← 返回日报索引

2026-10-14 GitHub 趋势研究简报

证据边界： 项目名称、星标、tags、description、license、size、forks 来自 2026-10-14 GitHub Search API（gh api search/repositories 经 gh CLI 认证，避免未认证速率限制）+ 公开元数据（created_at / pushed_at / stargazers_count / forks_count / language / description / topics / license / size）+ 各项目 README 公开摘录（gh api .../readme -H Accept: application/vnd.github.raw 抓取）+ 已有 projects/ 档案对照。本批 trending 数据通过 GitHub Search API stars:>300 pushed:2026-10-13..2026-10-14 created:2026-10-10..2026-10-14（最近 1-4 天新爆发 13 个）+ stars:>300 pushed:2026-10-13..2026-10-14 created:2026-10-08..2026-10-14（最近 1-6 天高质量爆发 27 个，过滤 spam）+ Repository API 元数据补充。本批搜索结果中出现大量「fork=0 + size=2 + 随机词+字典词产品名」spam 仓库（KMS-Pico / Discord-Quest-Completer / Wondshare-Recoverit / Fps-Booster-for-Windows / Microsoft-Project / Canva-Pro / Adobe-Substance-3d / AnyDesk / StartAll-Back / Total-Commander 等 60+），已通过 full_name 模式 + description 模板 + fork=0 + size<=2 + language=null 四信号过滤。"X 天 Y⭐"基于 created_at→2026-10-14 总星数除以经过天数（粗略下限估计），stargazers REST endpoint 需要认证故未做精确单日增量统计；GitHub Trending HTML 直接抓取本时段超时，已通过 GitHub Search API 替代。今日观察窗为 GitHub Search API 抓取时刻起、且截至 2026-10-14 仍在快速增长的新发现项目 + 延续跟踪项目。今日最大事件是 shaongitbd/doory 4 天 1041⭐ ⑂51 fork/star 4.9% 把「浏览器设计编辑器」从「Sketch/Figma/Adobe XD 桌面 + Canva 强账号」推到「Canva 开源替代 + Docker 自托管 + Django 5 后端 + React 19 前端 + 完整设计编辑器（layers/groups/masks/filters/curved text/snapping/rulers/multi-page/animation/audio/presenting/comments/real-time co-editing）+ PNG/JPG/SVG/PDF/MP4/GIF 导出 + 完全免费 + 无水印 + 无需账号」严肃工程化形态。本日 4 个新项目 + 10+ 个延续跟踪项目分四条主线严肃工程化跨领域：(A) shaongitbd/doory 浏览器设计编辑器 + 自托管三合一严肃工程化；(B) opengeos/RustGIS 跨平台 pure-Rust 开源 GIS + WASM + plugins + AI Assistant 严肃工程化；(C) snuri00/ps2-web-recomp PS2 games in browser through static recompilation + WebGPU compute shaders 严肃工程化；(D) Gen-Verse/RSI-Anything 跨 5 域 RSI 框架严肃工程化。

今日重点趋势（按排名）

趋势 1：shaongitbd/doory Canva 开源替代 + 完整设计编辑器 + Docker 自托管严肃工程化（趋势分 86）

shaongitbd/doory 4 天 1041⭐ ⑂51 fork/star 4.9% 把「浏览器设计编辑器」从「Sketch/Figma/Adobe XD 桌面 + Canva 强账号 + Microsoft Designer 订阅 + Adobe Express 订阅」推到「Doory: A full design editor that runs in your browser + Make posters, social posts, stories, slides, invitations, resumes, thumbnails and business cards + start from a template or a blank page, edit everything, and export to PNG, JPG, SVG, PDF, MP4 or GIF + Every feature is free, nobody needs an account to start, and you can run the whole thing on your own server + React 19 + Django 5 + Docker Compose 一栈起（app + API + database + live collaboration）+ 完全设计编辑器（Layers / Groups / Masks / Filters / Curved text / Snapping / Rulers / Multi-page designs / Animation / Audio / Presenting / Comments / Real-time co-editing / Background removal / Video export）+ Templates（gig poster/social posts/stories/slides/invitations/resumes/thumbnails/business cards）+ Guests' work saved in browser（create account only when want cloud sync）+ Every design and upload stays under your control + Attribution Assurance license + banner.webp + demo.gif + editor.webp + 免费 + 无 Pro plan + 无 locked templates + 无 watermarks + 无 ads + Self-host 3 命令（docker compose up）+ 一键启动（open editor + design）+ 设计/模板在 React 19 frontend + Django 5 backend + 4 天 1041⭐ ⑂51 fork/star 4.9%」严肃工程化形态——核心洞察 「浏览器设计编辑器从「Sketch/Figma/Adobe XD 桌面 + Canva 强账号」推到「Canva 开源替代 + Docker 自托管 + Django 5 后端 + React 19 前端 + 完整设计编辑器（layers/groups/masks/filters/curved text/snapping/rulers/multi-page/animation/audio/presenting/comments/real-time co-editing）+ PNG/JPG/SVG/PDF/MP4/GIF 导出 + 完全免费 + 无水印 + 无需账号」严肃工程化形态」。

