Skip to content

A local-first multi-agent app development platform, built on PenguinHarness

release: 0.2.3

Yaowei Zheng4b0f349df5bf
diff --git a/CHANGELOG.md b/CHANGELOG.md
index bc96a43e..dd5be693 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@
One brief line per release. Per-release detail lives in [`changelog/<version>/`](changelog/).
+- **0.2.3** — 2026-08-19. Conversations you can branch, take back and watch: forking a Session from a completed reply, recalling queued steering and follow-up messages into the composer, live per-row Session status in the sidebar, a compaction rework (tool results kept, interrupted compactions discarded, a summary that streams into a collapsed row, and compaction working on a Session resumed after a restart), thinking levels selectable wherever the work happens with a mid-chat switch that states its cost and a new `max` tier from AgentHub 0.4.4, custom-model protocol and vision detection alongside a per-model fast mode, the session row's hover pair back beside a right-click menu, drag-and-drop composer attachments, and in-place hot updates of a running harness. ([details](changelog/0.2.3/))
- **0.2.2** — 2026-08-11. Long-term Memory across sessions (per-agent user and Workspace scopes, index-only injection, a Memory tab), MCP Server support over stdio / Streamable HTTP / SSE with in-app management, Skills/Vault/Schedules behind Memory-style editable prompts and enable switches, dated config kernel versions with a smart no-loss Update beside restore, a hardened core runtime (unlimited default turns, context-window-derived caps, atomic Trace appends), desktop and proxy polish, a web UX batch, and a clean dependency audit. ([details](changelog/0.2.2/))
- **0.2.1** — 2026-08-04. The desktop app: an Electron shell over the unchanged server and Web App (signed-in on open, shared data root, per-root single-instance lock, three-platform installers mirrored to OSS), a landing download page with mirror-aware static links, download source selection for standalone installers and `penguin update`, compaction that survives malformed summaries with retries and costs finally visible, a random seeded admin password with login throttling, skills manageable from agent settings, per-Project new-chat defaults, a Traces page scaled onto a SQLite trace index, and the server's last IO hotspots closed. ([details](changelog/0.2.1/))
- **0.2.0** — 2026-08-03. One sealed installer bundle per target serving online and offline installs alike — now mirrored to Alibaba Cloud OSS with automatic fallback — truncated tool output recovered from the Session scratchpad, chat navigation for long conversations (shell-style ↑ input recall, stacked sticky run headers, a minimap tick rail with hover previews), steering that survives reloads and carries file attachments, a DB-served session list with paged sidebar groups, and the refreshed qianwenai / OpenRouter model catalog defaulting to deepseek-v4-flash. ([details](changelog/0.2.0/))
diff --git a/CHANGELOG.zh.md b/CHANGELOG.zh.md
index e1a8de9c..107237a1 100644
--- a/CHANGELOG.zh.md
+++ b/CHANGELOG.zh.md
@@ -4,6 +4,7 @@
每个版本一行简述。逐版本的细节见 [`changelog/<version>/`](changelog/)。
+- **0.2.3** — 2026-08-19。可以分叉、收回与观察的对话:从已完成的回复 fork 出新 Session,把排队中的转向与追问消息召回输入框重新编辑再发送,侧边栏逐行的实时 Session 状态,一次压缩重构(工具结果先跑完再压缩、被中断的压缩连同半成品摘要一并丢弃、摘要边生成边流入一个默认折叠的行、客户端重启后恢复的 Session 也能压缩),随处可选的思考等级、会话中切换前先说明代价的确认,以及 AgentHub 0.4.4 带来的 `max` 新档位,自定义模型的协议与视觉能力探测并配上逐模型的 fast mode,会话行重新回到悬停位的归档/删除按钮与右键菜单,聊天区拖拽上传附件,以及运行中实例的原地热更新。([细则](changelog/0.2.3/))
- **0.2.2** — 2026-08-11。跨 Session 的长期 Memory(逐 Agent 的用户与 Workspace 两个作用域,只注入索引,配 Memory 标签页),经 stdio / Streamable HTTP / SSE 的 MCP Server 支持并可在应用内管理,Skills/Vault/Schedules 改用 Memory 式的可编辑提示词与启用开关,带日期的配置内核版本以及「恢复」旁边那个不丢改动的智能更新,加固后的 core 运行时(默认轮次无限、由上下文窗口推导的限额、原子的 Trace 追加),桌面端与代理的打磨,一批 Web 体验改进,以及一次干净的依赖审计。([细则](changelog/0.2.2/))
- **0.2.1** — 2026-08-04。桌面应用:在未改动的服务端与 Web App 之上的 Electron 外壳(打开即登录、共享数据根目录、按根目录的单实例锁、三平台安装包镜像到 OSS),带镜像感知静态链接的落地页下载页,独立安装脚本与 `penguin update` 的下载源选择,能在畸形摘要下靠重试存活、且成本终于可见的压缩,随机预置的管理员密码配登录节流,可从 Agent 设置管理的 Skill,逐 Project 的新会话默认值,架在 SQLite trace 索引之上因而可扩展的 Traces 页,以及服务端最后几处 IO 热点的关闭。([细则](changelog/0.2.1/))
- **0.2.0** — 2026-08-03。每个目标一个封装好的安装包,在线与离线安装通用——现已镜像到阿里云 OSS 并带自动回退——被截断的工具输出可从 Session 草稿区恢复,面向长对话的聊天导航(shell 风格的 ↑ 输入召回、层叠的粘性运行标题、带悬停预览的缩略图刻度轨),能在刷新后存活并携带文件附件的转向,由数据库提供、侧边栏分组分页的会话列表,以及刷新后默认 deepseek-v4-flash 的 qianwenai / OpenRouter 模型目录。([细则](changelog/0.2.0/))
diff --git a/changelog/unreleased/2026-08-11-cli-model-remove.md b/changelog/0.2.3/2026-08-11-cli-model-remove.md
similarity index 100%
rename from changelog/unreleased/2026-08-11-cli-model-remove.md
rename to changelog/0.2.3/2026-08-11-cli-model-remove.md
diff --git a/changelog/unreleased/2026-08-11-cli-model-remove.zh.md b/changelog/0.2.3/2026-08-11-cli-model-remove.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-11-cli-model-remove.zh.md
rename to changelog/0.2.3/2026-08-11-cli-model-remove.zh.md
diff --git a/changelog/unreleased/2026-08-11-trace-index-compaction-rotate.md b/changelog/0.2.3/2026-08-11-trace-index-compaction-rotate.md
similarity index 100%
rename from changelog/unreleased/2026-08-11-trace-index-compaction-rotate.md
rename to changelog/0.2.3/2026-08-11-trace-index-compaction-rotate.md
diff --git a/changelog/unreleased/2026-08-11-trace-index-compaction-rotate.zh.md b/changelog/0.2.3/2026-08-11-trace-index-compaction-rotate.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-11-trace-index-compaction-rotate.zh.md
rename to changelog/0.2.3/2026-08-11-trace-index-compaction-rotate.zh.md
diff --git a/changelog/unreleased/2026-08-14-macos-app-icon-size.md b/changelog/0.2.3/2026-08-14-macos-app-icon-size.md
similarity index 100%
rename from changelog/unreleased/2026-08-14-macos-app-icon-size.md
rename to changelog/0.2.3/2026-08-14-macos-app-icon-size.md
diff --git a/changelog/unreleased/2026-08-14-macos-app-icon-size.zh.md b/changelog/0.2.3/2026-08-14-macos-app-icon-size.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-14-macos-app-icon-size.zh.md
rename to changelog/0.2.3/2026-08-14-macos-app-icon-size.zh.md
diff --git a/changelog/unreleased/2026-08-14-model-group-protocol.md b/changelog/0.2.3/2026-08-14-model-group-protocol.md
similarity index 100%
rename from changelog/unreleased/2026-08-14-model-group-protocol.md
rename to changelog/0.2.3/2026-08-14-model-group-protocol.md
diff --git a/changelog/unreleased/2026-08-14-model-group-protocol.zh.md b/changelog/0.2.3/2026-08-14-model-group-protocol.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-14-model-group-protocol.zh.md
rename to changelog/0.2.3/2026-08-14-model-group-protocol.zh.md
diff --git a/changelog/unreleased/2026-08-15-session-fork.md b/changelog/0.2.3/2026-08-15-session-fork.md
similarity index 100%
rename from changelog/unreleased/2026-08-15-session-fork.md
rename to changelog/0.2.3/2026-08-15-session-fork.md
diff --git a/changelog/unreleased/2026-08-15-session-fork.zh.md b/changelog/0.2.3/2026-08-15-session-fork.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-15-session-fork.zh.md
rename to changelog/0.2.3/2026-08-15-session-fork.zh.md
diff --git a/changelog/unreleased/2026-08-16-chat-completion-status.md b/changelog/0.2.3/2026-08-16-chat-completion-status.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-chat-completion-status.md
rename to changelog/0.2.3/2026-08-16-chat-completion-status.md
diff --git a/changelog/unreleased/2026-08-16-chat-completion-status.zh.md b/changelog/0.2.3/2026-08-16-chat-completion-status.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-chat-completion-status.zh.md
