OhMyAgent Desktop v2.2.0-beta1

中文

• 使用 Tauri 2 重构桌面端,以缩减安装包大小和资源消耗

English

• Rebuilt the desktop app with Tauri 2 to reduce installer size and resource consumption

OhMyAgent Desktop v2.1.0

v2.1.0 (stable) hardens the self-evolution pipeline, upgrades the embedded pi-mono framework, and slims down the desktop installer.

🧠 Self-Harness Hardening

• Fix all 19 code review findings in the self-harness pipeline
• Harden failure detection, auto-apply monitoring, approval policy, and rate limiting
• Robust skill editing with rollback support across all editable surfaces

🔁 Skill Self-Evolution

• Close the feedback loop: skill metrics now feed back into proposal generation
• Proposal generator supports re-promotion, skill refinement, and prompt/tool/config proposals
• Evolved skills are tracked and surfaced in the WebUI

💾 Memory Recall

• Embedding client with retry and backoff
• Safe truncation for context assembly
• Memory injection gating to filter low-relevance memories

⬆️ pi-mono v0.82.1

• Upgrade embedded pi-mono from v0.80.10 to v0.82.1
• Restore fallbackModels propagation in Agent.createLoopConfig

🖥 Desktop Build

• Smaller installer: filter wrong-platform native binaries (jieba, sharp), drop MSI target
• Fix server startup crash in packaged Electron builds

---

v2.1.0 正式版加固自进化管线，升级嵌入式 pi-mono 框架，并进一步缩小桌面安装包。

🧠 Self-Harness 加固

• 修复自进化管线中的全部 19 个 review 发现
• 加固失败检测、自动应用监控、审批策略与频率限制
• 技能编辑支持回滚，覆盖所有可编辑界面

🔁 技能自进化

• 闭环反馈：技能指标现在反馈到提案生成
• 提案生成器支持重新提升、技能优化及 prompt/工具/配置提案
• 进化的技能在 WebUI 中跟踪展示

💾 记忆召回

• 嵌入客户端支持重试与退避
• 上下文组装的安全截断
• 记忆注入门控，过滤低相关度记忆

⬆️ pi-mono v0.82.1

• 嵌入式 pi-mono 从 v0.80.10 升级至 v0.82.1
• 恢复 fallbackModels 在 Agent.createLoopConfig 中的传递

🖥 桌面构建

• 更小的安装包：过滤错误平台的原生二进制（jieba、sharp），移除 MSI 目标
• 修复打包版 Electron 启动时的服务器崩溃

OhMyAgent Desktop v2.1.0-beta1

v2.1.0-beta1 hardens the self-evolution pipeline and closes the skill evolution feedback loop.

🧠 Self-Harness Hardening

• Fix all 19 code review findings in the self-harness pipeline
• Harden failure detection, auto-apply monitoring, approval policy, and rate limiting
• Robust skill editing with rollback support across all editable surfaces

🔁 Skill Self-Evolution

• Close the feedback loop: skill metrics now feed back into proposal generation
• Proposal generator supports re-promotion, skill refinement, and prompt/tool/config proposals
• Evolved skills are tracked and surfaced in the WebUI

💾 Memory Recall

• Embedding client with retry and backoff
• Safe truncation for context assembly
• Memory injection gating to filter low-relevance memories

🧪 Tests

• New harness hardening, integration, skill-evolution, and embedding-client test suites

---

v2.1.0-beta1 加固自进化管线，并闭环技能自进化反馈。

🧠 Self-Harness 加固

• 修复自进化管线中的全部 19 个 review 发现
• 加固失败检测、自动应用监控、审批策略与频率限制
• 技能编辑支持回滚，覆盖所有可编辑界面

🔁 技能自进化

• 闭环反馈：技能指标现在反馈到提案生成
• 提案生成器支持重新提升、技能优化及 prompt/工具/配置提案
• 进化的技能在 WebUI 中跟踪展示

💾 记忆召回

• 嵌入客户端支持重试与退避
• 上下文组装的安全截断
• 记忆注入门控，过滤低相关度记忆

🧪 测试

• 新增 harness 加固、集成、技能进化、嵌入客户端测试套件

OhMyAgent Desktop v2.0.1-beta1

v2.0.1-beta1 upgrades the embedded pi-mono framework from v0.80.10 to v0.82.1 and improves desktop build packaging.