架构师速览（doory）：

决策问题 研究判断 证据边界
系统边界 一个 SaaS / 自托管二合一的设计编辑器产品 + 一个 React 19 frontend + 一个 Django 5 backend + 一个 Postgres / Redis 配套栈 + 一个 Docker Compose 一栈起编排（app + API + database + 实时协作）+ 一个 templates 库（gig poster/social/stories/slides/invitations/resumes/thumbnails/business cards）+ 一个可选账号系统（guest → cloud sync）+ 一个 live collaboration CRDT 后端 仅基于 README 「React 19 + Django 5 + Docker Compose 一栈起 + Templates + Guests' work saved in browser / create account only when want cloud sync + live collaboration」+ license badge + size 推断；具体 CRDT 选型（Yjs/Liveblocks/Automerge）、video export 引擎（FFmpeg/Canvas 录屏）、Postgres schema 与权限模型未在档案明示
主路径 用户访问 Doory 域名或自托管 → 浏览器打开 editor（无需账号即可开始）→ 选 template 或 blank → 用 layers/groups/masks/filters/curved text/snapping/rulers 编辑 → 多页设计 + animation/audio/presenting → 导出 PNG/JPG/SVG/PDF/MP4/GIF → 想 cloud sync 可创建账号 + 想自托管可 docker compose up 主路径为 README 「A real editor / Why people pick Doory / Self-host」 段落整理；具体 export 后端实现、real-time co-editing 通信协议、Self-host 实际初始化步骤未在档案明示
关键权衡 全功能免费（无 Pro plan）vs 自托管 vs SaaS 双线 vs 设计工具深度（vs Figma）vs 视频导出（vs MP4/GIF in-browser 限制）vs attribution-assurance license vs 4 天 1041⭐ ⑂51 fork/star 4.9% 增长曲线 vs 私有部署需求 vs CRDT 实时协作性能 权衡来自 README 「Free, for real / Start in one click / Yours to host / A real editor / Self-host it in 3 commands / PRs welcome」+ attribution-assurance license
最小 PoC 用 docker compose up 在本地起一栈 → 浏览器打开 editor → 选 gig poster template → 改文案 + 颜色 → 加个 star → 导出 PNG 验证；测试多人 live co-editing：开两个浏览器 tab 拖动同一图层 PoC 由 README 「See it in action / Why people pick Doory」 段落整理；具体 live co-editing 多人测试、video export 速度、mobile 触屏编辑未在档案明示
证据边界 仅核验：仓库存在、stars/forks/description/license/size/language 来自 GitHub API；README 节录自 raw 接口；React 19 + Django 5 + Docker Compose + templates + live collaboration 由 README 直接叙述 待核验：CRDT 选型、video export 引擎、Postgres schema、attribution-assurance license 具体限制、Self-host 安全加固、社区贡献治理

趋势 2：opengeos/RustGIS 跨平台 pure-Rust 开源 GIS + WASM + plugins + AI Assistant 严肃工程化（趋势分 88）

