spawns subprocess — heavy, slow startup, requires CLI installed
@@ -106,6 +125,7 @@ PPM is the **lightest path from phone to code** — a self-hosted, BYOK, multi-d
 | **Self-hosted PPM Cloud** | High | Docker image of PPM Cloud for enterprise/team. Same codebase, self-hosted config flag. `docker-compose up` and it works. LDAP/SSO integration. |
 | **PPM Marketplace** | High | Publish/install/update extensions. Browse community extensions. Revenue sharing for paid extensions. Clawbot can create extension → test → publish in minutes. |
 | **Stability & hardening** | Critical | Security audit, performance optimization, comprehensive test coverage (>80%), documentation for contributors, CI/CD pipeline. |
+| **Inline SQL** | Medium | Select text in Monaco → run as SQL. Connection picker in editor context menu. Results panel below editor. Leverages existing DB service. |
 
 ---
 
@@ -130,6 +150,7 @@ Features to pick from after v1.0. Will be reviewed and scheduled based on user f
 | **Cross-platform binaries** | Distribution | Compile macOS/Linux/Windows binaries via `bun build --compile`. `npx ppm` without Bun. |
 | **OLED dark mode** | UX | True black background for OLED screens. |
 | **Collaborative editing** | Social | Real-time multi-user file editing with CRDT (yjs/automerge). |
+| **Custom domain** | Cloud | Map custom domain to PPM Cloud tunnel URL. DNS CNAME + SSL via Let's Encrypt or Cloudflare. Access PPM at `code.yourdomain.com`. |
 
 ---
 
@@ -139,9 +160,10 @@ Features to pick from after v1.0. Will be reviewed and scheduled based on user f
 |---------|-------|-------------|--------|
 | **v0.7** | Multi-Account & Mobile | Account management, usage tracking, mobile UX | ✅ Current |
 | **v0.8** | Always On | PPM Cloud, auto-start, AI chat enhancements | Q2 2026 |
-| **v0.9** | Open Platform | Multi-provider AI, extension architecture, MCP | Q2–Q3 2026 |
-| **v0.10** | Intelligence | Skills API, built-in Clawbot, inline SQL | Q3 2026 |
-| **v1.0** | Production Ready | Self-hosted Cloud, Marketplace, stability | Q4 2026 |
+| **v0.9** | Open Platform | Multi-provider (Claude + Cursor), extension architecture, MCP | Q2–Q3 2026 |
+| **v0.10** | Enhanced Workflow | Chat history graph, worktree management | Q3 2026 |
+| **v0.11** | Intelligence | Hooks, Skills API, Clawbot, more providers (Gemini/Codex/Tier 3) | Q3–Q4 2026 |
+| **v1.0** | Production Ready | Self-hosted Cloud, Marketplace, stability, inline SQL | Q4 2026 |
 
 ---
 
@@ -149,7 +171,7 @@ Features to pick from after v1.0. Will be reviewed and scheduled based on user f
 
 1. **Own "phone to code"** — PPM wins on multi-device access. Don't chase Cursor/Windsurf feature parity.
 2. **PPM Cloud stays razor-thin** — Device registry + tunnel URLs only. No code storage. No cloud execution.
-3. **Multi-provider is tiered** — Full agentic = Claude SDK. Other providers get appropriate tier. Clean interface for future providers.
+3. **Multi-provider is tiered** — v0.9: Claude SDK (Tier 1) + Cursor (Tier 2). v0.11: Gemini, Codex, Tier 3. Clean interface for future providers.
 4. **Extensions keep core lightweight** — Features are opt-in. DB viewer, future tools = extensions. Core stays fast.
 5. **Clawbot enables the ecosystem** — Users create extensions with AI, publish to Marketplace. Zero-friction.
 6. **Self-hosted first, always** — Cloud is optional convenience. PPM works 100% offline/local.
@@ -160,7 +182,7 @@ Features to pick from after v1.0. Will be reviewed and scheduled based on user f
 
 | Item | Priority | Notes |
 |------|----------|-------|
-| Refactor ProviderInterface for multi-provider | High | Foundation for v0.9 |
+| ~~Refactor ProviderInterface for multi-provider~~ | ~~High~~ | ✅ Done on beta branch (v0.9.0-beta.5) |
 | Simplify ChatService streaming | Medium | Reduce async generator complexity |
 | Extract WebSocket common logic | Low | DRY for chat/terminal WS |
 | Round-robin cursor bug in AccountSelector | Medium | Positional cursor not advancing correctly |