⬆️ pi-mono v0.82.1

• Upgrade embedded pi-mono from v0.80.10 to v0.82.1, spanning ~12 upstream releases
• Restore fallbackModels propagation in Agent.createLoopConfig

🖥 Desktop Build

• Skip wrong-platform @img/sharp-* native binaries in bundle-deps
• Normalize Windows backslash paths for bundle-deps skip patterns
• Remove MSI target, switch to NSIS with zip compression

---

v2.0.1-beta1 升级嵌入式 pi-mono 框架至 v0.82.1，并改进桌面端构建打包。

⬆️ pi-mono v0.82.1

• 嵌入式 pi-mono 从 v0.80.10 升级至 v0.82.1，跨越约 12 个上游版本
• 恢复 fallbackModels 在 Agent.createLoopConfig 中的传递

🖥 桌面构建

• 跳过非当前平台的 @img/sharp-* 原生二进制文件
• 规范化 Windows 反斜杠路径的 bundle-deps 排除模式
• 移除 MSI 构建目标，改用 NSIS + zip 压缩

OhMyAgent Desktop v2.0.0

v2.0.0 is a major release introducing the Self-Harness autonomous optimization engine, interactive approval flows, and significant code quality improvements across the codebase.

🧠 Self-Harness System

OhMyAgent can now detect failures, propose optimizations, and apply improvements autonomously:

• Failure Detector — automatically triggers optimization workflows on agent errors
• LLM Optimizer — generates improvement proposals for skills, prompts, and tool configurations
• Interactive Approval — approval requests delivered via Feishu cards, Telegram inline keyboards, WeChat/QQ text
• Configurable Policies — confidence thresholds, mechanism allowlists, rate limiting, and cooldown controls
• WebUI Dashboard — full control panel with visual approval rules editor

Self-Harness is enabled by default; manage in Settings → Auto Evolution.

🛡️ Code Quality & Architecture

• Unified OhMyAgentError hierarchy with structured error handlers
• All console.* replaced with pino structured logging
• Config hot-reload with two-phase validate+apply pattern
• Tool registry caching, PreparedStatementCache, batch embedding/writing operations
• Configurable concurrency for embedding, LLM, and DB operations

🆕 New Capabilities

• Word Document skill — create, read, and edit docx files with polished templates and style presets
• Agnes Video V2.0 — keyframe and new params support for video generation
• Desktop updater — resume download + system proxy support

🧪 Tests

• 118 new tests added across agent-factory, memory-retriever, and policy-center (2473 tests, 163 test files)

---

v2.0.0 是一个大版本，引入 Self-Harness 自主优化引擎、交互式审批流程以及全面的代码质量提升。

🧠 Self-Harness 系统

OhMyAgent 现在可以自主检测失败、提出优化并应用改进：

• Failure Detector — 自动检测 agent 失败并触发优化流程
• LLM Optimizer — 为 skill、prompt、工具配置生成改进提案
• 交互式审批 — 通过飞书卡片、Telegram 内联键盘、微信/QQ 文本推送审批请求
• 可配置策略 — 置信度阈值、机制白名单、频率限制和冷却控制
• WebUI 控制台 — 完整的设置面板和可视化审批规则编辑器

Self-Harness 默认启用，请在设置 → Auto Evolution 中管理。

🛡️ 代码质量与架构

• 统一 OhMyAgentError 异常体系与结构化错误处理
• 全部 console.* 替换为 pino 结构化日志
• 配置热重载，两阶段校验+应用模式
• 工具注册表缓存、PreparedStatementCache、批量嵌入/写入操作
• 嵌入、LLM、数据库并发数可配置

🆕 新能力

• Word 文档技能 — 创建、读取、编辑 docx 文件，支持精美模板和样式预设
• Agnes Video V2.0 — 视频生成支持关键帧和新参数
• 桌面更新器 — 断点续传 + 系统代理支持

🧪 测试

• 新增 118 个测试（agent-factory、memory-retriever、policy-center），共 2473 测试、163 个测试文件

OhMyAgent Desktop v2.0.0-beta5

v2.0.0-beta5 is a comprehensive code quality release addressing review findings across all priority levels.

