Skip to content

Releases

DeepSeek++ 1.14.0

Choose a tag to compare

@github-actions github-actions released this 08 Aug 06:40

DeepSeek++ 1.14.0

1.14.0 是内联智能体交付体验升级与流式稳定性修复版本:最终回答与图表、代码产物改由 DeepSeek 原生渲染,推理过程分步展示,流式传输与代码块渲染更稳定。

主要变化

  • 内联智能体交付升级:最终回答改由 DeepSeek 页面原生渲染,mermaid/xychart 图表卡片、html/svg/xml 可运行代码块与原生预览面板直接展示;插件侧产物渲染流程移除。
  • 推理过程展示:逐轮捕获并展示思考过程,完成后自动折叠为「思考 · 第 N 步」条目;最终回答完整展开(渲染上限大幅提升),不再被折叠或截断。
  • 控制台定位与恢复:新控制台挂载到最新消息,虚拟列表窗口下的恢复定位更可靠,完整最终回复可正确恢复展开。
  • 流式与渲染稳定性:修复流式分帧可能丢失部分推理/正文增量的问题;代码块跨帧折叠不再破坏围栏内空行;代码运行状态在流式重渲染后保留;只有思考内容的回合不再重复提交(避免会话链分叉)。
  • 重生成授权修复:原生 regenerate 工具调用授权修复,仅成功执行过的调用作为授权证据,拒绝或中断的调用不再扩大授权范围。
  • 侧边栏:网页流式推理过程转发到侧边栏对话展示。
  • 权限变化:Chrome、Edge 和 Firefox 均不新增浏览器权限。

验证

  • npm run smoke:pow
  • npm run compile
  • npm pack --dry-run --workspace packages/shell-host
  • npm run build:all
  • npm run zip:all
  • npm run smoke:mcp
  • npm run smoke:shell
  • npm run verify:mcp:mock
  • npm run verify:automation
  • node scripts/web-search-smoke.mjs
  • npm run prompt:freeze
  • npm audit --audit-level=high --omit=dev
  • npm run ci:quality

Chrome Web Store 更新说明

内联智能体交付体验升级:最终回答与生成的图表、代码产物改由 DeepSeek 原生渲染展示,推理过程分步呈现并可折叠;修复流式传输可能丢失部分内容、代码块渲染被破坏等问题,代码运行结果在流式刷新后仍保留。Chrome、Edge 和 Firefox 均不新增浏览器权限。

DeepSeek++ 1.13.1

Choose a tag to compare

@github-actions github-actions released this 06 Aug 23:54

DeepSeek++ 1.13.1

1.13.1 是内联智能体体验打磨与对话导出稳定化版本:内联智能体的输出渲染与原生 Markdown 一致、工具状态展示更真实,对话导出的 HTML 格式与下载链路更可靠。

主要变化

  • 内联智能体渲染修复:Agent 输出改为与原生 Markdown 渲染一致的紧凑块级排版;剥离工具调用后的多余空行自动折叠,标题层级与正文对齐。
  • 内联智能体 UI/UX 打磨:工具执行状态如实展示(成功/失败/进行中),流式输出滚动更安全,暂停时显示中性页脚;图标改为 SVG 并统一类型与圆角体系,交互反馈更即时,键盘与读屏无障碍更好;启动失败不再无声。
  • 对话导出稳定化:HTML 导出包含输入/产出的统计信息,导出与下载链路更稳定,下载管理更可靠。
  • 权限变化:Chrome、Edge 和 Firefox 均不新增浏览器权限。

验证

  • npm run smoke:pow
  • npm run compile
  • npm pack --dry-run --workspace packages/shell-host
  • npm run build:all
  • npm run zip:all
  • npm run smoke:mcp
  • npm run smoke:shell
  • npm run verify:mcp:mock
  • npm run verify:automation
  • node scripts/web-search-smoke.mjs
  • npm run prompt:freeze
  • npm audit --audit-level=high --omit=dev
  • npm run ci:quality

Chrome Web Store 更新说明

内联智能体体验打磨:Agent 输出渲染与原生 Markdown 一致,工具执行状态展示更真实,交互反馈更即时、无障碍更好;对话导出的 HTML 格式与下载链路更稳定。Chrome、Edge 和 Firefox 均不新增浏览器权限。

DeepSeek++ 1.13.0

Choose a tag to compare

@github-actions github-actions released this 05 Aug 19:20

DeepSeek++ 1.13.0

1.13.0 是内联智能体引擎与本地 Skill 管理升级版本:内联智能体换用更稳定的新运行引擎并支持官方 API 后端,本地文件夹导入的 Skill 升级为索引导入并支持重关联与自动激活,后台性能与对话导出链路也更可靠。

