You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Added per-agent execution-style personality presets and custom styles, with the selected style available from the Workflow quick-actions menu.
Moved approval settings and execution-style controls into the quick-actions menu; automatic context compression is now opt-in for new Workflows.
Workflow and cost analysis:
Preserved user execution requirements, task objectives, and compression handoff context across context compression.
Added cache hit-rate visibility in cost analysis, MCP tool loading, and access to durable user-message history for Agents.
Proxy and chat reliability:
Added support for multiple proxy servers per API key.
Chat requests now retry transient API failures with exponential backoff, and conversations can retain a persistent cleared-context marker.
🪄 Improvements
Workflow experience:
Improved context-compression state and token usage feedback, terminal state handling, message pagination, message and tool-group ordering, and queued-message delivery.
Refined ask_user rendering and response handling; restored input focus after selecting a path and clarified relative file references.
Configuration and interface:
Refined proxy model-selection layout and quick-menu styling.
Japanese is no longer supported as an interface locale; it remains available for chat translation and language detection.
🐞 Bug Fixes
Workflow reliability:
Fixed preservation of ask_user waiting state, compressed context, durable message ordering, tool-group order, and task objectives after compression.
Fixed stale terminal compression indicators, terminal-session startup projections, and user messages queued while a sub-agent is running.
Shell command policy: Prevented awk, sed, and grep scripts or patterns from being incorrectly treated as file-path arguments and blocked.