Qwen Code v0.24.7-nightly.20261003.2c591ecc08

## 变更内容 * 修复(core)：使 Code Mode 文本与延迟工具发现保持一致，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/12990 * 修复(permissions)：遵循已批准的跨目录工具调用，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/12987 * 修复(cli)：补全建议加载时不再吞掉 Enter 键（#13015），作者 @qwen-code-dev-bot，详见 https://github.com/QwenLM/qwen-code/pull/13021 * 修复(cli)：隔离守护进程扩展设置加载与共享进程状态，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13022 * 测试(managed-agent)：记录 Hosted 延迟基线，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12945 * 新增(agents)：添加本地 workspace-agent 协作，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/11206 * 新增(serve)：实现通用 Broker 提供商控制，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12868 * 新增(managed-agent)：启用 G0 公共 Workspace 文件回合，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12955 * 修复(serve)：以 512 KiB 的分块大小上传会话附件，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/12923 * 测试(managed-agent)：固定 M4 关闭锚点和已发布定义（M4 后续工作），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12995 * 修复(cli)：限制重新启动的子进程超出其监管进程生命周期的时长，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13036 * 测试(managed-agent)：控制提供商重试和发布顺序，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13061 * 测试(managed-agent)：在回合认领测试中停止与恢复扫描器竞争，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13032 * 修复(core)：在使用统计 RUM 上传中遵循 NO_PROXY，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13023 * 修复(runtime-broker)：将被拒绝的提供商启动响应为 unknown，而不是 prepared，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13064 * 修复(runtime-broker)：当原始 Runtime 无法响应时保留 UNKNOWN 响应，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13069 * 新增(sdk-java)：添加经过审计的 Hosted Workspace 操作员恢复机制，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/12977 * 修复(workflow)：提前拒绝不受支持的脚本和超大批次，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13012 * 修复(hooks)：向模型传递工具调用前和 ACP 失败上下文，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13006 * 修复(core)：将已传递的通知回合排除在 ACP 回退序号之外，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13029 * 修复(core)：扩展生命周期事件遵循使用统计退出设置，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12789 * 新增(memory)：将 Mem0 与主 CLI 捆绑发布，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12891 * 新增(agents)：为 workspace agent 添加 A2A 访问和共享功能，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12851 * 新增(managed-agent)：请求 Hosted 工具审批（D6a），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13071 * 修复(web-shell)：防止统计信息头部重叠和原始 JSON 输出，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13091 * 修复(managed-agent)：确定任务事件和取消语义，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12998 * 修复(managed-agent)：防止排队中的 Hosted 审批过期，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13098 * 新增(managed-agent)：添加持久化远程 Shell 结果传递，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/12894 * 新增(managed-agent)：实现私有 Hosted MCP Runtime（H1），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12946 * 性能(sdk-java)：直接将 Runtime Broker JSON 序列化为字节，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13108 * 测试(cli)：为结算重试的 waitFor 设置显式超时（#13034），作者 @qwen-code-dev-bot，详见 https://github.com/QwenLM/qwen-code/pull/13058 * 测试(managed-agent)：明确哪些协调器拒绝路径是真实存在的，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13099 * 文档(managed-agent)：在 E2E README 中标明真实工作进程入口，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13093 * 新增(managed-agent)：提供持久化权限 Actions（D6b），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13101 * 测试(managed-agent)：完成 G0 准入后续工作，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13095 * 修复(web-shell)：保留大型上下文使用量快照，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13097 * 修复(managed-agent)：完成租户范围 403 契约，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13117 * 新增(managed-agent)：添加经过验证的 W1a Workspace 冷恢复，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13088 * 新增(web-shell)：添加可折叠的轨迹瀑布图，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13081 * 测试(managed-agent)：固定提供商控制拒绝保护逻辑，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13118 * 修复(managed-agent)：恢复 Hosted MCP 发布重试，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13109 * 修复(managed-agent)：通过有限验证恢复发布过期状态，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13114 * 测试(core)：在不删除测试的情况下压缩 Core 测试套件，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/13007 * 修复(cli)：发布设置时不再出现文件缺失窗口，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13119 * 修复(web-shell)：将内联 chip 注释保留在 chip 的实际范围内，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12992 * 修复(cli)：持久化已计划的 cron 提示，作者 @XIQIXIQIXIQI，详见 https://github.com/QwenLM/qwen-code/pull/8838 * 测试(managed-agent)：覆盖 G0 启动和缓存的拒绝情况，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13116 * 测试(managed-agent)：完成参与者范围失败诊断，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13127 * 新增(managed-agent)：将持久化工具结果投影到 WebShell，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13037 * 测试(core)：在召回扫描延迟门禁中将 GitHub 托管通道视为共享通道，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13094 * 新增(managed-agent)：在私有 ACP 子进程中托管 Managed 会话（M2），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13131 * 新增(managed-agent)：添加托管文件历史记录和撤销功能，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13110 * 修复(runtime-broker)：通过丢失清理续租恢复认领（#13017），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13115 * 测试(integration)：针对提供商延迟消除 monitor 工具调用 E2E 测试的波动（#13001），作者 @qwen-code-dev-bot，详见 https://github.com/QwenLM/qwen-code/pull/13005 * 测试(integration)：在发送 SIGKILL 终止 ACP 子进程前订阅会话事件流（#12925），作者 @qwen-code-dev-bot，详见 https://github.com/QwenLM/qwen-code/pull/12930 * 测试(integration)：在 E2E 测试框架中默认禁用 Managed 自动记忆（#13009），作者 @qwen-code-dev-bot，详见 https://github.com/QwenLM/qwen-code/pull/13013 * 测试(managed-agent)：补齐 H0b 记录测试装置缺口，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13120 * 修复(core,cli)：限制捆绑技能列表的大小；在没有提供商总数时移除 OpenTUI 的分类标题，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13159 * 新增(web-shell)：在 Managed 面板中显示并响应 Hosted 工具审批，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13107 * 修复(core)：在提醒行中完善延迟工具选择规则，并为 tool_call 添加经过审核的 schema 门禁，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13020 * 新增(managed-agent)：实现 Hosted 回合接管和 G1 故障转移 E2E，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13083 * 新增(core)：在 Code Mode 中支持自由格式输入，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/13150 * 修复(core)：使 exec 输出与 Codex 保持一致，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/13153 * 新增(managed-agent)：实现持久化 Hosted Hooks（H2），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13129 * 新增(core)：允许在 Code Mode 中并发调用 Bash，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/13151 * 修复(managed-hooks)：限制 Hook 准入和冷恢复成本，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13136 * 杂项：移除会隐藏 web-shell 源码的无锚点 workspace/ 忽略规则，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13155 * 新增(managed-agent)：保护由 Session 所有的工具输出回收，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13084 * 修复(serve)：准确报告 Hosted Hook 拒绝和冲突，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13137 * 修复(managed-agent)：验证撤销凭据并披露备份限制，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13144 * 修复(ci)：稳定托管浏览器冒烟测试门禁，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13172 * CI(sdk-java)：确保 Flyway 版本唯一性并监控失败的主分支运行，将可信 PR 通道路由至 ECS，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/12965 * 新增(managed-agent)：添加离线 W1b 恢复包，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13138 * 修复(web-shell)：不再让无法读取全局 QWEN.md 的记忆面板替换该文件，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13154 * 新增(managed-agent)：可靠关闭与 Workspace 绑定的会话，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13135 * 修复(managed-agent)：将会话关闭迁移重新编号为 V32，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13223 * 修复(web)：将快速模型固定到所选的提供商端点，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13227 * 新增(core)：从 models.dev 目录解析模型限制和模态，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/11959 * 修复(web-shell)：停止向无法响应的查看者提供 Managed 审批，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13165 * 新增(agents)：添加隔离的远程 Qwen Runtime 主机，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12582 * 新增(managed-agent)：存储不可变的 AgentDefinition 修订版本（Stage D8a），作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13142 * 测试(managed-agent)：在 ManagedEventReplayTest 中禁用 MockMvc 结果打印，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13229 * 修复(web-shell)：在离线重试时保留身份验证片段，作者 @jabrailkhalil，详见 https://github.com/QwenLM/qwen-code/pull/13176 * 修复(core)：允许在 workflow 元声明之前出现开头注释（#12217），作者 @holny，详见 https://github.com/QwenLM/qwen-code/pull/12651 * 修复(cli)：剪贴板粘贴使用 Linux 专用的不可用提示（#12504），作者 @holny，详见 https://github.com/QwenLM/qwen-code/pull/12507 * 新增(managed-agent)：归档并删除已关闭的 Workspace 会话，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13194 * 修复(core)：不再将格式错误的工具调用参数误诊为 max_tokens 截断，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/12982 * 测试(core)：在单元测试级别固定 CodeModeOnly bridge 行为，作者 @waybarrios，详见 https://github.com/QwenLM/qwen-code/pull/12497 * 修复(serve)：仅释放由加载操作所拥有的 Hook Runtime，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13195 * 修复(core)：将失败的无提醒通知回合恢复为 interrupted_prompt，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13126 * 新增(qqbot)：接受传入的图片和视频，作者 @Eric-GoodBoy-Tech，详见 https://github.com/QwenLM/qwen-code/pull/12850 * 修复(acp)：切换模型时保留已配置的 OpenAI 身份验证选项，作者 @tanzhenxin，详见 https://github.com/QwenLM/qwen-code/pull/13152 * 修复(core)：Windows 锁定目录时通过复制交换扩展构件，作者 @niheaven，详见 https://github.com/QwenLM/qwen-code/pull/11889 * 