Original release notes
Highlights
- Added support for Agent Plugins v1 to extend agent capabilities. (#8834)
- Enabled native multi-agent workflows with read-only teammates via the /coordinate command. (#8804)
- Improved text selection with word-wise drag on double-click and line-wise extension on triple-click. (#8739)
- Fixed DashScope Qwen 3.8 request failures by preventing conflicting reasoning settings. (#8525)
- Enhanced Web Shell interactivity with persistent chevrons, better hover states, and inline agent metrics. (#8780)
- Added OpenTelemetry session lifecycle events to improve observability of session creation and shutdown. (#8616)
Breaking Changes
No known breaking changes.
Complete Change List
Features
- feat(serve): bound daemon ACP NDJSON buffers (#8911) by @doudouOUC
- feat(extensions): support Agent Plugins v1 (#8834) by @callmeYe
- feat(ui): word-wise drag after double-click, line-wise extension after triple-click (#8739) by @qwen-code-dev-bot
- Adds standard OpenTelemetry session.start and session.end lifecycle events to improve observability of session creation and shutdown. (#8616) by @zjunothing
- Web Shell subagent rows are now more interactive with persistent chevrons, better hover states, and inline display of agent types and metrics. (#8780) by @carffuca
- Session list reads now properly propagate request cancellation to prevent disconnected clients from leaving expensive background scans running. (#8954) by @doudouOUC
- feat(desktop): add Aliyun OSS release mirror (#8976) by @yiliang114
- feat(web-shell): improve compact tool activity (#8973) by @ytahdn
- ACP sessions now use the unified Goal v3 runtime to support create, edit, pause, resume, and clear actions with improved turn scheduling. (#8732) by @qqqys
- The Web Shell sidebar now includes a Channels view to track integration sessions from DingTalk, Feishu, and WeCom alongside standard tasks. (#8457) by @BZ-D
- The /coordinate command now supports native multi-agent workflows with read-only teammates and automated result forwarding to the leader agent. (#8804) by @yiliang114
- The review skill's reverse audit now detects defects in modeled system layers like sandboxes by comparing runtime state semantics against the model. (#8956) by @wenshao
- Terminal window titles now display status symbols like and to indicate task state in multiplexers where color cues are unavailable. (#8970) by @qwen-code-dev-bot
- The /doctor memory command now reports tool result retention stats, including character counts and warnings for results exceeding 30k characters. (#8875) by @ZijianZhang989
- Background task notifications in the web shell are now localizable and display structured metadata within consistent chat-style bubbles. (#8989) by @ytahdn
- Web Shell now supports Qwen 3.8 reasoning controls, allowing users to toggle Thinking mode and select effort levels directly from the model chip. (#8974) by @callmeYe
Bug Fixes
fix(web-shell): Enforce prompt-safe session navigation (#8931) by @doudouOUC fix(desktop): consolidate 0.1.1 regressions (#8896) by @yiliang114 fix(serve): Keep restore request shapes distinct (#8933) by @doudouOUC fix(web-shell): improve ask user question keyboard interactions (#8876) by @carffuca Prevents DashScope Qwen 3.8 requests from failing by ensuring conflicting reasoning_effort and thinking_budget settings are not sent together. (#8525) by @DragonnZhang Fixes workspace path containment checks in tests to correctly handle canonicalized paths on macOS systems. (#8759) by @rbalachandar Ensures the repair agent is warned to rebuild dist/ on all retryable A/B exit paths to prevent trusting stale baseline artifacts. (#8958) by @wenshao Fixes parsing of dotted-minor Claude model aliases and adds token limit support for Opus 5 models. (#8585) by @netbrah Correctly identifies OpenAI SDK APIUserAbortError as a user cancellation to prevent false API error reporting when requests are aborted. (#8399) by @harjothkhara Virtual subagent session IDs now support reserved characters like colons and slashes to fix detail view resolution for certain provider task IDs. (#8717) by @carffuca Closed resource ownership gaps in the daemon ACP transport by validating envelopes earlier and preventing reuse of failed channel generations. (#8947) by @doudouOUC Same-session refresh operations are now transactional to ensure visible session state remains unchanged if a candidate restore fails or times out. (#8939) by @doudouOUC Extended defense against content-only thinking-tag leaks to all OpenAI-compatible providers to prevent unclosed tags from breaking streams. (#8818) by @yiliang114 Updated review body wording to clearly disclose coverage gaps and prevent contradictions when agents cannot certify the entire diff. (#8857) by @yiliang114 fix(web-shell): keep workspace picker suggestions closed (#8844) by @ytahdn fix(desktop): add safe area to macOS app icon (#8987) by @yiliang114 fix(webui): Close same-session refresh race gaps (#8990) by @doudouOUC fix(web-shell): Harden prompt admission ownership (#8955) by @doudouOUC fix(desktop): follow-up review fixes from #8896 (#8951) by <a class="user-mention notranslate" data-hovercard-type=