主要变化

  • 内联智能体引擎升级:内联智能体运行引擎整体升级,工具执行与授权链路更稳定,续跑与事件处理保持一致。配置 DeepSeek 官方 API Key 后,内联智能体自动改用官方 API 后端运行(支持推理过程);未配置时仍使用页面会话链路,与侧边栏聊天行为一致。
  • 本地 Skill 管理升级:本地文件夹导入的 Skill 改为索引导入——数据库中只保存索引卡、内容按需从磁盘读取;支持目录重定位后的重新关联与隐式自动激活。Skill 激活后,会话中的 shell 工作目录自动限定在该 Skill 目录内,防止越界操作。
  • 技能生态导入:支持从 pi/agentskills.io 生态导入 SKILL.md 技能包,沿用现有本地导入流程,不新增存储键。
  • 后台性能优化:后台脚本结构性瘦身,i18n 资源按分区裁剪;存储读写改用浏览器原生 IndexedDB,启动与响应更快、内存占用更低。
  • 对话导出修复:当前会话历史为空时导出会明确失败提示;复制输出失败时显示具体原因,不再无声失败。
  • 权限变化:Chrome、Edge 和 Firefox 均不新增浏览器权限。

感谢 @zhangweildlh 贡献的本地 Skill 管理升级(#457):本地文件夹索引导入、目录重定位重关联、隐式自动激活与 Skill 会话 cwd 安全围栏。

验证

  • npm run smoke:pow
  • npm run compile
  • npm pack --dry-run --workspace packages/shell-host
  • npm run build:all
  • npm run zip:all
  • npm run smoke:mcp
  • npm run smoke:shell
  • npm run verify:mcp:mock
  • npm run verify:automation
  • node scripts/web-search-smoke.mjs
  • npm run prompt:freeze
  • npm audit --audit-level=high --omit=dev
  • npm run ci:quality

Chrome Web Store 更新说明

中文:

  • 内联智能体引擎升级:工具执行与续跑更稳定;配置官方 API Key 后自动改用官方 API 后端运行(支持推理过程),未配置时仍使用页面会话链路。
  • 本地 Skill 管理升级:本地文件夹导入改为索引导入,支持目录重定位后的重新关联与隐式自动激活;Skill 会话的 shell 工作目录自动限定在 Skill 目录内。
  • 支持从 pi/agentskills.io 生态导入 SKILL.md 技能包。
  • 后台性能优化:启动与响应更快、内存占用更低。
  • 对话导出在会话历史为空或复制失败时给出明确提示。
  • 不新增浏览器权限。

English:

  • The inline agent runs on an upgraded engine with more stable tool execution and continuation; when an official DeepSeek API key is configured it automatically uses the official API backend (with reasoning), otherwise it keeps using the page session.
  • Local Skill management is upgraded: folder imports now use an index-based flow with re-association after relocation and implicit auto-activation, and Skill sessions confine shell working directories to the Skill folder.
  • SKILL.md skill packs from the pi/agentskills.io ecosystem can be imported through the existing local import pipeline.
  • Background performance is improved: faster startup and responses with lower memory use.
  • Conversation export now shows clear messages when session history is empty or copying fails.
  • No new browser permissions.

DeepSeek++ 1.12.2

Choose a tag to compare

@github-actions github-actions released this 04 Aug 19:23

DeepSeek++ 1.12.2

1.12.2 是工具执行透明化与生产体验修复版本:shell_exec 等工具失败不再黑箱、可一键导出诊断日志、长文产出自动保存、复制完整输出、导出内容范围可选,并让内联智能体的续跑计数与实际步数一致。

主要变化

  • 工具执行透明化:shell_exec 等工具失败时展示实际执行的命令与失败原因;授权被拒时显示具体原因(如授权关闭、过期、会话不匹配、工具禁用)。设置 → 数据 新增「导出诊断日志」按钮,一键导出最近一次会话内的工具执行、授权与错误记录,方便排查问题。
  • 内联智能体续跑计数更准确:达到自动续跑上限暂停时,提示文案改为报告实际完成的步数,与「发送『继续』续跑」的提示口径一致。
  • 长文产出自动保存:Agent 产出长文时自动保存为产物并显示下载入口,无需再手动要求调用下载工具。
  • 复制完整输出:回复操作区新增「复制完整输出」按钮,一次复制 DeepSeek 的完整回复,不再受官方复制按钮限制。
  • 对话导出范围选择:导出 Markdown/HTML 时可选「仅输入+产出」或「全部内容」(含工具调用、续跑信息、思考过程、插件注入信息),并记住上次选择,便于排查与日常归档两用。
  • 权限变化:Chrome、Edge 和 Firefox 均不新增浏览器权限。