新增(managed-agent)：恢复安全的已退役工具输出收集，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13225 * 修复(memory)：提取过程中保留 Markdown 强调样式，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13240 * 新增(managed-agent)：允许 Workspace 绑定会话的创建者提交、取消和重命名，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13112 * 测试(managed-agent)：添加工具输出收集部署门禁，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13090 * 新增(channels)：添加 IMAP 和 SMTP 电子邮件通道，作者 @destire-mio，详见 https://github.com/QwenLM/qwen-code/pull/12939 * 新增(sdk-java)：添加高置信度 SpotBugs 门禁、CodeQL Java 扫描和 Maven Dependabot，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13216 * 修复(managed-agent)：保留写入器和发布纪元截止时间，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13192 * 新增(managed-agent)：默认启用持久化本地进程和可信重启恢复，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13211 * 修复(web-shell)：跳过损坏的 Managed SSE 帧并合并间隙重新同步，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13206 * 新增(mobile)：添加范围限定的麦克风授权（第 2 阶段），作者 @jabrailkhalil，详见 https://github.com/QwenLM/qwen-code/pull/12127 * 性能(extensions)：在加载选定详情之前先加载列表元数据，作者 @callmeYe，详见 https://github.com/QwenLM/qwen-code/pull/12640 * 测试(feishu)：将媒体分发覆盖测试与外部 API 隔离，作者 @jabrailkhalil，详见 https://github.com/QwenLM/qwen-code/pull/13170 * 新增(managed-agent)：在 Runtime 工作进程中运行 Managed 会话工具（M5a），作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13167 * 修复(managed-agent)：在被动接管时接管 Runtime Session，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13173 * 修复(memory)：保持 MEMORY.md 索引链接目标可解析，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13156 * 修复(cli)：加强设置失败处理和沙箱命令流，作者 @doudouOUC，详见 https://github.com/QwenLM/qwen-code/pull/13140 * 修复(channels)：配置丢失后清理所属工作进程，作者 @XIQIXIQIXIQI，详见 https://github.com/QwenLM/qwen-code/pull/11071 * 测试(web-shell)：固定单次跳过的重新同步边界和停滞阈值，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13277 * 修复(ci)：提高 serve-ab 握手预算，报告已失效的 CodeQL nightly，将可信 PR 通道路由至 ECS，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13270 * 修复(workflow)：区分传输重试等待和停滞，作者 @qqqys，详见 https://github.com/QwenLM/qwen-code/pull/13274 * 修复(core)：根据剩余窗口限制主回合输出截断下限，作者 @yiliang114，详见 https://github.com/QwenLM/qwen-code/pull/13286 * 测试(managed-agent)：修复 managed-agent-server E2E 运行器的真实模型模式，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13263 * 修复(channels)：将 channel-email 固定到 channel-base 0.24.7，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13305 * 修复(ci)：以 cone 目录形式稀疏检出所有者脚本，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13306 * 修复(ci)：通过 REST assignees 端点分配 PR 所有者，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13310 * 修复(serve)：在 Shell 发布器排空和确认丢失期间保持 Hosted Sessions 可用，作者 @wenshao，详见 https://github.com/QwenLM/qwen-code/pull/13304 ## 新贡献者 * @waybarrios 在 https://github.com/QwenLM/qwen-code/pull/12497 中完成了首次贡献 * @niheaven 在 https://github.com/QwenLM/qwen-code/pull/11889 中完成了首次贡献 **完整变更日志**：https://github.com/QwenLM/qwen-code/compare/v0.24.7...v0.24.7-nightly.20261003.2c591ecc08

## What's Changed * fix(core): align Code Mode text with lazy tool discovery by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/12990 * fix(permissions): honor approved cross-directory tool calls by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/12987 * fix(cli): stop swallowing Enter while completion suggestions load (#13015) by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/13021 * fix(cli): isolate daemon extension settings loads from shared process state by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13022 * test(managed-agent): record Hosted latency baseline by @wenshao in https://github.com/QwenLM/qwen-code/pull/12945 * feat(agents): add local workspace-agent collaboration by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/11206 * feat(serve): implement generic Broker provider controls by @wenshao in https://github.com/QwenLM/qwen-code/pull/12868 * feat(managed-agent): enable G0 public Workspace file turns by @wenshao in https://github.com/QwenLM/qwen-code/pull/12955 * fix(serve): upload session attachments in 512 KiB chunks by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/12923 * test(managed-agent): Pin M4 close anchors and the published definition (M4 follow-up) by @wenshao in https://github.com/QwenLM/qwen-code/pull/12995 * fix(cli): Bound how long a relaunched child outlives its supervisor by @wenshao in https://github.com/QwenLM/qwen-code/pull/13036 * test(managed-agent): gate provider retries and release ordering by @wenshao in https://github.com/QwenLM/qwen-code/pull/13061 * test(managed-agent): stop racing the recovery scanner in turn-claim tests by @wenshao in https://github.com/QwenLM/qwen-code/pull/13032 * fix(core): honor NO_PROXY for usage-statistics RUM uploads by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13023 * fix(runtime-broker): answer