@ww-ai-lab/openclaw 2026.3.23

🎉 @ww-ai-lab/openclaw 2026.3.23

同步上游 openclaw/openclaw v2026.3.13 以来的 789 个提交，保留自定义 qwen/metaso web_search 供应商。

---

⚡ 重要：微信插件 (WeChat Plugin) 兼容性支持

本版本新增了 openclaw/plugin-sdk 主入口对第三方通道插件所需函数的导出，完整兼容 @tencent-weixin/openclaw-weixin 微信插件。

安装微信插件只需一行命令：

npx -y @tencent-weixin/openclaw-weixin-cli@latest install

新增导出的函数：

• buildChannelConfigSchema — 通道配置 schema 构建
• normalizeAccountId — 账号 ID 规范化
• withFileLock — 文件锁
• createTypingCallbacks — 输入状态回调
• resolveDirectDmAuthorizationOutcome — DM 鉴权
• resolveSenderCommandAuthorizationWithRuntime — 命令鉴权
• resolvePreferredOpenClawTmpDir — 临时目录
• stripMarkdown — Markdown 清理

---

上游更新亮点 (openclaw v2026.3.13)

• fix(compaction): 压缩后 token 计数健全性检查
• fix(telegram): 线程化 SSRF 媒体传输策略
• fix(session): 会话重置时保留 lastAccountId/lastThreadId
• fix(agents): 重放时丢弃 Anthropic thinking blocks
• fix(agents): 避免在大小写不敏感挂载上重复注入 memory 文件
• feat(android): 聊天设置 UI 重设计
• feat(ios): 新增 onboarding 欢迎引导页
• fix(ollama): 隐藏原生推理-only 输出
• Docker: 添加 OPENCLAW_TZ 时区支持
• 及更多修复和改进...

本地自定义

• 保留 qwen/metaso web_search 供应商（作为扩展插件架构）
• 包名: @ww-ai-lab/openclaw
• 修复 bundled 插件运行时依赖安装脚本兼容 @ww-ai-lab/openclaw 包名

v2026.3.19

Changes

• Sync upstream openclaw/openclaw main branch (349 commits)
• Migrate metaso/openai-search extensions to new plugin-sdk architecture
• Add qwen -> openai-search provider mapping for backward compatibility

npm

npm install @ww-ai-lab/openclaw@2026.3.19

@ww-ai-lab/openclaw 2026.3.18

同步上游 openclaw/openclaw（14 个提交），保留自定义 qwen/metaso web_search 供应商。

上游变更

• Discord/Slack/Telegram/WhatsApp: 将 action runtime 迁移到 extension
• Mattermost: 规范化插件导入
• Tlon/Nostr: 修复插件 SDK 导入边界
• Plugins: 移除一方遗留消息发现 shim
• Plugin SDK: 添加遗留消息发现辅助函数

本地修复

• 添加 resolveRuntimeWebSearchProviders 函数（适配上游插件架构重构）

@ww-ai-lab/openclaw 2026.3.17

同步上游 openclaw/openclaw（15 个提交），保留自定义 qwen/metaso web_search 供应商。

上游主要变更：

• 预防性上下文溢出检测（agent tool loop）
• Ollama "prompt too long" 上下文溢出错误检测
• 插件 speech provider 注册功能
• provider catalog 重构（移至 extensions）
• 频道 setup helper 重构（Slack/WhatsApp/Signal/iMessage/Telegram/Discord）
• TTS provider 注册表 + ElevenLabs/Microsoft speech 插件
• 新增多个 provider-catalog 文件（BytePlus、HuggingFace、Kilocode 等）

@ww-ai-lab/openclaw 2026.3.16

同步上游 openclaw/openclaw 328 个提交，保留自定义 qwen/metaso web_search 供应商。

主要变化

• 适配上游 web search 插件化架构重构，将 metaso 和 openai-search (qwen) 迁移为独立 plugin extensions
• 新增 extensions/metaso/ 和 extensions/openai-search/ 插件
• 上游新增 feishu structured card actions、openshell sandbox、多项安全优化和 lazy-load 改进
• 上游新增 firecrawl web search provider

@ww-ai-lab/openclaw 2026.3.15-1

Summary

Synced with upstream openclaw/openclaw v2026.3.13 (latest stable), with fork-specific customizations preserved.