rename to changelog/0.2.3/2026-08-16-chat-completion-status.zh.md
diff --git a/changelog/unreleased/2026-08-16-docs-quickstart-and-nav-order.md b/changelog/0.2.3/2026-08-16-docs-quickstart-and-nav-order.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-docs-quickstart-and-nav-order.md
rename to changelog/0.2.3/2026-08-16-docs-quickstart-and-nav-order.md
diff --git a/changelog/unreleased/2026-08-16-docs-quickstart-and-nav-order.zh.md b/changelog/0.2.3/2026-08-16-docs-quickstart-and-nav-order.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-docs-quickstart-and-nav-order.zh.md
rename to changelog/0.2.3/2026-08-16-docs-quickstart-and-nav-order.zh.md
diff --git a/changelog/unreleased/2026-08-16-recall-queued-messages.md b/changelog/0.2.3/2026-08-16-recall-queued-messages.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-recall-queued-messages.md
rename to changelog/0.2.3/2026-08-16-recall-queued-messages.md
diff --git a/changelog/unreleased/2026-08-16-recall-queued-messages.zh.md b/changelog/0.2.3/2026-08-16-recall-queued-messages.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-16-recall-queued-messages.zh.md
rename to changelog/0.2.3/2026-08-16-recall-queued-messages.zh.md
diff --git a/changelog/unreleased/2026-08-18-admin-password-reset.md b/changelog/0.2.3/2026-08-18-admin-password-reset.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-admin-password-reset.md
rename to changelog/0.2.3/2026-08-18-admin-password-reset.md
diff --git a/changelog/unreleased/2026-08-18-admin-password-reset.zh.md b/changelog/0.2.3/2026-08-18-admin-password-reset.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-admin-password-reset.zh.md
rename to changelog/0.2.3/2026-08-18-admin-password-reset.zh.md
diff --git a/changelog/unreleased/2026-08-18-cli-thinking-and-output-collapse.md b/changelog/0.2.3/2026-08-18-cli-thinking-and-output-collapse.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-cli-thinking-and-output-collapse.md
rename to changelog/0.2.3/2026-08-18-cli-thinking-and-output-collapse.md
diff --git a/changelog/unreleased/2026-08-18-cli-thinking-and-output-collapse.zh.md b/changelog/0.2.3/2026-08-18-cli-thinking-and-output-collapse.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-cli-thinking-and-output-collapse.zh.md
rename to changelog/0.2.3/2026-08-18-cli-thinking-and-output-collapse.zh.md
diff --git a/changelog/unreleased/2026-08-18-compaction-continuity-and-streaming.md b/changelog/0.2.3/2026-08-18-compaction-continuity-and-streaming.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-compaction-continuity-and-streaming.md
rename to changelog/0.2.3/2026-08-18-compaction-continuity-and-streaming.md
diff --git a/changelog/unreleased/2026-08-18-compaction-continuity-and-streaming.zh.md b/changelog/0.2.3/2026-08-18-compaction-continuity-and-streaming.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-compaction-continuity-and-streaming.zh.md
rename to changelog/0.2.3/2026-08-18-compaction-continuity-and-streaming.zh.md
diff --git a/changelog/unreleased/2026-08-18-custom-model-protocol-detect.md b/changelog/0.2.3/2026-08-18-custom-model-protocol-detect.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-custom-model-protocol-detect.md
rename to changelog/0.2.3/2026-08-18-custom-model-protocol-detect.md
diff --git a/changelog/unreleased/2026-08-18-custom-model-protocol-detect.zh.md b/changelog/0.2.3/2026-08-18-custom-model-protocol-detect.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-custom-model-protocol-detect.zh.md
rename to changelog/0.2.3/2026-08-18-custom-model-protocol-detect.zh.md
diff --git a/changelog/unreleased/2026-08-18-desktop-dev-instance.md b/changelog/0.2.3/2026-08-18-desktop-dev-instance.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-desktop-dev-instance.md
rename to changelog/0.2.3/2026-08-18-desktop-dev-instance.md
diff --git a/changelog/unreleased/2026-08-18-desktop-dev-instance.zh.md b/changelog/0.2.3/2026-08-18-desktop-dev-instance.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-desktop-dev-instance.zh.md
rename to changelog/0.2.3/2026-08-18-desktop-dev-instance.zh.md
diff --git a/changelog/unreleased/2026-08-18-hot-update.md b/changelog/0.2.3/2026-08-18-hot-update.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-hot-update.md
rename to changelog/0.2.3/2026-08-18-hot-update.md
diff --git a/changelog/unreleased/2026-08-18-hot-update.zh.md b/changelog/0.2.3/2026-08-18-hot-update.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-hot-update.zh.md
rename to changelog/0.2.3/2026-08-18-hot-update.zh.md
diff --git a/changelog/unreleased/2026-08-18-minimax-m3-preset.md b/changelog/0.2.3/2026-08-18-minimax-m3-preset.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-minimax-m3-preset.md
rename to changelog/0.2.3/2026-08-18-minimax-m3-preset.md
diff --git a/changelog/unreleased/2026-08-18-minimax-m3-preset.zh.md b/changelog/0.2.3/2026-08-18-minimax-m3-preset.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-minimax-m3-preset.zh.md
rename to changelog/0.2.3/2026-08-18-minimax-m3-preset.zh.md
diff --git a/changelog/unreleased/2026-08-18-model-catalog-refresh.md b/changelog/0.2.3/2026-08-18-model-catalog-refresh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-model-catalog-refresh.md
rename to changelog/0.2.3/2026-08-18-model-catalog-refresh.md
diff --git a/changelog/unreleased/2026-08-18-model-catalog-refresh.zh.md b/changelog/0.2.3/2026-08-18-model-catalog-refresh.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-model-catalog-refresh.zh.md
rename to changelog/0.2.3/2026-08-18-model-catalog-refresh.zh.md
diff --git a/changelog/unreleased/2026-08-18-model-fast-mode.md b/changelog/0.2.3/2026-08-18-model-fast-mode.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-model-fast-mode.md
rename to changelog/0.2.3/2026-08-18-model-fast-mode.md
diff --git a/changelog/unreleased/2026-08-18-model-fast-mode.zh.md b/changelog/0.2.3/2026-08-18-model-fast-mode.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-model-fast-mode.zh.md
rename to changelog/0.2.3/2026-08-18-model-fast-mode.zh.md
diff --git a/changelog/unreleased/2026-08-18-models-header-icon-actions.md b/changelog/0.2.3/2026-08-18-models-header-icon-actions.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-models-header-icon-actions.md
rename to changelog/0.2.3/2026-08-18-models-header-icon-actions.md
diff --git a/changelog/unreleased/2026-08-18-models-header-icon-actions.zh.md b/changelog/0.2.3/2026-08-18-models-header-icon-actions.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-models-header-icon-actions.zh.md
rename to changelog/0.2.3/2026-08-18-models-header-icon-actions.zh.md
diff --git a/changelog/unreleased/2026-08-18-proxy-socks.md b/changelog/0.2.3/2026-08-18-proxy-socks.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-proxy-socks.md
rename to changelog/0.2.3/2026-08-18-proxy-socks.md
diff --git a/changelog/unreleased/2026-08-18-proxy-socks.zh.md b/changelog/0.2.3/2026-08-18-proxy-socks.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-proxy-socks.zh.md
rename to changelog/0.2.3/2026-08-18-proxy-socks.zh.md
diff --git a/changelog/unreleased/2026-08-18-remote-claude-code-driving.md b/changelog/0.2.3/2026-08-18-remote-claude-code-driving.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-remote-claude-code-driving.md
rename to changelog/0.2.3/2026-08-18-remote-claude-code-driving.md
diff --git a/changelog/unreleased/2026-08-18-remote-claude-code-driving.zh.md b/changelog/0.2.3/2026-08-18-remote-claude-code-driving.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-remote-claude-code-driving.zh.md