a refused provider start as unknown instead of prepared by @wenshao in https://github.com/QwenLM/qwen-code/pull/13064 * fix(runtime-broker): keep the UNKNOWN answer when the original Runtime cannot answer by @wenshao in https://github.com/QwenLM/qwen-code/pull/13069 * feat(sdk-java): Add audited Hosted Workspace operator recovery by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/12977 * fix(workflow): reject unsupported scripts and oversized batches early by @qqqys in https://github.com/QwenLM/qwen-code/pull/13012 * fix(hooks): deliver pre-tool and ACP failure context to the model by @qqqys in https://github.com/QwenLM/qwen-code/pull/13006 * fix(core): keep delivered notification turns out of ACP rewind ordinals by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13029 * fix(core): honor usage-statistics opt-out for extension lifecycle events by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12789 * feat(memory): bundle Mem0 with the main CLI by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12891 * feat(agents): add A2A access and sharing for workspace agents by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12851 * feat(managed-agent): ask for Hosted tool approvals (D6a) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13071 * fix(web-shell): prevent stats header overlap and raw JSON output by @qqqys in https://github.com/QwenLM/qwen-code/pull/13091 * fix(managed-agent): Settle task event and cancel semantics by @wenshao in https://github.com/QwenLM/qwen-code/pull/12998 * fix(managed-agent): guard queued Hosted approval expiry by @wenshao in https://github.com/QwenLM/qwen-code/pull/13098 * feat(managed-agent): Add durable remote Shell result delivery by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/12894 * feat(managed-agent): Implement private Hosted MCP runtime (H1) by @wenshao in https://github.com/QwenLM/qwen-code/pull/12946 * perf(sdk-java): Serialize runtime broker JSON directly to bytes by @wenshao in https://github.com/QwenLM/qwen-code/pull/13108 * test(cli): give the settlement retry waitFors an explicit timeout (#13034) by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/13058 * test(managed-agent): state which coordinator refusal paths are real by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13099 * docs(managed-agent): name the real worker entry in the E2E README by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13093 * feat(managed-agent): Serve durable permission Actions (D6b) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13101 * test(managed-agent): close the G0 admission follow-ups by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13095 * fix(web-shell): preserve large context usage snapshots by @qqqys in https://github.com/QwenLM/qwen-code/pull/13097 * fix(managed-agent): Complete tenant scope 403 contract by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13117 * feat(managed-agent): add verified W1a workspace cold restore by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13088 * feat(web-shell): add a collapsible trajectory waterfall by @qqqys in https://github.com/QwenLM/qwen-code/pull/13081 * test(managed-agent): Pin provider control refusal guards by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13118 * fix(managed-agent): Recover Hosted MCP release retries by @wenshao in https://github.com/QwenLM/qwen-code/pull/13109 * fix(managed-agent): Recover publication expiry with bounded verification by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13114 * test(core): compress the Core test suite without removing tests by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/13007 * fix(cli): Publish settings without a missing-file window by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13119 * fix(web-shell): keep inline chip annotations on the chip's real range by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12992 * fix(cli): persist scheduled cron prompts by @XIQIXIQIXIQI in https://github.com/QwenLM/qwen-code/pull/8838 * test(managed-agent): Cover G0 startup and cached refusals by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13116 * test(managed-agent): finish actor-scope failure diagnostics by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13127 * feat(managed-agent): project durable tool results to WebShell by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13037 * test(core): treat GitHub-hosted lanes as shared in the recall scan latency gate by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13094 * feat(managed-agent): Host Managed sessions in a private ACP child (M2) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13131 * feat(managed-agent): Add hosted