感谢 @zhangweildlh 对 v1.12.0 的详细使用反馈(#494),本版针对其中的工具执行、导出、复制与长内容链路问题逐项落地。

验证

  • npm run smoke:pow
  • npm run compile
  • npm pack --dry-run --workspace packages/shell-host
  • npm run build:all
  • npm run zip:all
  • npm run smoke:mcp
  • npm run smoke:shell
  • npm run verify:mcp:mock
  • npm run verify:automation
  • node scripts/web-search-smoke.mjs
  • npm run prompt:freeze
  • npm audit --audit-level=high --omit=dev
  • npm run ci:quality

Chrome Web Store 更新说明

中文:

  • 工具执行更透明:失败时显示实际执行的命令与原因,授权被拒显示具体原因;设置中新增「导出诊断日志」,可一键导出排障信息。
  • 对话导出支持内容范围选择:可选「仅输入+产出」或「全部内容」,并记住上次选择。
  • 内联智能体续跑计数与实际步数一致;长文产出自动保存为产物并提供下载。
  • 回复操作区新增「复制完整输出」,一次复制完整回复。
  • 不新增浏览器权限。

English:

  • Tool execution is more transparent: failures show the actual command and reason, authorization denials show the specific cause, and a new "Export diagnostic logs" option in Settings exports troubleshooting data in one click.
  • Conversation export now supports a content scope choice: "inputs + outputs only" or "everything", remembering your last selection.
  • Inline agent continuation counts now match actual steps; long outputs are auto-saved as artifacts with a download entry.
  • A new "Copy full output" button copies the complete reply in one click.
  • No new browser permissions.

DeepSeek++ 1.12.0

Choose a tag to compare

@github-actions github-actions released this 04 Aug 08:47

DeepSeek++ 1.12.0

1.12.0 是 Agent 工作台与文件引用体验更新,让内联智能体运行更可靠、步骤清晰可见,并支持在聊天中通过 @ 引用可信目录中的文件。

主要变化

  • 结构化 Agent 步骤界面:Agent 运行时展示清晰的步骤与全局运行指示,循环结束不再无声无息,重复步骤有重试上限,避免无限循环;窗口类工具(如浏览器控制)的结果按窗口展示,续跑提示更准确。
  • 原生搜索协调:Agent 使用 DeepSeek 原生智能搜索时,不再触发扩展自带的联网搜索工具,避免重复搜索。
  • 可信目录与 @ 文件引用:新增“项目文件”设置页,可选择带扫描预算的本地可信目录;在聊天中用 @ 打开引用面板,把目录中的图片送入对话。只在你选择并保存的目录内生效,路径越界的内容会被忽略。
  • 文件引用健壮性:图片 MIME 规范化避免同一图片重复上传;扩展上下文失效(如重载扩展后旧页面残留)时不再误报记忆保存失败。
  • 用量图表修正:30 天 Token 趋势图 X 轴只显示日期,标签不再挤出图表面板。
  • MCP 兼容性:MCP 客户端能力声明改为符合协议规范的有效值,与校验客户端能力的服务端连接更稳定。
  • 页面加载清理:扩展页面不再输出 modulepreload 预加载链接,消除浏览器控制台中的相关噪音。
  • 权限变化:Chrome、Edge 和 Firefox 均不新增浏览器权限。

验证

  • npm run smoke:pow
  • npm run compile
  • npm pack --dry-run --workspace packages/shell-host
  • npm run build:all
  • npm run zip:all
  • npm run smoke:mcp
  • npm run smoke:shell
  • npm run verify:mcp:mock
  • npm run verify:automation
  • node scripts/web-search-smoke.mjs
  • npm run prompt:freeze
  • npm audit --audit-level=high --omit=dev
  • npm run ci:quality

Chrome Web Store 更新说明

中文:

  • 内联智能体运行更可靠:新增结构化步骤界面与全局运行指示,循环结束可见、重复步骤有重试上限,使用 DeepSeek 原生智能搜索时不再重复触发扩展自带联网搜索。
  • 新增可信目录与 @ 文件引用:在设置中选择一个本地目录后,可在聊天中用 @ 引用目录中的图片;只在你选择的目录内生效,越界内容会被忽略。
  • 修复 30 天 Token 趋势图 X 轴溢出,以及扩展上下文失效时误报保存失败的问题;改善与部分 MCP 服务的兼容性。
  • Chrome 版本不新增浏览器权限。

English:

  • Agent runs are more reliable: structured step UI with a global running indicator, visible loop endings, bounded step retries, and no duplicate built-in web search when DeepSeek native intelligent search is active.
  • New trusted directory and @ file references: after choosing a local directory in Settings, you can reference images from it in chat with @; only your chosen directory is scanned and out-of-bounds paths are ignored.
  • Fixes the 30-day token trend X-axis overflow and false save-failure reports after an extension context becomes invalid; improves compatibility with some MCP services.
  • The Chrome package adds no new browser permissions.