rename to changelog/0.2.3/2026-08-18-remote-claude-code-driving.zh.md
diff --git a/changelog/unreleased/2026-08-18-session-details-and-copy-ux.md b/changelog/0.2.3/2026-08-18-session-details-and-copy-ux.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-session-details-and-copy-ux.md
rename to changelog/0.2.3/2026-08-18-session-details-and-copy-ux.md
diff --git a/changelog/unreleased/2026-08-18-session-details-and-copy-ux.zh.md b/changelog/0.2.3/2026-08-18-session-details-and-copy-ux.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-session-details-and-copy-ux.zh.md
rename to changelog/0.2.3/2026-08-18-session-details-and-copy-ux.zh.md
diff --git a/changelog/unreleased/2026-08-18-sidebar-title-scroll-reveal.md b/changelog/0.2.3/2026-08-18-sidebar-title-scroll-reveal.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-sidebar-title-scroll-reveal.md
rename to changelog/0.2.3/2026-08-18-sidebar-title-scroll-reveal.md
diff --git a/changelog/unreleased/2026-08-18-sidebar-title-scroll-reveal.zh.md b/changelog/0.2.3/2026-08-18-sidebar-title-scroll-reveal.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-sidebar-title-scroll-reveal.zh.md
rename to changelog/0.2.3/2026-08-18-sidebar-title-scroll-reveal.zh.md
diff --git a/changelog/unreleased/2026-08-18-subagent-thinking-level.md b/changelog/0.2.3/2026-08-18-subagent-thinking-level.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-subagent-thinking-level.md
rename to changelog/0.2.3/2026-08-18-subagent-thinking-level.md
diff --git a/changelog/unreleased/2026-08-18-subagent-thinking-level.zh.md b/changelog/0.2.3/2026-08-18-subagent-thinking-level.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-subagent-thinking-level.zh.md
rename to changelog/0.2.3/2026-08-18-subagent-thinking-level.zh.md
diff --git a/changelog/unreleased/2026-08-18-thinking-switch-guard.md b/changelog/0.2.3/2026-08-18-thinking-switch-guard.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-thinking-switch-guard.md
rename to changelog/0.2.3/2026-08-18-thinking-switch-guard.md
diff --git a/changelog/unreleased/2026-08-18-thinking-switch-guard.zh.md b/changelog/0.2.3/2026-08-18-thinking-switch-guard.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-thinking-switch-guard.zh.md
rename to changelog/0.2.3/2026-08-18-thinking-switch-guard.zh.md
diff --git a/changelog/unreleased/2026-08-18-web-code-highlight-bundle.md b/changelog/0.2.3/2026-08-18-web-code-highlight-bundle.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-web-code-highlight-bundle.md
rename to changelog/0.2.3/2026-08-18-web-code-highlight-bundle.md
diff --git a/changelog/unreleased/2026-08-18-web-code-highlight-bundle.zh.md b/changelog/0.2.3/2026-08-18-web-code-highlight-bundle.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-web-code-highlight-bundle.zh.md
rename to changelog/0.2.3/2026-08-18-web-code-highlight-bundle.zh.md
diff --git a/changelog/unreleased/2026-08-18-web-drag-drop-upload.md b/changelog/0.2.3/2026-08-18-web-drag-drop-upload.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-web-drag-drop-upload.md
rename to changelog/0.2.3/2026-08-18-web-drag-drop-upload.md
diff --git a/changelog/unreleased/2026-08-18-web-drag-drop-upload.zh.md b/changelog/0.2.3/2026-08-18-web-drag-drop-upload.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-18-web-drag-drop-upload.zh.md
rename to changelog/0.2.3/2026-08-18-web-drag-drop-upload.zh.md
diff --git a/changelog/unreleased/2026-08-19-agenthub-0.4.4.md b/changelog/0.2.3/2026-08-19-agenthub-0.4.4.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-agenthub-0.4.4.md
rename to changelog/0.2.3/2026-08-19-agenthub-0.4.4.md
diff --git a/changelog/unreleased/2026-08-19-agenthub-0.4.4.zh.md b/changelog/0.2.3/2026-08-19-agenthub-0.4.4.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-agenthub-0.4.4.zh.md
rename to changelog/0.2.3/2026-08-19-agenthub-0.4.4.zh.md
diff --git a/changelog/unreleased/2026-08-19-compact-after-restart.md b/changelog/0.2.3/2026-08-19-compact-after-restart.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-compact-after-restart.md
rename to changelog/0.2.3/2026-08-19-compact-after-restart.md
diff --git a/changelog/unreleased/2026-08-19-compact-after-restart.zh.md b/changelog/0.2.3/2026-08-19-compact-after-restart.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-compact-after-restart.zh.md
rename to changelog/0.2.3/2026-08-19-compact-after-restart.zh.md
diff --git a/changelog/unreleased/2026-08-19-compaction-stats-row-order.md b/changelog/0.2.3/2026-08-19-compaction-stats-row-order.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-compaction-stats-row-order.md
rename to changelog/0.2.3/2026-08-19-compaction-stats-row-order.md
diff --git a/changelog/unreleased/2026-08-19-compaction-stats-row-order.zh.md b/changelog/0.2.3/2026-08-19-compaction-stats-row-order.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-compaction-stats-row-order.zh.md
rename to changelog/0.2.3/2026-08-19-compaction-stats-row-order.zh.md
diff --git a/changelog/unreleased/2026-08-19-desktop-change-password.md b/changelog/0.2.3/2026-08-19-desktop-change-password.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-desktop-change-password.md
rename to changelog/0.2.3/2026-08-19-desktop-change-password.md
diff --git a/changelog/unreleased/2026-08-19-desktop-change-password.zh.md b/changelog/0.2.3/2026-08-19-desktop-change-password.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-desktop-change-password.zh.md
rename to changelog/0.2.3/2026-08-19-desktop-change-password.zh.md
diff --git a/changelog/unreleased/2026-08-19-dev-skill-blog-media.md b/changelog/0.2.3/2026-08-19-dev-skill-blog-media.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-dev-skill-blog-media.md
rename to changelog/0.2.3/2026-08-19-dev-skill-blog-media.md
diff --git a/changelog/unreleased/2026-08-19-dev-skill-blog-media.zh.md b/changelog/0.2.3/2026-08-19-dev-skill-blog-media.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-dev-skill-blog-media.zh.md
rename to changelog/0.2.3/2026-08-19-dev-skill-blog-media.zh.md
diff --git a/changelog/unreleased/2026-08-19-gpt-5.6-sol-display-name.md b/changelog/0.2.3/2026-08-19-gpt-5.6-sol-display-name.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-gpt-5.6-sol-display-name.md
rename to changelog/0.2.3/2026-08-19-gpt-5.6-sol-display-name.md
diff --git a/changelog/unreleased/2026-08-19-gpt-5.6-sol-display-name.zh.md b/changelog/0.2.3/2026-08-19-gpt-5.6-sol-display-name.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-gpt-5.6-sol-display-name.zh.md
rename to changelog/0.2.3/2026-08-19-gpt-5.6-sol-display-name.zh.md
diff --git a/changelog/unreleased/2026-08-19-larger-attachments.md b/changelog/0.2.3/2026-08-19-larger-attachments.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-larger-attachments.md
rename to changelog/0.2.3/2026-08-19-larger-attachments.md
diff --git a/changelog/unreleased/2026-08-19-larger-attachments.zh.md b/changelog/0.2.3/2026-08-19-larger-attachments.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-larger-attachments.zh.md
rename to changelog/0.2.3/2026-08-19-larger-attachments.zh.md
diff --git a/changelog/unreleased/2026-08-19-penguin-dev-script-entry.md b/changelog/0.2.3/2026-08-19-penguin-dev-script-entry.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-penguin-dev-script-entry.md
rename to changelog/0.2.3/2026-08-19-penguin-dev-script-entry.md