🛡️ Robustness

• Unified OhMyAgentError hierarchy and structured error handlers replacing ad-hoc error handling
• Replace all console.* with pino structured logging across 15 files
• Add structured logging to previously silent catch blocks in 9 core files

⚡ Performance

• Tool registry version-based cache and PreparedStatementCache with LRU eviction
• embedBatch / writeBatch bulk operations for memory embedding and writing
• Configurable concurrency for embedding, LLM, and DB operations via AppConfig.performance

🔧 Architecture

• Config hot-reload with two-phase validate+apply pattern (ConfigManager)
• Skill dependency/conflict validation and SkillMetricsService
• Database composite indexes migration for query optimization
• Replace unbounded Map with LRUCache in skill router, audit all Map usage

🧪 Tests

• 118 new tests: agent-factory, memory-retriever, policy-center (2473 tests pass, 163 test files)

🎨 Code Style

• Add .prettierrc, .editorconfig, and format scripts

---

v2.0.0-beta5 是一个全面的代码质量版本，处理了所有优先级的 review 发现。

🛡️ 健壮性

• 统一 OhMyAgentError 异常体系和结构化错误处理器
• 15 个文件中的 console.* 全部替换为 pino 结构化日志
• 9 个核心文件中静默 catch 块增加结构化日志

⚡ 性能

• 工具注册表版本缓存和 PreparedStatementCache（LRU 驱逐）
• 记忆嵌入和写入的批量操作（embedBatch / writeBatch）
• 嵌入/LLM/数据库的并发数可通过 AppConfig.performance 配置

🔧 架构

• 配置热重载，两阶段校验+应用模式（ConfigManager）
• Skill 依赖/冲突校验和 SkillMetricsService
• 数据库复合索引迁移，优化查询性能
• 替换无界 Map 为 LRUCache，审计所有 Map 使用

🧪 测试

• 新增 118 个测试：agent-factory、memory-retriever、policy-center（2473 测试通过，163 个测试文件）

🎨 代码风格

• 添加 .prettierrc、.editorconfig 和格式化脚本

OhMyAgent Desktop v2.0.0-beta4

v2.0.0-beta4 fixes desktop auto-update issues: resume download, system proxy support, and markdown rendering in update toast.

---

v2.0.0-beta4 修复桌面端自动升级问题：支持断点续传、系统代理，以及更新 toast 中的 Markdown 渲染。

OhMyAgent Desktop v2.0.0-beta3

v2.0.0-beta3 brings a new document editing skill, fixes for file attachments and WSL media display, and restores fallback model retry logic.

🆕 Word Document Skill

• New skills/word-document/ with SKILL.md and docx-helper.mjs CLI — edit, read, and create docx files via shell tool
• Auto-installs docxmlater on first use; supports --api, --read, --run modes

🐛 Fixes

• Fallback model retry — restore fallback model loop broken by pi-mono v0.80.10 upgrade; agent now correctly retries with fallback models on primary failure
• Attachment links — fix user-uploaded file links disappearing from agent responses
• WSL media/files — fix webui_send_media images not showing and file downloads failing on WSL, including download URL token refresh
• Heading bold — replace addHeading with bold createParagraph in docx output to ensure correct rendering
• URL overflow — constrain long URLs within chat message bubbles
• Update detection — switch to semantic version comparison with beta awareness

🔧 Maintenance

• Migrate pnpm.onlyBuiltDependencies from package.json to .npmrc for pnpm v11+ compatibility

---

v2.0.0-beta3 带来文档编辑技能，修复文件附件和 WSL 媒体显示问题，并恢复备用模型重试逻辑。

🆕 Word 文档技能

• 新增 skills/word-document/，提供 docx-helper.mjs CLI — 通过 shell 工具编辑、读取、创建 docx 文件
• 首次使用自动安装 docxmlater；支持 --api、--read、--run 模式

