- Clear session.revert when server sends undefined (fixes inability to send messages after revert)
- Handle message.removed and message.part.removed SSE events to reload messages
- Ensures reverted messages are removed from UI in real-time without requiring app restart
- Add compact command with proper API parameters (providerID, modelID)
- Implement undo command matching TUI behavior:
- Find previous user message before revert point
- Restore reverted message to prompt input
- Filter messages client-side based on session.revert field
- Add init command with proper hex-encoded message ID generation
- Add session.revert field handling:
- Parse revert field from API responses in fetchSessions and createSession
- Update revert field via SSE session.updated events
- Filter messages during render when revert point is set
- Add SSE event handlers for session.compacted and session.error
- Add force reload option to loadMessages for post-compact refresh
- Messages now filter instantly based on revert state without API reload
- Remove sending() state blocking to allow queueing multiple messages
- Add optimistic message creation with temp IDs for immediate UI feedback
- Implement QUEUED badge display matching TUI behavior (accent color, bold)
- Compute queued state: user message ID > last assistant message ID
- Clear prompt input immediately before async send for better UX
- Fix duplicate messages by properly finding and replacing temp messages
- Fix duplicate parts by clearing optimistic parts when real message arrives
- Fix state mutation bugs in message.part.updated handler (immutable updates)
- Fix state mutation bugs in message.updated handler (immutable updates)
Matches TUI implementation for consistent user experience across clients.
- Add double-escape debounce pattern with 1-second timeout
- First escape shows warning, second escape aborts session
- Fix session busy check to handle undefined timeCompleted field
- Add abortSession API call to sessions store
- Show visual feedback in prompt input during debounce
- Remove Escape from filtered keys in prompt-input auto-focus
- Add required 'url' field to file parts in API request
- Use url, mime, filename instead of path field
- Prevent duplicate attachments by checking if path already exists
- Show all files when picker opens with empty query (use space as query)
- Filter git files only when search query provided
- Enter key now works correctly when file list has items
- Create attachment type system with file, text, symbol, and agent sources
- Implement file picker with SDK integration (find.files, file.status)
- Add @ detection in prompt input to trigger file search
- Create attachment chips UI for managing attached files
- Add attachment state management per session
- Update message submission to include attachments
- Implement drag & drop support for adding files
- Show git-modified files first in file picker with +/- indicators
- Filter files as user types after @
- Clear attachments after successful message send
- Implement automatic agent/model selection for new sessions (first agent, prioritize Anthropic default)
- Extract and restore agent/model from last assistant message when resuming sessions
- Add subagent support: child sessions can select subagents, parent sessions cannot
- Display subagent badge in agent dropdown for identification
- Truncate agent descriptions to 50 characters in dropdown
- Improve model search to include full provider/model ID path
- Auto-select input text on focus for easy model search
- Add getDefaultModel() with priority: agent model → Anthropic → first provider
- Implement tool-specific rendering for all 14 tool types (read, edit, write, bash, webfetch, todowrite, task, etc.)
- Each tool shows contextually relevant information (file previews, diffs, command output, todo lists)
- Add metadata-driven content display using preview, diff, output, and todos from tool state
- Implement status-based rendering (pending, running, completed, error) with animations
- Create global state store for expandable items (tool calls and reasoning sections)
- Fix state persistence: expanded tool calls and reasoning sections remain expanded when new messages arrive
- Fix scroll position preservation during live message updates
- Fix reasoning toggle loop by replacing native details element with custom expandable
- Add comprehensive documentation in TOOL_CALL_IMPLEMENTATION.md
- Reduce font sizes for better readability in expanded tool content
- Add proper keying to For loops to prevent component recreation
- Match TUI patterns for tool names, actions, and content formatting