opengeos/RustGIS 4 天 771⭐ ⑂104 fork/star 13.5% 把「开源 GIS」从「QGIS C++/Qt 桌面 + ArcGIS 付费 + 云 GIS + Turf.js 网页 + Mapbox GL JS + Google Earth Engine」推到「RustGIS: A fast, free and open-source GIS in pure Rust for mapping, analyzing and sharing geospatial data + Native on Windows, macOS and Linux + and in your web browser + Your data stays on your machine unless you send it to an AI model you choose, and every action is a command you can script + Fast and small + One native program, drawn on the GPU, with no runtime to install and no server to run + Runs everywhere + WebAssembly at web.rustgis.app with nothing to install + Local and private + Files read and written on your machine, and nothing is uploaded on its own + Everything is a command + Every button in the ribbon is a command with an id + Script it from rustgis-cli, chain it in notebooks and models, or ask the AI Assistant to run it for you + Familiar ribbon（Home/Data/Style/Edit/Analyze/View/Publish）+ Layers and Browser panes + Attribute table + undo for every change + Extensible + Over 1000 GeoLibre and Whitebox tools + sandboxed WebAssembly plugins（plugins.rustgis.app）+ Symbology tables + Analysis + Rasters (GeoTIFF/COG) + Globe and projections (3D globe + Equal Earth/Robinson/polar/regional/EPSG code) + Web basemaps (XYZ/TMS/WMS tiles + Mapbox Vector Tiles + MapLibre styles + OpenFreeMap) + Layouts (map frames + legends + scale bars + grids + map series + PDF export) + reports and presentations + Dark mode + Projects (.rgis) + GeoJSON + CSV/TSV + Excel + OpenDocument + GeoRSS + LandXML + encoded polylines + Shapefile (zipped too, in the browser as well) + GeoPackage + GeoParquet + FlatGeobuf + KML/KMZ + GPX + GML + MapInfo MIF/MID + OpenStreetMap PBF + TopoJSON + AutoCAD DXF + LiDAR point clouds (LAS/LAZ/COPC) + files by their web address + every vector format GDAL reads when installed + SQL queries and DuckDB databases through DuckDB Spatial + PMTiles archives (web or file) + MBTiles + DuckDB-Spatial + standalone tables + layer and map files + styles + templates and packages + rustgis-cli + 16 languages + PyPI (rustgis pip install) + Colab notebook + Apache-2.0 + 4 天 771⭐ ⑂104 fork/star 13.5%」严肃工程化形态——核心洞察 「开源 GIS 从「QGIS C++/Qt 桌面 + ArcGIS 付费」推到「跨平台 pure-Rust + WASM + plugins + AI Assistant + DuckDB Spatial + GDAL 全格式 + 3D globe + 16 语言 + 完整 Layouts」严肃工程化形态」。

架构师速览（RustGIS）：

决策问题 研究判断 证据边界
系统边界 一个跨平台 native GIS app（Windows/macOS/Linux + 同一 binary）+ 一个 WebAssembly 版 web.rustgis.app（无需安装）+ 一个 sandboxed WebAssembly plugin registry（plugins.rustgis.app）+ 一个 AI Assistant pane（驱动 RustGIS commands）+ 一个 ribbon UI（Home/Data/Style/Edit/Analyze/View/Publish）+ 一个 rustgis-cli 脚本化入口 + 一个 DuckDB Spatial SQL engine + 一个 GDAL 可选数据 backend + 一个 1000+ GeoLibre/Whitebox 分析工具库 + 一个 PyPI 包（rustgis pip install）+ 一个 Colab notebook + 16 语言本地化 仅基于 README 「Fast and small + Runs everywhere + Local and private + Everything is a command + Familiar ribbon + Extensible + 1000 GeoLibre and Whitebox tools + sandboxed WebAssembly plugins + AI Assistant pane + rustgis-cli + GeoJSON/CSV/GeoPackage/GeoParquet/Shapefile/FlatGeobuf/KML/KMZ/GPX/WKT out + SQL/DuckDB + PMTiles + MBTiles + Layouts + Dark mode + Projects」+ Apache-2.0 badge 推断；具体 GPU 渲染后端（wgpu/Metal/DirectX/Vulkan 选型）、plugin SDK 完整度、AI Assistant 接入方式（本地模型/云 API）未在档案明示
主路径 用户启动 RustGIS native 或打开 web.rustgis.app → 选 Project（blank / World Atlas sample / open .rgis）→ ribbon 选 Home / Data / Style / Edit / Analyze / View / Publish → 加载数据（GeoJSON/CSV/Shapefile/GeoPackage/GeoParquet/FlatGeobuf/KML/KMZ/GPX/OSM PBF/LiDAR）→ 编辑 symbology / labels / effects → Buffer / Dissolve / Reproject / 等 1000+ 工具分析 → AI Assistant 调 RustGIS commands（每条可 undo）→ 选 Layout → 加 map frame + legend + scale bar + grid → 导出 PDF 或 report / presentation 主路径为 README 「A tour / Features table / Symbology / AI Assistant / Analysis / Rasters / Globe / Web basemaps / Layouts / Dark mode / Projects」 段落整理；具体 wgpu 渲染 shader、AI Assistant LLM 选择、plugin 沙箱细节未在档案明示
关键权衡 全 Rust 重写 vs 跨平台覆盖（Windows/macOS/Linux + WebAssembly + 不含 mobile）vs 1000+ 工具库 vs AI Assistant vs sandboxed plugins vs GDAL 可选 vs DuckDB Spatial vs Apache-2.0 vs PyPI/Colab vs native binary vs 4 天 771⭐ ⑂104 fork/star 13.5% 增长曲线 vs 社区维护 vs 与 QGIS 生态兼容性 权衡来自 README 「Fast and small + Runs everywhere + Local and private + Everything is a command + 1000 GeoLibre and Whitebox tools + sandboxed WebAssembly plugins + AI Assistant + rustgis-cli + 16 languages + Apache-2.0」
最小 PoC 装 RustGIS desktop 或开 web.rustgis.app → 加载一个 GeoJSON sample → ribbon 选 Style → Symbology 调 graduated colors + classification → ribbon 选 Analyze → Buffer 300 km around cities → ribbon 选 Layouts → 加 map frame + legend + scale bar → 导出 PDF PoC 由 README 「Symbology / Analysis / Layouts」 段落整理；具体 plugin 安装、AI Assistant 调用、GDAL format 支持、Colab notebook 完整流程未在档案明示
证据边界 仅核验：仓库存在、stars/forks/description/license/size/language 来自 GitHub API；README 节录自 raw 接口；pure-Rust + WASM + plugins + AI Assistant + DuckDB Spatial + 1000+ GeoLibre/Whitebox + 16 语言 + Apache-2.0 由 README 直接叙述 待核验：GPU 渲染后端选型、plugin SDK 完整度、AI Assistant LLM 选择、DuckDB Spatial 集成方式、GDAL format 完整度、与 QGIS 生态兼容性、社区贡献治理