file history and undo by @wenshao in https://github.com/QwenLM/qwen-code/pull/13110 * fix(runtime-broker): renew the recovery claim through lost cleanup (#13017) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13115 * test(integration): deflake the monitor tool call E2E against provider latency (#13001) by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/13005 * test(integration): subscribe to the session event stream before SIGKILLing the ACP child (#12925) by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/12930 * test(integration): disable managed auto-memory by default in E2E harnesses (#13009) by @qwen-code-dev-bot in https://github.com/QwenLM/qwen-code/pull/13013 * test(managed-agent): Close H0b record fixture gaps by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13120 * fix(core,cli): budget the bundled skill listing; drop OpenTUI's category heading without a provider total by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13159 * feat(web-shell): show and answer Hosted tool approvals in the Managed panel by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13107 * fix(core): deferred-tool selection rules in the reminder line and a reviewed-schema gate for tool_call by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13020 * feat(managed-agent): Hosted Turn takeover and G1 failover E2E by @wenshao in https://github.com/QwenLM/qwen-code/pull/13083 * feat(core): support Freeform input in Code Mode by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/13150 * fix(core): align exec output with Codex by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/13153 * feat(managed-agent): implement durable Hosted Hooks (H2) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13129 * feat(core): allow concurrent Bash calls in Code Mode by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/13151 * fix(managed-hooks): bound Hook admission and cold restore cost by @wenshao in https://github.com/QwenLM/qwen-code/pull/13136 * chore: drop the unanchored workspace/ ignore rule that hides web-shell sources by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13155 * feat(managed-agent): Protect Session-owned tool output retirement by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13084 * fix(serve): Report Hosted Hook denials and conflicts accurately by @wenshao in https://github.com/QwenLM/qwen-code/pull/13137 * fix(managed-agent): Validate undo receipts and disclose backup limits by @wenshao in https://github.com/QwenLM/qwen-code/pull/13144 * fix(ci): stabilize hosted browser smoke gates by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13172 * ci(sdk-java): guard Flyway version uniqueness and watch red main runs by @wenshao in https://github.com/QwenLM/qwen-code/pull/12965 * feat(managed-agent): Add offline W1b recovery bundles by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13138 * fix(web-shell): stop the memory panel replacing a global QWEN.md it could not read by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13154 * feat(managed-agent): reliably close workspace-bound sessions by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13135 * fix(managed-agent): renumber session-close migration to V32 by @wenshao in https://github.com/QwenLM/qwen-code/pull/13223 * fix(web): pin fast model to the selected provider endpoint by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13227 * feat(core): resolve model limits and modalities from a models.dev catalog by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/11959 * fix(web-shell): stop offering a Managed approval the viewer cannot answer by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13165 * feat(agents): add isolated remote Qwen runtime hosts by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12582 * feat(managed-agent): store immutable AgentDefinition revisions (Stage D8a) by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13142 * test(managed-agent): Disable MockMvc result printing in ManagedEventReplayTest by @wenshao in https://github.com/QwenLM/qwen-code/pull/13229 * fix(web-shell): preserve authentication fragment on offline retry by @jabrailkhalil in https://github.com/QwenLM/qwen-code/pull/13176 * fix(core): allow leading comments before workflow meta declaration (#12217) by @holny in https://github.com/QwenLM/qwen-code/pull/12651 * fix(cli): use Linux-specific unavailable message for clipboard paste (#12504) by @holny in https://github.com/QwenLM/qwen-code/pull/12507 * feat(managed-agent): archive and delete closed workspace sessions by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13194 * fix(core): stop misdiagnosing malformed tool-call args as max_tokens truncation