wisp-science v0.34.0

wisp-science v0.34.0

v0.34.0 聚焦科研工作区的可观测性、跨平台稳定性与长对话性能：新增工作区和上下文用量面板，支持项目级 AGENTS.md 指令与 Skill ZIP 导入；同时修复 Windows 子进程、运行输出、Agent 委派和多层界面交互中的一系列问题。

下载

• macOS（.dmg，已签名 + 公证）：请选择 Apple Silicon（aarch64）或 Intel（x86_64）对应架构；兼容版本可从应用内更新。
• Windows 安装包（MSI / NSIS）：当前尚未签名，可能触发 SmartScreen；可选择「仍要运行」。

见下方 Assets。

核心亮点

用量与记忆管理

• 新增按类别拆分的上下文用量面板，可查看当前会话的上下文构成与详细占用。
• 设置页新增工作区存储用量仪表盘，并区分项目存储；记忆浏览器现在可切换任意项目。
• 修复记忆工作区选择器，并简化笔记界面与清除操作样式。

Agent、Skill 与项目指令

• Agent 可加载项目根目录的 AGENTS.md 指令，使项目约定随工作区自动生效。
• Skill 支持从 ZIP 包安全导入；Skill 绑定的子 Agent 不再因 token 预算进入死锁，默认委派预算改为不限额。
• 延迟 MCP 工具可正确授权给委派 Agent；降级的委派结果不再被丢弃。
• Skill portfolio 规划改为由模型驱动。

性能与大文件体验

• 流式回复期间可增量渲染 Markdown，并隔离 transcript 渲染热点、优化行 key 与克隆路径。
• 大文本预览只读取有界的文件头部，避免为预览加载整个文件。
• 修复运行输出轮询时的抖动与自动滚动问题，并让运行计时和聊天位置保持实时更新。

Windows 与执行稳定性

• Windows 子进程默认隐藏控制台窗口，并串行化 Git 启动，避免窗口闪烁、锁竞争和异步运行时阻塞。
• 系统托盘新增重启操作。
• 执行工具不再获取不必要的项目锁；R/Python 解释器可从 PATH 之外的常见位置发现。
• OpenAI 兼容服务探测支持更多 API 路径。

界面与交互修复

• 修复聊天粘贴、更新交互、Agent 菜单闪烁、上下文菜单越界和多层设置 Escape 返回行为。
• 中途编辑消息前新增破坏性操作确认。
• 优化上下文用量面板宽度、设置页按钮与弹窗关闭样式。

完整变更

自 v0.33.0 以来的全部提交见：
v0.33.0...v0.34.0

What's Changed

• Fix Skill-bound sub-Agent token budget deadlock (#648) by @xuzhougeng in #649
• fix(ui): keep context menus inside the viewport near window edges by @xuzhougeng in #652
• fix: discover R/Python interpreters beyond PATH and surface them in settings (#651) by @xuzhougeng in #653
• feat: default delegated Agent budgets to unlimited by @xuzhougeng in #655
• feat(skills): import ZIP packages by @xuzhougeng in #656
• fix(ui): keep run output panels pinned to the tail across poll rebuilds (#654) by @xuzhougeng in #657
• feat: confirm before destructive mid-conversation message edit by @xuzhougeng in #658
• fix: preserve delegated task deliveries as degraded results instead of discarding them by @xuzhougeng in #660
• feat: add categorized context usage panel by @xuzhougeng in #662
• fix(ui): keep Run elapsed time and long-chat position live by @xuzhougeng in #664
• fix(ui): align context usage panel with composer width by @xuzhougeng in #665
• Make Skill Portfolio planning model-driven by @xuzhougeng in #667
• fix(settings): separate project storage usage by @xuzhougeng in #668
• feat(settings): add workspace usage dashboard by @xuzhougeng in #669
• fix(ui): stabilize chat, paste, and update interactions by @xuzhougeng in #673
• feat(core): support project AGENTS.md instructions by @xuzhougeng in #674
• fix(preview): load only a bounded head for large text files by @xuzhougeng in #677
• fix(ui): keep agent options stable while streaming by @xuzhougeng in #679
• feat(ui): inspect context usage details by @xuzhougeng in #680
• feat(windows): add tray restart action by @xuzhougeng in #682
• fix: probe OpenAI-compatible chat completion paths by @JohnnyChen1113 in #681
• refactor(ui): split main.rs and app_support.rs into cohesive modules by @xuzhougeng in #683
• fix(ui-tests): deflake reasoning stream and long-scroll tests by @xuzhougeng in #685
• perf(ui): make transcript row keys O(1) and clone rows lazily by @xuzhougeng in #684
• fix(ui): stop run output panels from jittering each poll by @xuzhougeng in #687
• fix(ui): measure completed turn wall time by @xuzhougeng in #688
• fix(windows): hide subprocess consoles on startup by @xuzhougeng in #689
• perf(ui): isolate transcript rendering hot paths by @xuzhougeng in #690
• perf(ui): render Markdown during streaming by @xuzhougeng in #691

New Contributors

• @JohnnyChen1113 made their first contribution in #681

Full Changelog: v0.33.0...v0.34.0