Skip to content

Agentero v0.5.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 07 Aug 10:56
· 822 commits to main since this release
v0.5.0
0f2cf20

概要

v0.5.0 以 PDF 版面理解与阅读体验、知识库 Doctor、翻译与 Agent 一键安装 为主线：

  • 解析论文中插图、表格、算法、公式
  • 并支持版面全文翻译，补充BYOK翻译服务
  • CLI 性能 与 doctor页面
  • Agent 一键安装
  • 提升 PDF 渲染性能

新功能

  • PDF 版面分析: 接入 PP-DocLayoutV3，打开论文或入库后可自动/后台解析，结果缓存为 source/layout.json，并生成与侧栏同构的 layout-index.json。（dc154164、fb6e318c、f05d5bb0）
  • 插图 / 公式侧栏: 阅读侧栏按图片、表格、算法与带编号公式分组展示，支持版面控件与分析进度，悬停可发起视觉提问。（61ffcf56、21799ee6）
  • 版面全文翻译: 按阅读顺序批量翻译正文/摘要/标题/图注等区域，译块完成即叠纸色遮罩，并适配暗色模式。（96260eab、2a95f440）
  • Vault Doctor: 设置页与 CLI 提供只读诊断（目录结构、Catalog、双链、论文 aliases），修复前需确认；可生成英文/中文别名、探测坏链并 git 风格 diff 批量写入，也可把提示词交给 Agent；支持别名检查忽略/恢复。（#210、#217、22049895）
  • 商业翻译 BYOK 与 DeepLX: 设置中可配置商业翻译提供商并探测可用性；默认免费链路纳入 DeepLX，状态以彩色圆点展示。（42da9797、eb256418、0b9e64f9）
  • Agent 静默安装 / 升级: 设置里分别展示 Agent 与 ACP 状态，支持无终端确认的安装、安装 ACP 与已装 CLI 升级；常用 Agent 列表位置与文案一并调整。（536ebaf0、57f931c9、37a0dd22）
  • 划词钉到对话: PDF 划词提问会写入 agent-trace，在页边形成可回看的对话钉与选区卡片。（7cf52f7a）
  • equation-annotation Skill: 内置公式/变量批注 Skill，可生成中英对照的 Annotation.md 术语表。（3a54eadb）
  • CLI layout 与区域 mark: agentero layout list/get 读取版面索引；mark add --region 等支持区域锚点批注，默认高亮色与桌面一致为黄色。（41e4d340、47111be0）
  • 诊断遥测（可退出）: 可选诊断上报（崩溃、错误日志、设备与已装 ACP 等），本地队列批量上传；设置 → 通用 → 隐私可关闭或手动发送。（96605306）
  • 标题栏更新标签: 有可用更新时显示「新版本」标签，点击下载安装并重启。（11866065）
  • 后台任务进度环: 后台任务以进度环展示，悬停或点击展开详情。（b35a7f58）
  • PDF 独立配色: 阅读器可独立切换配色，不必跟随系统主题。（51766376）
  • 编辑器属性入口: Frontmatter / Properties 移入工具栏，阅读整理路径更短。（17091517）
  • 侧栏 arXiv 链接: 论文信息区补充 arXiv 相关服务入口。（9f3375fb）

改进

  • PDF 渲染性能: 优先 Worker 引擎（失败回退主线程），限制 base-coat / DPR，放大 tile、预渲染外围环、合并缩放步进，减轻滚动闪白与卡顿。（60d06459、9985326c、9af0991e）
  • PDF 阅读控件: 适应缩放移至底部分页旁；选区卡片贴钉定位；视觉批注缩略图进侧栏；背景任务默认收成环。（5a1a22a5、0cc51ae6、178e9106）
  • Catalog 单一数据源: 不再写入 metadata.json / reading-meta.json，论文元数据以 .agentero/catalog.sqlite 为准，页数等直接读 PDF。（1dd7839a、5528b349）
  • 内置 Skill 按 version 升级: 以 frontmatter version 比较替代 hash；版本更低才升级，用户更高/自定义版本保留。（#224）
  • 去掉 Zen 模式: Agent 专注布局入口移除，面板与快捷方式统一到常规壳层。（ae47149f）
  • Agent 排队计数: 作曲区显示已排队提示数量。（9e7b7732）
  • Zotero 集合目录默认开启: 迁移导入默认按集合建文件夹。（0635b867）

修复

  • Agent 历史会话恢复: 打开历史不再空白；同会话续聊不再拆成多条；续聊偶发只显示提问的问题已处理。（#221）
  • AskUserQuestion 转发: Agent 向用户提问后的回答会正确回传会话。（2905443a）
  • ACP 斜杠命令名: 去掉命令名中多余的 $，避免 slash 转发错误。（990dd3d5）
  • Zotero Connector 附件: 正确声明 filesEditable 并解析 PDF，修复仅写入元数据、缺附件（如 IEEE 落地页）的情况。（d22f47fe）
  • 索引文件夹下嵌套论文: 组织目录带 index NOTES.md 时不再吞掉子论文单元。（b7b62ac8）
  • 非论文标签保留信息栏: 切到非 paper 标签时论文信息不被清空。（717a5bc0）
  • CLI 入库后库刷新: 文件监听/导入后元数据与论文库及时更新。（2e594f85）
  • 版面分析细节: 同栏公式合并、双栏顺序、图文误标、可无 ONNX 从 layout.json 重分析等多项稳定性修复。（c1800c4a、6ace73e7、8c5cefa7）

已解决问题

下列在区间内关闭、且已由上方条目覆盖的 issue 不再重复展开：#193、#198、#209、#215、#182、#201、#200、#203、#225、#218、#204、#219、#170（区域部分）、#148（体验相关）。

Summary

v0.5.0 centers on PDF layout understanding, Vault Doctor, and translation / Agent install polish. Papers can be analyzed in the background on open or import, with a figures-and-formulas rail and progressive full-text layout translation. Settings and CLI diagnose and repair aliases and broken wikilinks without silent rewrites. Commercial BYOK translation, DeepLX defaults, silent Agent CLI install/upgrade, PDF performance work, and Catalog-as-single-source metadata land together.

