release: Memmy v1.1.6

.github/release-notes/v1.1.3.md

@@ -14,3 +14,18 @@
 
 - Memmy Desktop and Agent are version `1.1.3`. The bundled Memory service, viewer, and `memmy-memory` CLI use the independent version `2.1.1`.
 - Memory continues running after Desktop exits unless **Stop Memory when quitting** is enabled. Standalone Memory CLI installations retain their operating-system service management.
+
+## 修复
+
+- 修复 Windows 和 macOS 安装包中的 Memory 启动、重启问题。桌面端直接管理随包提供的 Memory，避免 Windows 计划任务权限错误和 macOS 系统服务重启冲突。
+- 修复退出桌面端后，后台 Memory 无法从查看器重启的问题。桌面端连接断开后，Memory 仍可重新加载配置并重启服务，支持连续重启。
+- 桌面端会在旧 Memory 完成迁移、释放数据库锁后，将自己管理的运行时升级到随包版本；保留兼容的独立安装和较新版本。
+- 修复 Linux CLI 发布时，版本分支与标签同名导致构建取到旧分支的问题。
+- 自定义 BYOK 模型未被内置能力目录收录时，也可接收图片输入；明确标记为仅支持文本的模型仍沿用原有回退方式。
+- 账号模式下的 Memory 摘要改用专用摘要模型，并兼容旧配置迁移。
+- Windows Memory 后台服务改用隐藏启动器并将日志写入文件，避免登录和重启服务时留下命令窗口。任务定义兼容中文路径，结束计划任务后会在短暂清理窗口后停止其 Memory 子进程；安装、启动及复用已有兼容运行时时会更新旧启动器。旧计划任务因权限不足无法更新时，不再阻断桌面端使用或启动 Memory。
+
+## 升级说明
+
+- 桌面端和 Agent 版本为 `1.1.3`；随包提供的 Memory、查看器和 `memmy-memory` CLI 独立升级为 `2.1.1`。
+- 未开启“退出时停止 Memory”时，退出桌面端后 Memory 继续在后台运行。单独安装的 Memory CLI 保留系统服务管理方式。

.github/release-notes/v1.1.5.md

@@ -2,19 +2,31 @@
 
 ## Highlights
 
-- Duplicate attachment drops are ignored without hiding real validation errors.
-- WebUI and API uploads accept more than four attachments and files larger than the former per-file limits.
-- The request body limit for attachment uploads is 256 MB.
-- Includes the documentation, screenshot, and GitHub/Yunxiao synchronization updates already merged into main since v1.1.4.
+- Agent-source capture now keeps complete Codex turns, pairs tool calls with their surrounding turn, and filters the imported L1 history more accurately. New native scan and source-turn storage paths preserve the turn boundary and support large histories safely.
+- Memory can be packaged and run as a standalone executable. Runtime installation now handles packaged assets, native dependencies, version/help reporting, and transformer loading consistently.
+- Added StepFun and Xiaomi MiMo text providers, with catalog validation derived from the provider contract.
+- Expanded attachments so document content is represented by a manifest, image/document buffering is more reliable, and duplicate attachment drops are ignored safely.
+- Added the PDF, DOCX, PPTX, and XLSX document skills and their validation tooling, together with the required notices and platform payload manifests.
+- Improved world-model detail rendering, work-memory processing, model filtering, and desktop update/toolbar behavior.
+- Synced the release branch with the latest `main` changes, including the current Yunxiao GitHub synchronization workflow and documentation cleanup.
 
 ## Upgrade notes
 
 - Memmy Desktop and Agent are version `1.1.5`.
-- The bundled Memory service, viewer, and CLI remain on independent version `2.1.2`.
+- The bundled Memory service, viewer, and CLI use the independent version `2.1.3`.
 - Memory protocol version remains `1`.
 