What's Changed (upstream openclaw v2026.3.13)

Changes

• Android/chat settings: redesign with grouped device and media sections, refresh Connect and Voice tabs (#44894)
• iOS/onboarding: add first-run welcome pager before gateway setup (#45054)
• Browser/existing-session: add Chrome DevTools MCP attach mode for signed-in live Chrome sessions
• Browser/act automation: add batched actions, selector targeting, and delayed clicks
• Docker/timezone: add OPENCLAW_TZ for timezone override (#34119)
• Dependencies/pi: bump @mariozechner/pi-* to 0.58.0
• Cron/sessions: add sessionTarget: "current" and session:<id> support
• Telegram/message send: add --force-document for document uploads (#45111)
• Feishu/streaming: add reasoning stream support and structured card headers (#29938)
• Gateway/health monitor: configurable stale-event thresholds and restart limits (#42107)
• Android/mobile: system-aware dark theme across all screens (#46249)

Fixes

• Compaction: fix token count sanity check (#28347)
• Session: preserve lastAccountId/lastThreadId on reset (#44773)
• Ollama: hide native reasoning output (#45330)
• Discord: handle gateway metadata fetch failures (#44397)
• Windows: suppress console windows during restart (#44842)
• Slack: add interactive reply directives (#44607)
• Cron: prevent nested lane deadlocks (#45459)
• Control UI: fix insecure auth, chat-new-messages class, dashboard shutdown reasons
• Build: deduplicate plugin-sdk chunks to fix ~2x memory regression (#45426)
• And many more bug fixes (see upstream CHANGELOG for full list)

Unreleased (upstream HEAD)

• Commands/btw: add /btw side questions for quick tool-less answers (#45444)
• Feishu/cards: identity-aware structured card headers and note footers (#29938)
• Gateway/health monitor: per-channel and per-account overrides (#42107)
• Android/dark theme across all screens (#46249)

What's Changed (fork-specific)

• feat(web-search): upgrade qwen provider to generic openai-search provider (OpenAI-compatible search API)
• feat: add provider usage caching to prevent duplicate requests
• fix(llm-task): strip empty tools array for providers that reject it
• fix(update): align package name to prevent openclaw update from installing upstream
• fix(version): support @ww-ai-lab/openclaw package name in version resolver
• fix(test): adapt symlink fallback test for @ww-ai-lab/openclaw
• chore: update README and package.json links for WW-AI-Lab/openclaw fork

Install

npm install -g @ww-ai-lab/openclaw@latest

Full Changelog

v2026.3.13-1...v2026.3.15-1

@ww-ai-lab/openclaw 2026.3.13-1

同步上游 openclaw/openclaw（30 个提交），保留并升级自定义 web_search 供应商。

Changes

• 升级 qwen 供应商为通用 openai-search 供应商，支持任何 OpenAI-compatible Search API
• 新增 openaiSearch 配置块（toolName, enableSearch, searchParam 等选项）
• 保持向后兼容：旧 qwen 配置和 provider 名仍可使用（deprecated alias）
• 更新 secret resolution 支持 OPENAI_SEARCH_API_KEY 环境变量

Upstream sync

• fix: fall back to a startup entry for windows gateway install
• fix(feishu): fail closed on webhook signature checks
• fix(memory): stop forcing Windows qmd cmd shims
• fix: stop false cron payload-kind warnings in doctor
• fix(cron): compare raw value not trimmed in normalizePayloadKind
• fix(cron): avoid false legacy payload kind migrations
• Various test hardening improvements

@ww-ai-lab/openclaw 2026.3.12

同步上游 openclaw/openclaw main 分支（上游最新版本 v2026.3.8 + 后续 61 个提交），保留自定义 qwen/metaso web_search 供应商。

主要上游变更包括：

• fix(discord): enforce users/roles allowlist in reaction ingress
• fix: harden openai websocket replay
• fix(cli): handle scheduled gateway restarts consistently
• fix(whatsapp): trim leading whitespace in direct outbound sends
• fix: harden state dir permissions during onboard
• fix(gateway): enforce browser origin check regardless of proxy headers
• 以及更多安全加固和 bug 修复

@ww-ai-lab/openclaw 2026.3.11-1

Changes

• fix(llm-task): strip empty tools array for providers that reject it (e.g. DashScope/bailian)
  • Some OpenAI-compatible providers return 400 when tools: [] is sent. When disableTools: true, the empty array is now stripped from the request context.
• Updated allowedModels config to include custom/qwen3.5-397b-a17b and custom/qwen3.5-plus

同步上游 openclaw/openclaw，保留自定义 qwen/metaso web_search 供应商。

@ww-ai-lab/openclaw 2026.3.11

同步上游 openclaw/openclaw（106个提交），保留自定义 qwen/metaso web_search 供应商。

主要变更（来自上游）

• 安全加固: gateway 认证、sandbox fs-bridge、system.run 审批等多处安全强化
• Gateway: 修复 conversation reset、本地认证 SecretRef 解析
• 节点安全: fail closed 处理未绑定的 interpreter approvals
• 重构: 统一 sandbox fs bridge mutations、gateway 凭证规划

Fork 自定义功能

• qwen web_search 供应商: 支持阿里云千问搜索（qwen3.5-plus 等模型）
• metaso web_search 供应商: 支持秘塔搜索