Release Notes - v0.8.3

🎉 新版本亮点 / Highlights

本次更新修复了五个问题并带来一项体验改进：多 Agent 路由在 sharedMemoryAcrossConversations 配置下的路由错误、发送图片时的异常问题、发送人昵称和群名称未正确传递给 AI 的问题，以及 AI 卡片流式更新（progressive updates）失效的问题；同时优化了消息队列繁忙时的用户体验。
想结合 OpenClaw Gateway 新版本（2026.3.22+）使用，可手动更新为0.8.3-beta.1版本。

This release fixes five issues and brings one UX improvement: a compatibility issue with newer OpenClaw Gateway versions (2026.3.22+), incorrect multi-Agent routing when sharedMemoryAcrossConversations is enabled, an image sending failure, sender nickname and group name not being correctly passed to the AI, and AI card progressive updates not working; plus an improved experience when the message queue is busy.
To use it with the new OpenClaw Gateway version (2026.3.22+), you can manually update to version 0.8.3-beta.1.

🐛 修复 / Fixes

兼容 OpenClaw Gateway 新版本 / Compatible with Newer OpenClaw Gateway Versions
修复在 OpenClaw Gateway 2026.3.22+ 版本下安装插件时报错 ERR_PACKAGE_PATH_NOT_EXPORTED: Package subpath './plugin-sdk/compat' is not defined by "exports" 的问题。根因是 src/runtime.ts 使用了已被新版 SDK 移除的 openclaw/plugin-sdk/compat 子路径，现已改为从 openclaw/plugin-sdk 主入口导入，对所有版本（2026.3.1+）均兼容。
Fixed an error ERR_PACKAGE_PATH_NOT_EXPORTED: Package subpath './plugin-sdk/compat' is not defined by "exports" when installing the plugin under OpenClaw Gateway 2026.3.22+. The root cause was src/runtime.ts importing from the openclaw/plugin-sdk/compat sub-path which was removed in newer SDK versions. Now imports from the openclaw/plugin-sdk main entry, compatible with all versions (2026.3.1+).
AI 卡片流式更新延迟 / AI Card Progressive Updates Delayed
优化了 AI 卡片流式更新的响应速度。改动前，onReplyStart 中 await startStreaming() 串行等待 AI Card 创建完成（约 500ms~1s），期间收到的 partial reply 全部被丢弃，Card 创建好之后才开始流式更新；同时节流间隔 1000ms 过于保守，对于 2 秒内完成的短回复几乎跳过所有中间更新。改动后，AI Card 创建改为 fire-and-forget 模式与 AI 生成并行，onPartialReply 到来时等待 Card 就绪后立即更新，节流间隔调整为 500ms，流式内容能更早、更频繁地呈现给用户。
Improved AI card progressive update responsiveness. Previously, await startStreaming() in onReplyStart waited serially for AI Card creation (~500ms–1s), discarding all partial replies received during that window; additionally, the 1000ms throttle interval was too conservative, skipping nearly all intermediate updates for short replies. After the fix, AI Card creation runs in fire-and-forget mode in parallel with AI generation; onPartialReply waits for the card to be ready and then updates immediately, with the throttle interval reduced to 500ms for earlier and more frequent streaming updates.
消息重复处理（钉钉服务端重发穿透）/ Duplicate Message Processing on DingTalk Server Resend
修复了 AI 处理耗时超过 ~60 秒时，钉钉服务端重发消息导致 Bot 重复处理的问题。根因是去重逻辑仅使用 headers.messageId（WebSocket 协议层投递 ID，每次重发都是新值），未检查 data.msgId（业务层消息 ID，重发时保持不变），导致重发消息穿透去重缓存。修复后引入双层去重：协议层（headers.messageId）拦截同一次投递的重复回调，业务层（data.msgId）拦截服务端重发，两个 ID 同时标记，任意一个命中即可拦截。
Fixed an issue where DingTalk server resent messages (triggered when AI processing exceeds ~60 seconds) caused the Bot to process the same message twice. The root cause was that deduplication only checked headers.messageId (WebSocket protocol-layer delivery ID, which changes on every resend), ignoring data.msgId (business-layer message ID, which stays the same on resend). After the fix, a two-layer deduplication is applied: protocol-layer (headers.messageId) blocks duplicate callbacks from the same delivery, and business-layer (data.msgId) blocks server resends. Both IDs are marked simultaneously so either one can trigger deduplication.
多 Agent 路由与 sharedMemoryAcrossConversations 冲突 / Multi-Agent Routing Conflict with sharedMemoryAcrossConversations
修复了配置 sharedMemoryAcrossConversations: true 时，多群分配不同 Agent 的 bindings 全部路由到同一个 Agent 的问题。根因是路由匹配错误地使用了 sessionPeerId（已被覆盖为 accountId）而非真实的 peer 标识。修复后，路由匹配使用专用的 peerId 字段（不受会话隔离配置影响），session 构建使用 sessionPeerId，两者职责严格分离。
Fixed an issue where all bindings routing different groups to different Agents would resolve to the same Agent when sharedMemoryAcrossConversations: true was configured. The root cause was that routing matched against sessionPeerId (overridden to accountId) instead of the real peer identifier. After the fix, routing uses the dedicated peerId field (unaffected by session isolation config), while session construction uses sessionPeerId, with strict separation of responsibilities.
发送图片失败 / Image Sending Failure
修复了发送图片时出现异常的问题。(#316)
Fixed an issue where sending images would fail with an error. (#316)
发送人昵称与群名称未正确传递给 AI / Sender Nickname and Group Name Not Passed to AI
修复了会话上下文中 SenderName 字段错误传入用户 ID（而非昵称）、GroupSubject 字段错误传入群 ID（而非群名称）的问题。修复后，AI 能正确获取发送人的钉钉昵称和所在群的名称，有助于 AI 更好地理解对话场景。
Fixed an issue where the SenderName field in the session context was incorrectly set to the user ID instead of the display name, and GroupSubject was set to the group ID instead of the group title. After the fix, the AI correctly receives the sender's DingTalk nickname and the group name, enabling better contextual understanding.

✨ 功能与体验改进 / Features & Improvements

消息队列繁忙时的即时排队反馈 / Instant Queue Acknowledgement When Busy
当用户快速连续发送消息、上一条仍在处理中时，新消息现在会立即弹出一条 AI Card 气泡，显示"上一条还没结束，这条我已经记下，稍后按顺序继续处理"等提示文案，同时贴上"思考中"表情，让用户第一时间知道消息已被接收并排队。等轮到该消息处理时，这条气泡会原地更新为最终回复内容，不会额外多发一条消息。
When a user sends messages in quick succession while the previous one is still being processed, the new message now immediately shows an AI Card bubble with a queuing acknowledgement (e.g. "Still working on the last one, I've noted this down and will continue in order"), along with a "thinking" emoji reaction. When it's the message's turn to be processed, the same bubble is updated in place with the final reply — no extra message is sent.

📥 安装升级 / Installation & Upgrade

# 通过 npm 安装最新版本 / Install latest version via npm
openclaw plugins install @dingtalk-real-ai/dingtalk-connector

# 或升级现有版本 / Or upgrade existing version
openclaw plugins update dingtalk-connector

# 通过 Git 安装 / Install via Git
openclaw plugins install https://github.com/DingTalk-Real-AI/dingtalk-openclaw-connector.git

🔗 相关链接 / Related Links

发布日期 / Release Date：2026-03-23
版本号 / Version：v0.8.3
兼容性 / Compatibility：OpenClaw Gateway 0.4.0+