Skip to content

1.5.13

Latest

Choose a tag to compare

@github-actions github-actions released this 05 Jul 10:51

1.5.13 Agent Chat Performance ✨

⚡ Agent Performance (#420)

  • Improved streaming refresh performance during long conversations, reducing UI re-render pressure when models output quickly.
  • Fixed memory growth after heavy Subagent use: completed task execution history could be duplicated and kept accumulating.
  • Smoother typing in long sessions by cutting unnecessary re-renders of the history area and input editor.
  • Fixed Agent runtime state lingering in memory after deleting long conversations.
  • Faster timeline rendering for long chats and Agent streaming: history now uses a more stable row-level layout, with less churn on older messages while a reply is generating; Chat and Quick Ask history paths are also improved for input, retry, branching, and tool results.

🚀 Startup

  • Reduced initialization work at plugin startup (#425).

✨ Experience Improvements

  • Removed misleading skill-read prompts when all Skills are disabled, so the model is no longer nudged to read non-existent skill paths.
  • Improved Tab completion context so the model better understands cursor position and surrounding text; more leading context is kept so suggestions fit headings, paragraphs, and prior context more naturally.
  • Removed the empty-state card shadow that showed as clipped gray edges on narrow mobile screens (#430).
  • Fixed some OpenAI-compatible models rejecting plain-text-only messages: text-only requests now fall back to a more compatible legacy string format without affecting image or PDF inputs.
  • Better Obsidian pop-out window behavior: fixed the Agent settings dropdown and opening chat in a right split incorrectly jumping back to the main window (#432) (#434).

1.5.13 Agent 对话性能优化 ✨

⚡ Agent 性能 (#420)

  • 优化长对话生成期间的流式刷新性能,减少快速模型输出时的界面重渲压力。
  • 修复长时间使用 Agent 并频繁派遣 Subagent 后,已完成任务的执行历史可能被重复保留、导致内存持续增长的问题。
  • 优化长对话中的输入体验,减少输入时历史消息区和编辑器组件的无意义重渲,缓解长会话下输入框卡顿。
  • 修复删除长对话后部分 Agent 运行时状态仍留在内存中的问题,降低长期使用后的内存累积风险。
  • 优化长对话和 Agent 流式回复时的时间线渲染:历史消息以更稳定的行级结构展示,减少当前回复生成时对旧消息列表的无意义刷新;同时改进 Chat 与 Quick Ask 的历史消息渲染路径,降低长会话中输入、重试、分支和工具结果展示时的卡顿风险。

🚀 启动优化

  • 优化插件启动时的初始化工作,减少启动阶段的加载开销 (#425)。

✨ 体验改进

  • 移除无技能场景下的技能读取误导提示,避免关闭全部 Skills 后模型仍被提示尝试读取不存在的 skill path。
  • 优化 Tab 补全的上下文设计,让模型更明确地理解光标位置与前后文关系;保留更完整的光标前内容,使补全结果更容易贴合标题、段落结构和前文语境。
  • 移除聊天空状态卡片的阴影样式,修复移动端窄屏下左右被裁切成灰边的视觉问题 (#430)。
  • 修复部分 OpenAI-compatible 模型无法处理纯文本消息的问题:纯文本请求会自动使用兼容性更好的旧式字符串格式,同时不影响图片和 PDF 等多模态输入。
  • 改进 Obsidian 弹出窗口兼容性:修复 Agent 设置下拉菜单和右侧分屏打开聊天时错误回到主窗口的问题 (#432) (#434)。