Features

  • PDF layout analysis: PP-DocLayoutV3 runs automatically after open/import, caches source/layout.json, and writes a sidebar-aligned layout-index.json. (dc154164, fb6e318c, f05d5bb0)
  • Figures / formula rail: Image, table, algorithm, and numbered-formula sections in the reading sidebar, with layout controls, progress, and hover visual ask. (61ffcf56, 21799ee6)
  • Layout bulk translate: Translates body/abstract/headers/captions in reading order with progressive paper-themed overlays, including dark-mode matching. (96260eab, 2a95f440)
  • Vault Doctor: Read-only checks for structure, Catalog, wikilinks, and paper aliases in Settings and CLI, with confirmed repairs, wikilink probe + git-style diffs, Agent prompt handoff, and alias ignore/restore. (#210, #217, 22049895)
  • Commercial BYOK translation & DeepLX: Configure and probe commercial providers; free defaults include DeepLX; status shows as color dots. (42da9797, eb256418, 0b9e64f9)
  • Silent Agent install / upgrade: Separate Agent vs ACP status, install / install-ACP / upgrade without a confirming terminal, plus settings list polish. (536ebaf0, 57f931c9, 37a0dd22)
  • Pin PDF selection chats: Selection asks write agent-trace marks so quotes become page-side conversation pins and cards. (7cf52f7a)
  • equation-annotation skill: Bundled skill for equation/variable glossaries producing bilingual Annotation.md. (3a54eadb)
  • CLI layout & region marks: agentero layout list/get plus mark add --region (and list/get/delete); default mark color yellow. (41e4d340, 47111be0)
  • Opt-out telemetry: Optional diagnostics (crashes, errors, device, installed ACP agents) with a local queue; opt out or send from Settings → General → Privacy. (96605306)
  • Title-bar update tag: Shows a “New version” tag when an update is available; click to download, install, and restart. (11866065)
  • Background task ring: Tasks show as a progress ring that expands on hover or click. (b35a7f58)
  • Independent PDF color scheme: Toggle reader colors without following the app theme. (51766376)
  • Editor properties in toolbar: Frontmatter/properties live in the editor toolbar. (17091517)
  • Sidebar arXiv links: Paper info adds arXiv service links. (9f3375fb)

Improvements

  • PDF render performance: Prefer the worker engine with main-thread fallback; cap base-coat/DPR, enlarge tiles, pre-render outer rings, and coalesce zoom steps. (60d06459, 9985326c, 9af0991e)
  • PDF reading chrome: Fit-zoom next to the page bar; selection cards track pins; visual-trace thumbs in sidebars; background tasks stay as a ring. (5a1a22a5, 0cc51ae6, 178e9106)
  • Catalog as sole metadata source: Stop writing metadata.json / reading-meta.json; Catalog SQLite owns paper metadata; page counts come from the PDF. (1dd7839a, 5528b349)
  • Skill upgrades by version: Frontmatter version replaces hash checks; only lower versions auto-upgrade so custom higher versions stay. (#224)
  • Remove zen mode: Drop the focused Agent layout path; panels and shortcuts stay on the normal shell. (ae47149f)
  • Queued prompt count: Composer shows how many prompts are queued. (9e7b7732)
  • Zotero collection folders on by default: Migration import defaults to collection folders. (0635b867)

Fixes

  • Agent history restore: Opening history is no longer empty; continuing a chat no longer splits sessions; occasional “question-only” resumes are fixed. (#221)
  • AskUserQuestion forwarding: User answers to Agent questions are forwarded into the session. (2905443a)
  • ACP slash command names: Strip stray $ from slash command names so commands route correctly. (990dd3d5)
  • Zotero Connector PDFs: Report filesEditable and resolve attachments so saves are not metadata-only (e.g. IEEE stamp pages). (d22f47fe)
  • Nested papers under index folders: Index NOTES.md folders no longer hide nested paper units. (b7b62ac8)
  • Paper info on non-paper tabs: Switching to non-paper tabs no longer clears the paper info pane. (717a5bc0)
  • Library refresh after CLI import: Metadata and library views update after CLI imports / file events. (2e594f85)
  • Layout analysis stability: Same-line formula merge, two-column order, dual-label cleanup, and re-analyze from layout.json without ONNX. (c1800c4a, 6ace73e7, 8c5cefa7)

Resolved Issues

Issues closed in-range and already covered above are not repeated: #193, #198, #209, #215, #182, #201, #200, #203, #225, #218, #204, #219, #170 (region path), #148 (related UX).

Full changelog: v0.4.1...v0.5.0

Assets 32
Loading