diff --git a/changelog/unreleased/2026-08-19-penguin-dev-script-entry.zh.md b/changelog/0.2.3/2026-08-19-penguin-dev-script-entry.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-penguin-dev-script-entry.zh.md
rename to changelog/0.2.3/2026-08-19-penguin-dev-script-entry.zh.md
diff --git a/changelog/unreleased/2026-08-19-session-row-actions.md b/changelog/0.2.3/2026-08-19-session-row-actions.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-session-row-actions.md
rename to changelog/0.2.3/2026-08-19-session-row-actions.md
diff --git a/changelog/unreleased/2026-08-19-session-row-actions.zh.md b/changelog/0.2.3/2026-08-19-session-row-actions.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-session-row-actions.zh.md
rename to changelog/0.2.3/2026-08-19-session-row-actions.zh.md
diff --git a/changelog/unreleased/2026-08-19-stop-during-compaction.md b/changelog/0.2.3/2026-08-19-stop-during-compaction.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-stop-during-compaction.md
rename to changelog/0.2.3/2026-08-19-stop-during-compaction.md
diff --git a/changelog/unreleased/2026-08-19-stop-during-compaction.zh.md b/changelog/0.2.3/2026-08-19-stop-during-compaction.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-stop-during-compaction.zh.md
rename to changelog/0.2.3/2026-08-19-stop-during-compaction.zh.md
diff --git a/changelog/unreleased/2026-08-19-thinking-label-and-compaction-notice.md b/changelog/0.2.3/2026-08-19-thinking-label-and-compaction-notice.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-thinking-label-and-compaction-notice.md
rename to changelog/0.2.3/2026-08-19-thinking-label-and-compaction-notice.md
diff --git a/changelog/unreleased/2026-08-19-thinking-label-and-compaction-notice.zh.md b/changelog/0.2.3/2026-08-19-thinking-label-and-compaction-notice.zh.md
similarity index 100%
rename from changelog/unreleased/2026-08-19-thinking-label-and-compaction-notice.zh.md
rename to changelog/0.2.3/2026-08-19-thinking-label-and-compaction-notice.zh.md
diff --git a/changelog/0.2.3/RELEASE.md b/changelog/0.2.3/RELEASE.md
new file mode 100644
index 00000000..edb86a46
--- /dev/null
+++ b/changelog/0.2.3/RELEASE.md
@@ -0,0 +1,63 @@
+PenguinHarness 0.2.3 — conversations you can branch, take back and watch: fork a Session from a completed reply, recall a queued steering or follow-up message into the composer to edit and resend, live per-row Session status in the sidebar, a compaction that streams its summary into a row you can open instead of passing as a silent gap, thinking levels selectable wherever the work happens with a mid-chat switch that states its cost first, and custom model entries that detect their own protocol and vision support.
+
+## Install
+
+**Desktop app**: grab your platform's installer from [penguin.ooo/download](https://penguin.ooo/download) — served from the OSS mirror when it is reachable, GitHub otherwise. The macOS builds are Developer ID signed and notarized, and as of this release the Windows installers are Authenticode-signed too, so neither platform needs a first-launch unblock.
+
+CLI / server (Linux, macOS; bundled Node runtime):
+
+```sh
+curl -fsSL https://penguin.ooo/install.sh | sh
+penguin web
+```
+
+Windows (PowerShell):
+
+```powershell
+irm https://penguin.ooo/install.ps1 | iex
+penguin web
+```
+
+Or via npm (needs Node >= 24):
+
+```sh
+npm install -g @prismshadow/penguin-cli
+```
+
+## Highlights
+
+**Fork a Session from a completed reply.** Completed root assistant replies gained a **Fork** action beside Copy: confirm, and the conversation is duplicated into a new Session up to that reply, keeping the Project, Agent, model, Workspace, approval mode and title while truncating the transcript at the selected completed Task. Forks own independent Trace shards and a recursive scratchpad snapshot, and local image and file marker paths are rewritten to the new Session id, so attachments keep rendering after either Session is deleted. The server rejects forged or unsafe Trace positions and refuses to fork while the source is running or compacting. Discarding a fork got quieter too: deleting the conversation you are looking at used to leave a `session_not_found` record behind on every delete, visible in the Cost Center's error panel, and the Session list now tombstones the ids it deletes so that lookup never fires.
+
+**Recall a queued message back into the composer.** A mid-run message still waiting — an undelivered steering message, or a follow-up queued behind the current run — can be pulled back into the input box, edited and resent. Every queued hint line above the composer gained an icon-only recall control; clicking it withdraws the message server-side and restores its original content in front of whatever is currently typed, with images and file attachments coming back as composer chips and a follow-up restoring the per-turn thinking level it was queued with. New `DELETE` routes for steering and follow-ups back it, `task_state` gained the pending follow-up list that rebuilds the hints after a reload, and every recall re-broadcasts, so the hint disappears in the other tabs too — including the one that sent the message.
+
+**The sidebar says what every Session is doing.** A row now carries an hourglass that turns while its Session is busy — amber for a compaction, gray for a run — and a green dot afterwards for as long as the last reply has not been read; a Session whose reply has been read shows nothing at all, so the only marks left in the list are the ones worth acting on. Status is live for every row rather than only for the conversation on screen: the server publishes each flip on the user-level event stream, so a scheduled task, a subagent or another tab lights up the row it belongs to, and a run that finishes while you are looking elsewhere leaves the unread dot instead of settling silently. Nothing polls, read state is tracked per Project in the browser, and both live states and the dot name themselves in their tooltip and accessible name in both locales.
+
+**A compaction you can watch, and one that runs when it should.** Compaction stopped being a silent gap in the transcript: the summary rides the output stream as it is written, inside a row collapsed by default exactly like a thinking block, which you expand to watch it being generated or to read it afterwards — and a reader joining mid-compaction picks up the prefix generated so far. The row also names what it did, so a step that drops the old context reads "Clear" rather than being announced as a compaction, and a discarded round is flagged as a compaction turn at all now instead of rendering as a bare card. Two defects went with it. A compaction triggered on a turn that ended in tool calls now runs those tools to completion and compacts with their results in their original call order, inventing nothing synthetic. A compaction the user quit out of is closed as a failed compaction the next time the Session loads, its half-written summary discarded rather than reconstructed — so messages sent after a compaction survive a reload and stale "(unknown tool)" cards are gone. And compaction works on a Session resumed after a client restart, where it was refused with "nothing to compact" however long the conversation actually was; availability is now answered from the state the Trace replay recovered, and each of the three refusals got its own error code so the reason arrives in the reader's language. The composer keeps its Stop button through all of it, so a compaction that is taking too long can be interrupted like any other run.