 ## 修复与更新
 
-- 重复添加的附件会静默忽略，真正的格式或读取错误仍会提示。
-- WebUI 和 API 上传不再限制附件数量及单文件大小，请求体上限调整为 256 MB。
-- 包含 v1.1.4 之后已合入 main 的文档、截图及 GitHub/云效同步更新。
+- Agent 来源采集现在按完整 Codex 对话轮次写入记忆，正确配对工具调用并改进 L1 历史过滤；新增原生扫描和 source-turn 存储路径，可更稳定地处理大规模历史记录。
+- Memory 支持打包为独立可执行程序，统一处理打包资源、原生依赖、版本/帮助信息和 transformers 加载。
+- 新增 StepFun 和 Xiaomi MiMo 文本模型提供商，模型目录校验改为从提供商契约派生。
+- 附件处理改为使用文档清单，改进图片/文档缓冲，并安全忽略重复的附件拖放。
+- 新增 PDF、DOCX、PPTX、XLSX 文档技能及其校验工具、许可说明和平台资源清单。
+- 改进世界模型详情渲染、工作记忆处理、模型过滤以及桌面端更新和工具栏布局。
+- 发布分支同步最新 `main`，包含当前云效 GitHub 同步流程及文档清理。
+
+## 升级说明
+
+- 桌面端和 Agent 版本为 `1.1.5`；随包提供的 Memory、查看器和 CLI 独立升级为 `2.1.3`。
+- Memory 协议版本保持为 `1`。

.github/workflows/computer-use-native.yml

@@ -0,0 +1,82 @@
+name: Computer Use native runtime
+
+on:
+  pull_request:
+    paths:
+      - ".github/workflows/computer-use-native.yml"
+      - "App/memmy-agent/package*.json"
+      - "App/memmy-agent/src/tools/computer-use/**"
+      - "App/shell/desktop/electron-builder*.yml"
+      - "scripts/internal/shared/check-open-computer-use.mjs"
+      - "scripts/internal/linux/**"
+      - "scripts/internal/mac/**"
+      - "tests/ocu-dev-install.test.mjs"
+      - "scripts/internal/win/build-nsis.sh"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  native:
+    strategy:
+      fail-fast: false
+      matrix:
+        runner: [ubuntu-24.04, ubuntu-24.04-arm, macos-15, windows-2025]
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 10
+    defaults:
+      run:
+        shell: bash
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+      - name: Install the pinned native package into an isolated directory
+        run: |
+          version="$(node -p "require('./App/memmy-agent/package.json').dependencies['open-computer-use']")"
+          npm install --prefix "$RUNNER_TEMP/ocu-native" --ignore-scripts --no-audit --no-fund "open-computer-use@$version"
+          node --input-type=module <<'NODE'
+          import fs from 'node:fs';
+          import path from 'node:path';
+          const root = path.join(process.env.RUNNER_TEMP, 'ocu-native/node_modules/open-computer-use');
+          const arch = process.arch === 'x64' ? 'amd64' : process.arch;
+          const relative = process.platform === 'darwin'
+            ? 'dist/Open Computer Use.app/Contents/MacOS/OpenComputerUse'
+            : `dist/${process.platform === 'win32' ? 'windows' : 'linux'}/${arch}/open-computer-use${process.platform === 'win32' ? '.exe' : ''}`;
+          fs.appendFileSync(process.env.GITHUB_ENV, `OCU_BINARY=${path.join(root, relative)}\n`);
+          NODE
+      - name: Prepare an isolated Linux desktop session
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y dbus-x11 xvfb
+          bash scripts/internal/linux/install-computer-use-deps.sh
+          bash scripts/internal/linux/install-computer-use-deps.sh
+      - name: Check Linux MCP and desktop backend
+        if: runner.os == 'Linux'
+        run: dbus-run-session -- xvfb-run -a node scripts/internal/shared/check-open-computer-use.mjs "$OCU_BINARY" --list-apps | tee ocu-native-result.json
+      - name: Check Windows MCP and desktop backend
+        if: runner.os == 'Windows'
+        run: node scripts/internal/shared/check-open-computer-use.mjs "$OCU_BINARY" --list-apps | tee ocu-native-result.json
+      - name: Validate the published macOS bundle and installer
+        if: runner.os == 'macOS'
+        run: |
+          codesign --verify --deep --strict "$RUNNER_TEMP/ocu-native/node_modules/open-computer-use/dist/Open Computer Use.app"
+          lipo "$OCU_BINARY" -verify_arch arm64 x86_64
+          node --test tests/ocu-dev-install.test.mjs
+      - name: Check macOS MCP discovery
+        if: runner.os == 'macOS'
+        run: node scripts/internal/shared/check-open-computer-use.mjs "$OCU_BINARY" | tee ocu-native-result.json
+      - uses: actions/upload-artifact@v4
+        if: always()
+        with:
+          name: ocu-native-${{ matrix.runner }}
+          path: ocu-native-result.json
+          if-no-files-found: warn
+      - name: Record validation boundary
+        if: always()
+        run: echo 'Native checks validate MCP discovery and, on Linux/Windows, desktop enumeration. Signed installers, permission prompts, screenshots and input still require desktop acceptance.' >> "$GITHUB_STEP_SUMMARY"