* fix(chat): prevent message send during IME composition
Add isComposing check to ChatInput to prevent messages from being sent
during Japanese/CJK character conversion. This fixes the issue where
pressing Enter to confirm IME composition would immediately send the
message instead of just completing the character conversion.
Changes:
- Add early return in handleKeyDown when IME is composing
- Add explicit isComposing check to Enter key handler
Fixes issue where Enter during IME composition sends incomplete text.
* fix(chat): improve IME composition detection with keyCode 229 fallback
Add isIMECompositionEvent helper function that checks both isComposing
and keyCode === 229, as recommended by MDN. This fixes IME input issues
on WebKit-based browsers (including Tauri WebView on macOS) where
compositionend may fire before keydown, causing isComposing to be false
when Enter key is processed.
Changes:
- Add isIMECompositionEvent helper function with dual detection
- Update handleKeyDown to use the new helper for IME checks
Fixes MacUI desktop app issue where pressing Enter during IME
composition would send the message instead of confirming the input.
* fix(agent-manager): prevent message send during IME composition
Apply the same IME composition fix as ChatInput to AgentManagerEmptyState.
Add isIMECompositionEvent helper and check it in handleKeyDown to prevent
accidental form submission when confirming IME input with Enter key.
* fix(agent-manager): add redundant IME check to Enter key handler
Add explicit isIMECompositionEvent check to the Enter key condition in
handleKeyDown. Although the early return already handles this, this
change ensures the implementation pattern matches ChatInput.tsx exactly
for consistency and safety.
* feat: add support for OpenCode permission.asked event
Implement UI handling for the new permission.asked event from OpenCode's
PermissionNext system, allowing tools configured with "ask" in opencode.json
to prompt user for approval.
Changes:
- Add permission.asked event handler in useEventStream
- Add patterns[] and always[] fields to Permission type
- Display patterns being requested in PermissionCard
- Show what "Always Allow" will auto-approve
This enables the permission asking feature where tools can require user
approval based on opencode.json configuration.
* feat: add button to open subAgent session from task tool
When an agent uses the task tool to create a subagent session,
a new "Open subAgent session" button is now displayed next to
the task tool output. Clicking this button navigates to the
child session in the chat view.
The button shows only when a sessionId is available in the task
metadata and uses the external link icon to indicate it opens
a separate session context.
Update all import statements to use @opencode-ai/sdk/v2
Modify Vite configurations to alias new SDK path
Update API client method signatures for SDK v2 compatibility
* feat: add /undo, /redo, /timeline slash commands and fork button
- Add /undo command to revert to previous user message
- Add /redo command to redo previously undone messages
- Add /timeline command to show conversation history
- Add fork button on user messages (hover) to create new session
- Add TimelineDialog component for navigating conversation history
- Show undo/redo/timeline in autocomplete when session exists
- Fix fork to use SDK session.fork API
- Silent no-op behavior for undo/redo when no messages (matches OpenCode CLI)
* fix: correct fork session endpoint and session switching
- Add hybrid SDK approach using v1 for existing methods and v2 for fork only
- v2 SDK has correct endpoint /session/{sessionID}/fork instead of broken /session/{id}/fork
- Fix forkFromMessage to use setCurrentSession instead of direct set
- Ensures both useSessionStore and useSessionManagementStore are consistent
- Fixes issue where forked sessions appeared empty until sending a new message
* fix: use direct fetch for fork to avoid v2 SDK build errors
- Remove v2 SDK imports and hybrid SDK approach
- Use direct fetch() call to /api/session/{sessionID}/fork endpoint
- Avoids type conflicts between v1 and v2 SDK
- Bypasses broken v1 SDK fork endpoint (wrong path parameter)
- All codebase now uses v1 SDK consistently
* fix: use jsonc-parser for config file parsing
- Replace strip-json-comments + JSON.parse with jsonc-parser
- Handles comments, trailing commas, and unquoted keys
- Matches OpenCode CLI config parsing behavior
- Fixes SyntaxError when reading global opencode.json config
* feat: improve timeline dialog with loading state and search
- Add loading spinner on fork button during operation (~30s on big sessions)
- Refresh session list after fork so new session appears in sidebar immediately
- Add search bar to filter messages by prompt text content
- Show 'No messages found' when search has no results
- Disable fork button while forking to prevent duplicate operations
- Added skill scope helpers and CRUD operations for skills in opencodeConfig.ts.
- Introduced API endpoints for skill management in main.tsx and index.js.
- Enhanced server-side logic to support skill discovery, creation, updating, and deletion.
- Implemented supporting file operations for skills, including reading, writing, and deleting files.
- Updated package.json to use the latest version of @opencode-ai/sdk.
- Improved development server hot module replacement (HMR) by reusing OpenCode processes.
- Introduced queued message mode with batching and idle auto-send, including support for attachments.
- Added a toggle for queue mode in OpenChamber settings with persistence across sessions.
- Updated various components to handle queued messages, including ChatInput and QueuedMessageChips.
- Modified settings payload and desktop settings to include queue mode configuration.
- Enhanced message sending logic to support additional parts for queued messages.
The app now opens to a new chat by default on startup
New chats no longer create a session until you send your first message
Fixed mobile and VSCode sessions handling for better consistency
madebyjun