+
+**Thinking level, chosen where the work is — and a switch that says what it costs.** The composer's picker stopped applying a mid-chat change silently: a pick that would move the level opens a confirmation stating in one sentence that switching lowers the provider's prompt-cache hit rate and raises the cost of the next request, and offers three ways out — "Compact, then switch", which runs a compaction and applies the level once it ends; "Switch anyway"; and Cancel. The picked level is now stored on the Session, so it survives a reload and outlives a later Agent-config edit. AgentHub 0.4.4 added a `max` tier above `xhigh`, selectable in the Web pickers, valid for `run_subagent` and accepted everywhere a level is stored or sent, and every tier is now labelled with the value it actually sends. `run_subagent` gained a `thinking_level` argument so the model can run a cheap mechanical subtask lower and hard analysis higher, and the CLI gained `--thinking` on `run` and `chat` plus a `/thinking` command that reports the level the next turn will run at and overrides subsequent ones. Widening that enum changed the default tool entry, so every existing Agent takes a kernel update — automatic wherever the stored entry was not hand-edited.
+
+**Model entries that find out for themselves.** Pointing the harness at your own endpoint stopped being guesswork. Custom and user-defined groups are no longer pinned to one protocol: a **Detect** action in the model dialog probes which of AgentHub's generic protocol clients a base URL serves — `openai-responses`, then `ant-messages`, then `openai-chat`, first hit wins — with minimal invalid requests that bill nothing and need no model id, so even a keyless endpoint can be classified; confirming the dialog while the protocol is still unset detects first, and falls back to `openai-chat` rather than saving an entry that cannot start. A second probe beside the vision switch sends one 1x1 PNG and a one-word prompt and turns the switch on when the model answers — a real billed completion, so it runs only on an explicit press. Model entries also gained a **fast mode**, opting a model's session requests into the provider's faster serving tier at premium pricing (`service_tier: "priority"` on OpenAI-protocol clients, `speed: "fast"` on Anthropic-protocol ones). It is offered only where the resolved client can actually serve it, switching it on asks for confirmation and names what the premium is not reflected in — the entry's recorded prices, and so every cost surface — and a model that rejects the parameter anyway fails the run immediately with a message naming the fix instead of retrying.
+
+**Attachments up to 100MB, with the ceiling in the admin's hands.** The composer's file attachments were capped at 10MB per file and 12MB per message. The defaults are now 100MB and 120MB, and an admin sets both from an Upload limits dialog in the user menu, bounded to 1–200MB and applied without a restart. The number has one home rather than four: the server owns it, `GET /api/me` reports it, and the composer validates against what it was told and quotes that same figure when it refuses a file, so a client can no longer permit what the server will reject. The HTTP body cap is derived from the per-message total instead of sitting at a fixed 20MB, because a draft chat has no Session to upload to and attachments therefore ride the task request as base64. Inline images are the exception and are capped at 20MB: their bytes enter the conversation and the Trace, which is re-read whole on every history page and resume, and providers reject far smaller ones anyway — a file attachment is passed to the model by path, so its size never lands in the context.
+
+## Notable in this release
+
+- **Session rows.** The hover pair — archive and delete, one click each — returned to the row's trailing slot, and the full set (pin, rename, archive, delete) moved onto right-click, press-and-hold and Shift+F10, anchored to the row rather than to a viewport corner. The list-options menu took leading glyphs on all four rows, and a title clipped by the sidebar's width scrolls its hidden tail into view on hover or keyboard focus, holding still for 0.3s first so a pointer crossing the sidebar leaves every row as it found it.
+- **Web App polish.** Files dragged onto the chat area now attach to the composer, and a file dropped anywhere else no longer navigates the tab away from the running app; exited background processes became removable from the session details card; copy buttons confirm with the check icon alone, announced to screen readers through a live region; the Trace file row shows a single-line file name with the full path on a copy button beside it; and the models page keeps every group action reachable as an icon-only control on narrow rows.
+- **CLI.** `penguin config model remove` closes the model CRUD gap, deleting an entry with the credential inlined on it and reporting the resulting default; `penguin server reset-admin-password` recovers a forgotten admin password from the machine that owns the data root, re-arming the whole initial-password machinery; and the chat REPL collapses long tool output to its first and last four lines with a marker naming how many it hid, which `/verbose` and `penguin chat --verbose` switch off. Collapsing is display-only — the model, the Trace and the Web App receive the complete text either way, and `penguin run` never collapses.
+- **Models.** The catalog was refreshed against AgentHub's line-up: Gemini 3.7, GLM-5.3 and the GPT-5.6 generation joined the direct groups along with their OpenRouter counterparts, plus Grok 4.6 and DeepSeek V4 Pro 0813; a new MiniMax group ships a direct `MiniMax-M3` preset; the direct and gateway DeepSeek rows were re-priced from the current official lists; the delisted `inclusionai/ling-3.0-flash:free` row was dropped; OpenRouter's `openai/*` rows moved onto the Responses protocol; and the generic Chat Completions client was renamed `openai-chat`, with bare `openai` still a working alias that is normalized on read; and the direct `gpt-5.6` row is labelled **GPT-5.6 Sol**, naming the variant its id reaches and is billed as. Gateway heartbeats injected into a streaming response no longer fail the stream, ending the endless "reconnecting" through one-api-style proxies and OpenRouter.
+- **Hot update.** A running PenguinHarness, the desktop app included, can be updated in place: `POST /api/hmr/upgrade` replaces the backend platform, the CLI's command implementations and the Web App as one atomic version, with no restart and no reinstall, and live state such as a running terminal process survives the swap. The new platform must boot and the web dist must validate before anything is committed, a restart resumes the last committed version as a unit, and one rollback candidate is kept. The hot APIs require an admin session and are loopback-only unless served over HTTPS.
+- **A lighter code highlighter.** The first code block a conversation renders pulls ~69 KB gzip instead of ~308 KB: Shiki's JavaScript regex engine replaces the Oniguruma WASM engine, and an explicit language list replaces the 332-grammar registry. `dist/` drops from 11 MB across 304 chunks to 3.3 MB across 29. A fenced language outside that list renders as unhighlighted monospace.
+- **Proxy options accept SOCKS.** The admin proxy address takes any proxy URL undici's dispatcher does — `http://`, `https://` and the (experimental) `socks5://` / `socks://` schemes, credentials allowed — gated by constructing a probe agent rather than by a scheme list, so a value undici cannot speak is refused at write time instead of crashing every later startup. The bare `host[:port]` shorthand and the loopback exemption are unchanged.
+- **Traces.** The trace index's mtime gate now stats every known date directory rather than only the newest, so a shard that compaction rotated into an older directory is indexed instead of leaving the conversation stuck on the compaction banner until something else happened to trip the gate.
+- **Desktop.** A dedicated macOS icon master centres the brand artwork in the platform's safe area, matching the optical size of neighbouring Dock and Launchpad apps. The desktop window also stops offering **Change password**, which asked the one session that signs in by token rather than by password to replace a secret it was never shown; a browser session against the same server keeps the entry, and `penguin server reset-admin-password` gives a desktop-created root a password it can name.
+- **Docs and skills.** Installation was folded into Quickstart, which became an overview branching into one page per route (desktop app / CLI and Web App / SDK), the usage guides moved ahead of the core-design chapters, and the renderer gained tabbed code blocks, collapsible callouts and Shiki highlighting behind dynamic imports. The `remote-claude-code` skill moved to v2: user messages are relayed word-for-word, permission prompts and plan approvals are carried back to the user, multi-line messages travel through a tmux buffer, and menu navigation runs one key per call with a capture in between.
