Skip to content

v0.2.12

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 03 Jul 15:42
v0.2.12
1c05853
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

更新摘要

本次迭代核心

新增图片理解工具、工具审批栏、Remote Connect LAN IP 选择，并大幅增强 Computer Use、PPT Live、Flow Chat 长会话、终端和远程 SSH 会话体验。

关键特性

  • [图片理解工具] 新增 analyze_image 工具，Agent 可使用配置的图片理解模型分析本地或远程 Workspace 图片；聊天粘贴/上传图片、普通对话和 /btw 侧边问题都支持图片上下文。
  • [工具审批栏] Flow Chat 新增统一审批栏，支持批准、拒绝、带指令拒绝、权限选项、倒计时和一键开启自动执行；文件修改类工具会进入审批流程。
  • [Computer Use 升级] 统一 computer_use 工具入口，并大幅增强 macOS / Windows 桌面自动化能力，包括后台点击、输入、拖拽、窗口截图、应用状态读取和权限提示。
  • [Computer Use 工具卡] Flow Chat 新增专用 Computer Use 工具卡，可展示执行动作、目标应用、循环提示和权限问题，让桌面自动化过程更清楚。
  • [远程连接增强] Remote Connect LAN 模式支持选择具体本机 IPv4 网卡，二维码会使用所选 IP；远程 IM Bot 新增 /model 模型切换。
  • [SSH 连接体验升级] SSH 连接弹窗支持搜索已保存连接和 ~/.ssh/config 主机，并改进 Include / Match、带引号路径和表单回填流程。
  • [PPT Live 生成增强] PPT Live 将阶段事件和 Agent 流合并为统一时间线，隐藏无关工具噪声并显示实时进度，长时间生成时反馈更明确。
  • [PPTX 导出优化] editable PPTX 导出改进文字与背景分层，减少文字重叠、重复或丢失，提升后续编辑稳定性。
  • [长会话体验优化] Flow Chat 长会话采用窗口化历史加载，打开、上滑和搜索更轻量，并增加旧历史加载边界提示。
  • [终端体验升级] 终端快速输入改为顺序批量写入，并增强 IME 场景下的字符处理，减少 macOS / WKWebView 下快速输入丢字问题。
  • [Agent 工具增强] Read 支持读取文件尾部，Glob 会返回搜索范围、总匹配数和截断状态，MCP 大输出也会走统一的大结果存储策略。
  • [启动与主题响应优化] 缩短启动 splash handoff，压缩内置 Noto Sans SC 字体资源，并改善首屏主题一致性。

关键修复

  • 修复远程 SSH 会话在恢复、cron/session 工具、远程对话、历史重载和计划任务中可能丢失远程身份或误用本地路径的问题。
  • 修复模型切换期间仍可发送消息、前后端模型状态不同步、多模态能力丢失、图片理解默认模型异常等问题。
  • 优化 AI 连接测试，减少无意义重试，并提供更清晰的 TLS/证书、代理和网络错误提示。
  • 修复 Flow Chat 长会话发送后从列表消失、历史加载窗口不稳定、turn header 导航跳动和静态历史锚点跳转问题。
  • 修复 Slash 命令关闭、输入框焦点恢复、并发创建空会话、远程 Workspace 会话匹配等输入和会话稳定性问题。
  • 修复终端快速输入可能丢字、IME keyCode 229 输入异常、退出后仍写入导致错误级联等问题。
  • 修复图片预览被侧边栏遮挡、Markdown diff 文本重叠、ACP 工具卡片不可见、Companion pet 未读状态不清除等 UI 问题。
  • 修复 PPT Live 阶段标签错乱、生成流过噪、PPTX 导出文字和背景叠加异常等问题。
  • 修复 Windows Computer Use 截图、指针坐标、WGC/D3D11 参数和跨平台编译问题。
  • 修复 MCP 大工具输出被提前截断的问题，使其进入统一的大结果存储和预览流程。
  • 抑制 WebFetch readable extraction 中第三方 parser 噪声日志。
  • 补充 SECURITY 安全策略文档，并完善 OpenCode / Plugin runtime 相关架构说明。

What's Changed

Release Focus

This release adds image understanding, a Flow Chat tool approval bar, Remote Connect LAN IP selection, and major improvements to Computer Use, PPT Live, long-session Flow Chat, terminal input, and remote SSH
sessions.