🐛 修复

• 备用模型重试 — 恢复被 pi-mono v0.80.10 升级破坏的 fallback 模型循环
• 附件链接 — 修复用户上传的文件链接在 agent 回复中消失的问题
• WSL 媒体/文件 — 修复 webui_send_media 图片不显示和 WSL 下文件下载失败，含下载 URL token 刷新
• 标题粗体 — docx 输出中 addHeading 改为 bold createParagraph，确保粗体正确渲染
• URL 换行 — 约束长 URL 在聊天气泡内换行不溢出
• 更新检测 — 切换到语义化版本比较，支持 beta 版本识别

🔧 维护

• pnpm.onlyBuiltDependencies 从 package.json 迁移到 .npmrc，适配 pnpm v11+

OhMyAgent Desktop v2.0.0-beta2

v2.0.0-beta2 fixes the release notes: Self-Harness is enabled by default (harness.enabled: true), not disabled.

---

v2.0.0-beta2 修正 release notes：Self-Harness 默认启用（harness.enabled: true），而非关闭。

OhMyAgent Desktop v2.0.0-beta

v2.0.0-beta introduces the Interactive Self-Harness system — OhMyAgent can now detect failures, propose optimizations, and apply improvements with user approval. This is the first beta of the 2.x architecture.

🧠 Self-Harness Optimization Engine

• Failure Detector — automatically detects agent failures and triggers optimization workflows
• Harness Optimizer — LLM-powered optimization proposal engine, wired through completeSimple + model chain
• Approval Policy — configurable approval rules (auto-apply by confidence threshold, mechanism allowlist)
• Rate Limiter — cooldown and hourly caps to prevent runaway optimization cycles
• Auto-Apply Monitor — tracks applied improvements and their effectiveness
• Skill Editor — structured skill modification with validation

🎛️ WebUI Settings

• Harness Settings tab — full control over triggers, rate limits, approval rules, and model selection
• Approval Rules Editor — visual rule builder for confidence thresholds and allowed mechanisms
• Agent harness toggle — per-agent enable/disable with inherited defaults

💬 Interactive Approval Flow

• Feishu — interactive card with approve/reject/modify actions
• Telegram — inline keyboard button approval
• WeChat / QQ — text-based approval instructions
• All channels route approval responses back to the initiating conversation

🛠 Additional Improvements

• Config persistence — fixed camelCase/snake_case mismatch that caused harness settings to be lost on restart
• Beta update channel — "Include beta versions" checkbox in Settings → About for early access to pre-releases
• i18n — 32 harness-specific locale keys in zh-CN and en

⚠️ Beta Notice

Self-Harness is disabled by default. Enable it in Settings → Auto Evolution to start using optimization features.

---

v2.0.0-beta 引入 交互式 Self-Harness（自我进化）系统 — OhMyAgent 现在可以检测失败、提出优化方案，并经用户批准后自动改进。这是 2.x 架构的首个测试版。

🧠 Self-Harness 优化引擎

• Failure Detector — 自动检测 Agent 失败并触发优化流程
• Harness Optimizer — LLM 驱动的优化提案引擎，通过 completeSimple + 模型链路实现
• Approval Policy — 可配置的审批规则（按置信度阈值、机制白名单自动批准）
• Rate Limiter — 冷却时间和每小时上限，防止失控循环
• Auto-Apply Monitor — 跟踪已应用的改进及其效果
• Skill Editor — 结构化的 skill 修改与校验

🎛️ WebUI 设置

• Harness 设置面板 — 全面控制触发器、频率限制、审批规则和模型选择
• Approval Rules 编辑器 — 可视化的规则构建器，配置置信度阈值和允许的改进机制
• Agent 开关 — 按 Agent 独立启用/禁用

💬 交互式审批流程

• 飞书 — 交互式卡片，支持批准/拒绝/修改
• Telegram — 内联键盘按钮审批
• 微信 / QQ — 文本指令审批
• 所有渠道的审批响应均路由回发起会话

🛠 其他改进

• 配置持久化 — 修复 camelCase/snake_case 不匹配导致重启后 harness 设置丢失的问题
• Beta 更新通道 — 设置 → 关于中新增「包含测试版」复选框
• i18n — 32 个 harness 相关本地化键值（中英文）

⚠️ 测试版说明

Self-Harness 默认关闭。请在设置 → Auto Evolution 中启用以使用优化功能。