+- **Contributor tooling.** An unpackaged desktop run takes a `PenguinHarness-Dev` identity and its own data root, so hacking on the shell no longer collides with an installed release build; and `pnpm penguin` at the repo root runs the CLI again after the entry split left it importing a module that ran nothing; and the development skill now records where a blog post's screenshots live, so the split between this repo and the community one survives the next person writing a post.
+- **Breaking (thinking levels on DeepSeek).** AgentHub 0.4.4 aligned the DeepSeek client with the effort vocabulary DeepSeek actually accepts: `low` now sends `low` where it sent `high`, and `xhigh` now sends `high` where it sent `max`. A DeepSeek Session left on `xhigh` reasons less deeply than it did — pick `max` to keep that depth; one on `low` reasons and costs less, and `high` is the closest match to its old behavior.
+- **Breaking (`/compact` error codes).** Two of the three `POST /compact` refusals moved off the shared `nothing_to_compact` code onto `compaction_not_configured` and `already_compacted`. `nothing_to_compact` keeps its meaning — no completed conversation turn yet — along with its HTTP status and the shape of the error body, so a client that only rendered `error.message` needs no change.
+
+## Requirements
+
+Linux or macOS (x64 / arm64), or Windows 10+ (x64). The desktop app and the CLI installers bundle their own runtime; installing from npm needs Node >= 24. All data stays under `~/.penguin/data`.
+
+Full detail: [changelog/0.2.3/](https://github.com/Prism-Shadow/penguin-harness/tree/main/changelog/0.2.3).
diff --git a/package.json b/package.json
index 22012b76..ff61c618 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
{
"name": "penguin-harness",
- "version": "0.2.2",
+ "version": "0.2.3",
"private": true,
"type": "module",
"description": "PenguinHarness \u2014 TypeScript AI Agent (SDK + CLI).",
diff --git a/packages/cli/package.json b/packages/cli/package.json
index 06a34f8d..e291f7a9 100644
--- a/packages/cli/package.json
+++ b/packages/cli/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-cli",
- "version": "0.2.2",
+ "version": "0.2.3",
"type": "module",
"description": "PenguinHarness CLI: interactive REPL and single-task runner over @prismshadow/penguin-core.",
"license": "Apache-2.0",
diff --git a/packages/core/package.json b/packages/core/package.json
index 6122d86e..442ac68c 100644
--- a/packages/core/package.json
+++ b/packages/core/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-core",
- "version": "0.2.2",
+ "version": "0.2.3",
"type": "module",
"description": "PenguinHarness core SDK: context_engine, OmniMessage protocol, LLM/Environment interfaces.",
"license": "Apache-2.0",
diff --git a/packages/core/src/index.ts b/packages/core/src/index.ts
index a9d70286..6cf4ff5a 100644
--- a/packages/core/src/index.ts
+++ b/packages/core/src/index.ts
@@ -59,7 +59,7 @@ export { Agent, createAgent } from "./agent.js";
export type { CreateAgentOptions, CreateSessionOptions, ResumeSessionOptions } from "./agent.js";
/** SDK version number. */
-export const VERSION = "0.2.2";
+export const VERSION = "0.2.3";
/** Release build date (UTC yyyy-mm-dd), stamped by the release workflow next to VERSION; null in a dev/source build. */
export const BUILD_DATE: string | null = null;
// Version-string helpers (shared by the CLI's `penguin update` and the server's update check).
diff --git a/packages/desktop/package.json b/packages/desktop/package.json
index cbdaf504..e313c1fe 100644
--- a/packages/desktop/package.json
+++ b/packages/desktop/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-desktop",
- "version": "0.2.2",
+ "version": "0.2.3",
"private": true,
"type": "module",
"description": "PenguinHarness desktop app: Electron shell running @prismshadow/penguin-server as a utilityProcess, window on http://localhost (same-origin HTTP/SSE, no private IPC).",
diff --git a/packages/docs/package.json b/packages/docs/package.json
index dc5ace9b..42d33b0d 100644
--- a/packages/docs/package.json
+++ b/packages/docs/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-docs",
- "version": "0.2.2",
+ "version": "0.2.3",
"private": true,
"type": "module",
"description": "PenguinHarness documentation site (React + Vite + Tailwind CSS): bilingual zh/en Markdown pages with light/dark themes and per-page Copy Markdown, deployed to GitHub Pages under /docs/ next to the landing page.",
diff --git a/packages/landing/content/blog/penguinharness-0-2-3.en.md b/packages/landing/content/blog/penguinharness-0-2-3.en.md
new file mode 100644
index 00000000..e7aa421f
--- /dev/null
+++ b/packages/landing/content/blog/penguinharness-0-2-3.en.md
@@ -0,0 +1,102 @@
+---
+title: "PenguinHarness 0.2.3: a better experience, and an agent you can watch work"
+date: 2026-08-19
+category: news
+excerpt: This release does two things — it makes what the agent is doing visible, and it files down the parts that were awkward to use. The sidebar now shows which conversation is running, context compaction went from a silent gap to a record you can open and read, any completed reply can fork into a new Session, and a queued message can be pulled back into the composer and edited before it goes out.
+---
+
+PenguinHarness 0.2.3 is out. It does two things. It makes what the agent is doing visible, and it files down the parts that were awkward to use.
+
+## The sidebar shows which conversation is running
+
+Until now exactly one row in the sidebar was telling the truth, and that was the conversation you had open. Every other row showed whatever the last list fetch returned. Leave a running conversation and it kept claiming to run; start one in the background and its row never changed at all.
+
+Session status now rides the user-level event stream, so every row is live. A turning hourglass means running. A green dot means it finished and you have not looked at it since. Nothing at all means you have.
+
+![Three sidebar rows showing an unread green dot, no marker, and a running hourglass](/blog-assets/penguinharness-0-2-3-sidebar-status-en.png)
+
+There was a trap hiding under that. Whether a Session has ever run is a field on the server's list row, and the event only updated the status, not that field. So a Session running its very first task went blank the moment the hourglass stopped. The field now travels with the status, and a first run settles onto the dot like any other.
+
+## Compaction is no longer a gap
+
+Context compaction used to be a blank stretch in the conversation. You knew it had happened. Now it is a row you can open, collapsed by default, the same way thinking is collapsed inside a task group.
+
+![A settled compaction row, collapsed](/blog-assets/penguinharness-0-2-3-compaction-collapsed-en.png)
+
+The summary streams in as it is written, so you can expand it and read exactly what the agent decided to keep.
+
+![The expanded compaction summary, with task, confirmed findings, next steps and caveats](/blog-assets/penguinharness-0-2-3-compaction-expanded-en.png)
+
+The timing changed too. When compaction triggers after the model has produced a tool call, it now waits for the tools to finish and compacts with their results in hand, instead of folding an unfinished exchange into the request. Quit the client mid-compaction and that round is marked failed, with its half-written content discarded on load. A round that clears the context rather than summarizing it no longer calls itself compaction; it says Clear.
+
+One recurring complaint is fixed as well. After restarting the client, a conversation full of history would still refuse to compact, reporting that there was nothing to compact. The engine object is built lazily, and while resumption did replay the turn count into the initial state, that value only takes effect in the engine's constructor, which had not run yet. The server's three different 409 reasons also became three distinct error codes, so a Chinese UI now shows a Chinese message instead of falling back to the English prose.
+
+## Any reply can fork into a new Session
+
+When a conversation goes sideways you used to have two options, live with it or start over. Every completed reply now carries a fork control, sitting next to copy.
+
+![The reply action row, with fork beside copy](/blog-assets/penguinharness-0-2-3-session-fork-actions-en.png)
+
+It asks first. On confirm it copies the conversation up to that reply into a new Session and leaves the original alone.
+
+![The confirmation dialog shown before forking](/blog-assets/penguinharness-0-2-3-session-fork-confirm-en.png)
+
+## A queued message can be pulled back
+
+Steering typed mid-run, and follow-ups waiting their turn, used to be gone the moment you sent them. Each now carries a curved-back arrow. Click it and the message returns to the composer, where you can edit it and send it again.