Key Features

  • [Image understanding] Added the analyze_image tool, allowing agents to inspect local or remote Workspace images with the configured vision model. Pasted and uploaded images now work as image context in normal
    chat and /btw side questions.
  • [Tool approval bar] Flow Chat now has a unified approval bar with approve, reject, reject with instruction, permission options, countdowns, and a shortcut for enabling auto execution. File-modifying tools now
    go through approval.
  • [Computer Use upgrade] Unified desktop automation under the computer_use tool and significantly improved macOS / Windows automation, including background clicks, typing, drag operations, window capture, app-
    state reading, and permission guidance.
  • [Computer Use tool card] Flow Chat now has a dedicated Computer Use card showing the action, target app, loop hints, and permission issues, making desktop automation easier to follow.
  • [Remote connection improvements] Remote Connect LAN mode can use a selected local IPv4 interface, and the QR code uses that IP. Remote IM Bot sessions now support /model switching.
  • [SSH connection upgrade] The SSH dialog supports searching saved connections and ~/.ssh/config hosts, with better handling for Include / Match, quoted paths, and form fill before connection.
  • [PPT Live generation improvements] PPT Live now merges phase events and Agent stream entries into one timeline, hides unrelated tool noise, and shows live activity during long-running generation.
  • [PPTX export improvements] Editable PPTX export improves text/background layering, reducing overlapping, duplicated, or missing text and making exported decks more stable to edit.
  • [Long-session improvements] Flow Chat long sessions now use windowed history loading, making open, upward scroll, and search paths lighter while showing older-history boundary states.
  • [Terminal experience upgrade] Terminal input is now batched into ordered writes and handles IME cases more reliably, reducing dropped characters during rapid input on macOS / WKWebView.
  • [Agent tool improvements] Read can read file tails, Glob reports scope, total matches, and truncation state, and large MCP outputs now use the shared large-result storage path.
  • [Startup and theme responsiveness] Shortened startup splash handoff, compressed bundled Noto Sans SC font assets, and improved first-paint theme consistency.

Key Fixes

  • Fixed remote SSH sessions losing remote identity or using local paths during restore, cron/session tools, remote dialog turns, history reloads, and scheduled jobs.
  • Fixed sending during model switching, inconsistent frontend/backend model sync, lost multimodal capabilities, and image-understanding default model issues.
  • Improved AI connection tests with fewer unnecessary retries and clearer TLS/certificate, proxy, and network diagnostics.
  • Fixed Flow Chat issues around long sessions disappearing after send, unstable history windows, turn header navigation, and static-history anchor jumps.
  • Fixed slash command dismissal, input focus recovery, duplicate empty-session creation, and remote Workspace session matching issues.
  • Fixed terminal dropped characters during rapid input, IME keyCode 229 handling, and write errors after terminal exit.
  • Fixed image preview being covered by the sidebar, Markdown diff text overlap, hidden ACP tool cards, and Companion pet unread state not clearing.
  • Fixed PPT Live phase-label confusion, noisy generation stream output, and PPTX text/background layering issues.
  • Fixed Windows Computer Use screenshot, pointer coordinate, WGC/D3D11 argument, and cross-platform build issues.
  • Fixed MCP large tool output being pre-truncated before it could enter the shared large-result storage and preview flow.
  • Suppressed noisy third-party parser logs during WebFetch readable extraction.
  • Added SECURITY policy documentation and expanded OpenCode / Plugin runtime architecture docs.

New Contributors

Full Changelog: v0.2.11...v0.2.12

Contributors

circlecrystal, snvtac, and su-yunhe
Assets 22
Loading