by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/12982 * test(core): pin CodeModeOnly bridge behavior at unit level by @waybarrios in https://github.com/QwenLM/qwen-code/pull/12497 * fix(serve): release only load-owned Hook Runtimes by @wenshao in https://github.com/QwenLM/qwen-code/pull/13195 * fix(core): recover a failed reminder-less notification turn as interrupted_prompt by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13126 * feat(qqbot): Accept inbound images and videos by @Eric-GoodBoy-Tech in https://github.com/QwenLM/qwen-code/pull/12850 * fix(acp): preserve configured OpenAI auth choice on model switches by @tanzhenxin in https://github.com/QwenLM/qwen-code/pull/13152 * fix(core): swap extension artifacts by copy when Windows locks the directory by @niheaven in https://github.com/QwenLM/qwen-code/pull/11889 * feat(managed-agent): Restore safe retired tool output collection by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13225 * fix(memory): preserve Markdown emphasis style during extraction by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13240 * feat(managed-agent): let a Workspace-bound Session's creator submit, cancel and rename by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13112 * test(managed-agent): Add tool output collection deployment gates by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13090 * feat(channels): add IMAP and SMTP email channel by @destire-mio in https://github.com/QwenLM/qwen-code/pull/12939 * feat(sdk-java): add SpotBugs high-confidence gate, CodeQL Java scan, Maven dependabot by @wenshao in https://github.com/QwenLM/qwen-code/pull/13216 * fix(managed-agent): Preserve writer and publication epoch deadlines by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13192 * feat(managed-agent): default durable local-process and trusted reboot recovery on by @wenshao in https://github.com/QwenLM/qwen-code/pull/13211 * fix(web-shell): skip corrupt managed SSE frames and merge gap resyncs by @wenshao in https://github.com/QwenLM/qwen-code/pull/13206 * feat(mobile): add scoped microphone consent (Phase 2) by @jabrailkhalil in https://github.com/QwenLM/qwen-code/pull/12127 * perf(extensions): load list metadata before selected details by @callmeYe in https://github.com/QwenLM/qwen-code/pull/12640 * test(feishu): isolate media dispatch coverage from external APIs by @jabrailkhalil in https://github.com/QwenLM/qwen-code/pull/13170 * feat(managed-agent): Run Managed session tools in a Runtime worker (M5a) by @wenshao in https://github.com/QwenLM/qwen-code/pull/13167 * fix(managed-agent): adopt the Runtime Session on a passive takeover by @wenshao in https://github.com/QwenLM/qwen-code/pull/13173 * fix(memory): keep MEMORY.md index link targets resolvable by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13156 * fix(cli): Harden settings failures and sandbox command streams by @doudouOUC in https://github.com/QwenLM/qwen-code/pull/13140 * fix(channels): clean up owned workers after config loss by @XIQIXIQIXIQI in https://github.com/QwenLM/qwen-code/pull/11071 * test(web-shell): pin the single-skip resync boundary and the stall threshold by @wenshao in https://github.com/QwenLM/qwen-code/pull/13277 * fix(ci): raise serve-ab handshake budget, report dead CodeQL nightlies, route trusted PR lanes to ECS by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13270 * fix(workflow): distinguish transport retry waits from stalls by @qqqys in https://github.com/QwenLM/qwen-code/pull/13274 * fix(core): cap the main-turn output clamp floor by the remaining window by @yiliang114 in https://github.com/QwenLM/qwen-code/pull/13286 * test(managed-agent): repair real-model mode of the managed-agent-server E2E runner by @wenshao in https://github.com/QwenLM/qwen-code/pull/13263 * fix(channels): pin channel-email on channel-base 0.24.7 by @wenshao in https://github.com/QwenLM/qwen-code/pull/13305 * fix(ci): sparse-check out the owner scripts as a cone directory by @wenshao in https://github.com/QwenLM/qwen-code/pull/13306 * fix(ci): assign PR owners through the REST assignees endpoint by @wenshao in https://github.com/QwenLM/qwen-code/pull/13310 * fix(serve): keep Hosted Sessions available across Shell publisher drain and lost acknowledgements by @wenshao in https://github.com/QwenLM/qwen-code/pull/13304 ## New Contributors * @waybarrios made their first contribution in https://github.com/QwenLM/qwen-code/pull/12497 * @niheaven made their first contribution in https://github.com/QwenLM/qwen-code/pull/11889 **Full Changelog**: https://github.com/QwenLM/qwen-code/compare/v0.24.7...v0.24.7-nightly.20261003.2c591ecc08

Qwen Code v0.24.7-nightly.20261003.2c591ecc08 预发布