+
+Recall and the engine's own dequeue are mutually exclusive, so a message is never both delivered and withdrawn. If the engine takes it in the same instant you click, you get a clear message saying so rather than a silent disappearance or a double send.
+
+## Thinking levels
+
+AgentHub 0.4.4 added a `max` tier, so the ladder is now low, medium, high, xhigh, max.
+
+![The thinking-level dropdown](/blog-assets/penguinharness-0-2-3-thinking-level-menu-en.png)
+
+Changing the level mid-conversation now asks first. Some providers implement thinking levels by injecting a different prompt prefix at the very front of the request, so switching part-way through invalidates the prefix cache over the whole history and the next request re-bills all of it at the uncached rate. The dialog offers three ways out: compact first and then switch, switch anyway, or cancel.
+
+![The confirmation shown when switching thinking level mid-conversation](/blog-assets/penguinharness-0-2-3-thinking-level-switch-guard-en.png)
+
+The level you pick is stored on the Session, so it survives a refresh and shows up in a second tab.
+
+## Adding a custom model no longer means guessing the protocol
+
+Since 0.4.2 AgentHub has carried three generic protocol clients — OpenAI Responses, Anthropic Messages, and OpenAI Chat Completions. You no longer have to work out which one your endpoint speaks. The detect action probes them in that order and takes the first that answers in its shape, then fills the protocol picker in for you. With no key typed it falls back to the environment variable, and when nothing matches you get one short message telling you to check the API key and the base URL.
+
+![The detect action at the top right of the base URL field, and the in-field picker listing the three protocols with the path each appends](/blog-assets/penguinharness-0-2-3-model-add-protocol-en.png)
+
+Vision support can be detected the same way. It sends a 1×1 image; a normal answer turns the setting on, an explicit image-related rejection turns it off, and anything else leaves your setting alone. That probe costs a real API call, so it only runs when you ask for it and never on save.
+
+Models that can serve fast mode gained a toggle, default off. Models that cannot do not show the toggle at all, so you cannot arm a switch that would fail every turn. Turning it on says plainly that fast mode bills at premium rates.
+
+## Also in this release
+
+The sidebar row's hover actions went back to archive and delete, with the full set moved to the right-click menu. Attachments and images can be dragged onto the chat area to upload, and a file attachment can now be 100MB rather than 10MB, with an admin able to change that from the user menu. Inline images keep a separate 20MB cap, because their bytes enter the conversation and the Trace, which is re-read whole on every history page; a file attachment is opened by path, so its size never reaches the context. The CLI gained `/thinking` and `--thinking`, and long tool output now collapses to its first and last lines with `/verbose` to see all of it — the model and the Trace always received the full text either way.
+
+The model catalog was refreshed: Gemini 3.7, GLM-5.3 and the GPT-5.6 family, plus their OpenRouter counterparts, with the nine `openai/*` rows pinned to the Responses protocol. DeepSeek pricing follows the current official rates, and the delisted `ling-3.0-flash` row is gone.
+
+One fix underneath is worth calling out. Gateways such as one-api and OpenRouter inject heartbeat events into a streaming response during a long generation. Those events used to reach the unknown-event guard and kill the stream, the harness classified that as a retryable failure, and the retry hit the next heartbeat — a long generation through such a gateway reconnected until it gave up. Every streaming client in AgentHub 0.4.3 and later skips them.
+
+## From running to legible
+
+These changes land on the same point. While an agent works you should be able to see what it is doing, what it chose to remember, and what the next step will cost. Session status, the compaction record, and the warning before a level switch all move something out of the logs and onto the screen.
+
+## Install and upgrade
+
+Desktop builds are on the site for every platform. Windows builds are signed as of this release; macOS has been signed and notarized since 0.2.2.
+
+Linux and macOS install from the command line.
+
+```sh
+curl -fsSL https://penguin.ooo/install.sh | sh
+penguin web
+```
+
+Windows uses PowerShell.
+
+```powershell
+irm https://penguin.ooo/install.ps1 | iex
+penguin web
+```
+
+The full record of what changed is in [changelog/0.2.3/](https://github.com/Prism-Shadow/penguin-harness/tree/main/changelog/0.2.3).
diff --git a/packages/landing/content/blog/penguinharness-0-2-3.zh.md b/packages/landing/content/blog/penguinharness-0-2-3.zh.md
new file mode 100644
index 00000000..0e04ef91
--- /dev/null
+++ b/packages/landing/content/blog/penguinharness-0-2-3.zh.md
@@ -0,0 +1,102 @@
+---
+title: "PenguinHarness 0.2.3:更好的用户体验,更透明的 Agent 运行"
+date: 2026-08-19
+category: news
+excerpt: 这一版集中在两件事上,一是让 Agent 正在做什么变得看得见,二是把用起来别扭的地方磨掉。侧栏能实时看出哪个会话在跑,上下文压缩从一段空白变成可以展开读的记录,任何一条回复都能分叉出新会话,排队的消息可以撤回来改了再发。
+---
+
+我们发布了 PenguinHarness 0.2.3。这一版集中在两件事上,一是让 Agent 正在做什么变得看得见,二是把用起来别扭的地方磨掉。
+
+## 侧栏能看出哪个会话在跑
+
+以前侧栏里只有一行的状态是准的,就是你当前打开的那个会话。其余每一行显示的都是上次拉取列表时的快照。你离开一个正在跑的会话,它会一直显示成在跑;而在后台开始跑的会话,那一行根本不会变。
+
+现在会话状态走用户级的事件流推送,每一行都是实时的。运行中是一个转动的沙漏,跑完了而你还没看过是一个绿点,看过之后就什么都不显示。
+
+![侧栏的三个会话,分别显示未读绿点、无标记和运行中的沙漏](/blog-assets/penguinharness-0-2-3-sidebar-status-zh.png)
+
+这里有个不显眼的坑。会话是否"跑过"是服务端列表行上的一个字段,而事件只更新状态,不更新它。所以一个第一次跑任务的新会话,沙漏一停就会什么都不显示。现在这个字段跟着状态一起推送,第一次跑完也能正常落到绿点上。
+
+## 压缩不再是一段空白
+
+上下文压缩以前是会话里的一段空白,你只知道它发生过。现在它是一行可以展开的记录,默认折叠,和任务组里的思考一样。
+
+![折叠状态的压缩行](/blog-assets/penguinharness-0-2-3-compaction-collapsed-zh.png)
+
+摘要在生成的时候就流式写进去,展开就能看到 Agent 到底保留了什么。
+
+![展开后的压缩摘要,包含任务、已确认的结论、下一步和注意事项](/blog-assets/penguinharness-0-2-3-compaction-expanded-zh.png)
+
+压缩的时机也改了。模型产生工具调用之后触发的压缩,现在会等工具执行完,带着工具返回一起压,而不是把一次没结束的交换塞进去。压缩到一半你退出程序,那一次会被标记为失败,重新加载时直接丢掉写了半截的内容。清空上下文这种模式不再顶着"压缩"两个字,它现在就叫清空。
+
+还修掉一个反复有人碰到的问题。客户端重启之后,明明满屏都是历史,压缩却报"当前上下文没有可压缩的内容"。原因是引擎对象是惰性构造的,会话从磁盘恢复时轮次数确实回填进了初始状态,但那个值只在引擎的构造函数里才会生效,而这时候引擎还没被造出来。顺带把服务端这类 409 拆成了三个错误码,中文界面下报错也是中文了。
+
+## 任何一条回复都能分叉出新会话
+
+对话跑歪了,以前要么继续将就,要么从头开一个。现在每条已完成的回复右边都有一个分叉按钮,就在复制旁边。
+
+![回复下方的操作行,分叉按钮在复制右边](/blog-assets/penguinharness-0-2-3-session-fork-actions-zh.png)
+
+点它会先问一句,确认之后把这条回复之前的对话原样复制成一个新会话,原来那个不动。
+
+![分叉前的确认弹窗](/blog-assets/penguinharness-0-2-3-session-fork-confirm-zh.png)
+
+## 排队的消息可以撤回来改
+
+任务跑着的时候插的话、排在后面的追问,以前发出去就收不回来了。现在它们旁边有一个回弯的箭头,点一下消息回到输入框,改完再发。
+
+撤回和引擎取用之间是互斥的,同一条消息不会既被送出去又被撤回来。如果引擎刚好在那一瞬间把它取走了,你会看到一条明确的提示,而不是消息悄悄消失或者发两遍。
+
+## 思考等级
+
+AgentHub 升到 0.4.4 之后多了一档 max,梯度现在是低、中、高、极高、最高。按钮上只写中文,下拉里带上英文的原始值。
+
+![思考等级的下拉菜单](/blog-assets/penguinharness-0-2-3-thinking-level-menu-zh.png)
+
+会话中途换等级会先弹一个框。有些模型服务商是靠在提示词最前面注入不同前缀来实现思考等级的,中途一换,整段历史的前缀缓存就失效了,下一次请求要按未缓存的价格重新计一遍。所以这个框给三个选择,先压缩再切换、直接切换、取消。
+
+![中途切换思考等级时的确认框](/blog-assets/penguinharness-0-2-3-thinking-level-switch-guard-zh.png)
+
+换完的等级会存在会话上,刷新页面、换个标签页都还在。
+
+## 加自定义模型不用再猜接口
+
+AgentHub 0.4.2 之后有三种通用协议客户端,OpenAI Responses、Anthropic Messages 和 OpenAI Chat Completions。填自定义 base URL 的时候,你现在不用自己判断对方是哪一种。检测按钮会按 Responses、Messages、Chat Completions 的顺序去探,第一个能对上的就是它,结果直接填进协议选择器里。没填 key 的时候会回退到环境变量,探不出来就弹一句提示让你查 key 和地址。
+
+![base URL 输入框右上角的检测协议按钮,以及框内列出三种协议和各自端点路径的选择器](/blog-assets/penguinharness-0-2-3-model-add-protocol-zh.png)
+
+视觉能力也能探。它会发一张 1x1 的图过去,正常应答就打开,模型明确拒绝图片就关掉,其他情况一律不动你的设置。这个探测会真实产生一次调用费用,所以只在你点的时候跑,保存模型的时候不会偷偷跑。
+
+支持 fast mode 的模型多了一个开关,默认关。不支持的模型上根本不显示这个开关,免得你打开之后每一轮都报错。打开的时候会提示一句,fast mode 是溢价计费的。
+
+## 还有一些
+
+会话列表的悬停操作改回了归档和删除,右键菜单里是完整的那一套。附件和图片可以直接拖到聊天区域上传,单个文件的上限从 10MB 放宽到 100MB,管理员可以在用户菜单里改这个数。内联图片是例外,单独限在 20MB,因为图片的字节会进对话正文和轨迹文件,而轨迹每次翻历史都要整个读一遍;文件附件不受这个约束,模型是按路径去打开它的。CLI 加了 `/thinking` 和 `--thinking`,长工具输出默认折叠成首尾几行,`/verbose` 展开,模型和轨迹拿到的一直是完整内容。
+
+模型目录这一版也刷了一遍。新增 Gemini 3.7、GLM-5.3、GPT-5.6 系列,以及它们在 OpenRouter 上的对应条目,其中 OpenAI 那九行走 Responses 协议。DeepSeek 的价格按官网最新的改过了,下架的 ling-3.0-flash 移除。
+
+底层还有一个值得说的修复。经过 one-api、OpenRouter 这类网关的长时间生成,网关会往流里塞心跳事件,而这些事件以前会被当成未知输出直接让整条流失败,harness 又把它判成可重试,重试再撞上下一个心跳,表现出来就是没完没了的"正在重连"。AgentHub 0.4.3 起所有流式客户端都会跳过这些心跳。
+
+## 从跑得动到看得懂
+
+前面这些改动落在同一件事上。Agent 干活的时候,你应该知道它在干什么、它把什么记住了、它下一步要花多少钱。会话状态、压缩记录、切换等级前的提示,都是把原本只有日志里才有的东西搬到界面上。
+
+## 安装与升级
+
+桌面端到官网下载对应平台的安装包。这一版开始 Windows 构建也带签名了,macOS 从 0.2.2 起就已经签名并公证。
+
+Linux 和 macOS 用命令行装。
+
+```sh
+curl -fsSL https://penguin.ooo/install.sh | sh
+penguin web
+```
+
+Windows 用 PowerShell。
+
+```powershell
+irm https://penguin.ooo/install.ps1 | iex
+penguin web
+```
+
+完整的改动记录在 [changelog/0.2.3/](https://github.com/Prism-Shadow/penguin-harness/tree/main/changelog/0.2.3)。
diff --git a/packages/landing/package.json b/packages/landing/package.json
index eb72a55d..a284dc52 100644
--- a/packages/landing/package.json
+++ b/packages/landing/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-landing",
- "version": "0.2.2",
+ "version": "0.2.3",
"private": true,
"type": "module",
"description": "PenguinHarness landing page (React + Vite + Tailwind CSS): multilingual, light/dark themes, benchmark showcase, and a local Markdown blog, deployed to GitHub Pages via GitHub Actions.",
diff --git a/packages/landing/test/blog.test.ts b/packages/landing/test/blog.test.ts
index 42375416..efd48d0f 100644
--- a/packages/landing/test/blog.test.ts
+++ b/packages/landing/test/blog.test.ts
@@ -96,7 +96,7 @@ describe("frontmatter mapping (author / pinned / category)", () => {
it("reads the pinned flag and sorts the pinned post first", () => {
for (const locale of ["en", "zh"] as const) {
const posts = postsFor(locale);
- expect(posts.length).toBe(16);
+ expect(posts.length).toBe(17);
// The launch post stays the single pinned post; newer posts sort under it by date.
expect(posts.filter((p) => p.pinned).map((p) => p.slug)).toEqual([
"introducing-penguinharness",
@@ -120,6 +120,7 @@ describe("frontmatter mapping (author / pinned / category)", () => {
it("filters by the news category, newest first", () => {
expect(postsFor("en", "news").map((p) => p.slug)).toEqual([
"introducing-penguinharness",
+ "penguinharness-0-2-3",
"penguinharness-0-2-2",
"penguinharness-0-2-1",
"penguinharness-0-2-0",
diff --git a/packages/server/package.json b/packages/server/package.json
index 6dd2a540..e811c012 100644
--- a/packages/server/package.json
+++ b/packages/server/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-server",
- "version": "0.2.2",
+ "version": "0.2.3",
"type": "module",
"description": "PenguinHarness web server: multi-user authentication and authorization, Session execution with SSE streaming, and usage accounting, built on @prismshadow/penguin-core.",
"license": "Apache-2.0",
diff --git a/packages/skills/package.json b/packages/skills/package.json
index b1a7998a..97bcebf0 100644
--- a/packages/skills/package.json
+++ b/packages/skills/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-skills",
- "version": "0.2.2",
+ "version": "0.2.3",
"type": "module",
"description": "PenguinHarness skill library: built-in SKILL.md documents and skill groups, decoupled from core.",
"license": "Apache-2.0",
diff --git a/packages/web/package.json b/packages/web/package.json
index 9433f070..706d3ee7 100644
--- a/packages/web/package.json
+++ b/packages/web/package.json
@@ -1,6 +1,6 @@
{
"name": "@prismshadow/penguin-web",
- "version": "0.2.2",
+ "version": "0.2.3",
"private": true,
"type": "module",
"description": "PenguinHarness web frontend SPA: React + Vite + Tailwind CSS, rendering conversations streamed over the OmniMessage protocol, plus Agent configuration, model configuration, usage stats, and Traces.",