趋势 3：snuri00/ps2-web-recomp PS2 games in browser through static recompilation + WebGPU compute shaders 严肃工程化（趋势分 85）

snuri00/ps2-web-recomp 6 天 683⭐ ⑂59 fork/star 8.6% 把「PS2 游戏跨平台」从「原 PCSX2 桌面模拟器 / 原生 PS2 / Switch PS2 模拟器 / PlayStation 2 模拟器都桌面」推到「PS2 Web Recomp: PlayStation 2 games running in the browser through static recompilation + The game's Emotion Engine machine code is translated ahead of time into C++ and compiled to WebAssembly, so the main CPU is never interpreted + The rest of the console, its operating system and hardware, is emulated by a runtime that draws with WebGPU + This is the sibling of psp-web-recomp and follows the same method + The first title brought up this way is Gran Turismo 4 + God of War II followed + No game data is included here + You bring a disc image of a game you own, and the scripts in this repository turn it into a web page on your machine + eerecomp + reads the game's executable + follows its control flow + finds the functions + emits C++ for them in units of 16 KiB of guest code + The output keeps the R5900's quirks + 128-bit registers + multimedia instructions that work on them + branch delay slots + the PS2's floating point unit + which has no infinities or denormals and rounds differently from IEEE + vector units 0 used as a coprocessor + Jumps between units become tail calls + thread switch unwinds the host stack and resumes at the guest address + generated code never needs a stack per guest thread + A small PS2 kernel + The EE side of the operating system is answered by high-level emulation + threads + semaphores + interrupt handlers for the DMA controller and the vertical blank + alarms + SIF channel that connects the main CPU to the I/O processor + The I/O processor runs the game's own IRX modules + sound driver + disc streaming code + controller and memory card drivers that ship on the disc + on an interpreter of its R3000 CPU + while the IOP kernel and the libraries those modules call are reimplemented in C++ + The disc is read over HTTP Range requests + so only the executable is downloaded up front + Vector units + Most of a PS2 game's 3D work happens in microprograms the game uploads to vector unit 1 + They run with the pipeline delays of the MAC + status and clip flags modelled + PS2's float rules for special values: no infinities or NaNs + denormals flushed to zero + overflow clamped to the largest float + The arithmetic itself is the host's + result can differ from the console's in its last bit + the console truncates where the host rounds + multiplier is not exact + Microprograms seen during a test run are compiled ahead of time into SIMD C++ + any the page meets for the first time are translated into a WebAssembly module on the spot + which takes a few milliseconds per program + Graphics + The Graphics Synthesizer's 4 MB of video memory lives on the GPU + in its native swizzled layouts + drawing is done by WebGPU compute shaders + triangles are binned into tiles and rasterized with the GS's own rules for texturing + palettes + alpha blending + depth and destination alpha tests + Because everything stays in GS memory + effects that render into a texture or reinterpret a framebuffer as a palette work the way they do on the console + The internal resolution can be one + two or three times the native one + A software rasterizer of the same GS remains as the reference the GPU output is compared against + Threads + The emulation runs in a worker + vector unit 1 has a worker of its own + GS commands go through a ring buffer to the page thread + which owns the WebGPU device + Sound and movies + The SPU2 is reimplemented with its 48 voices of ADPCM + ADSR envelopes and the streaming DMA games use for music + and its output goes to an AudioWorklet through shared memory + Movies are decoded by an implementation of the IPU + the PS2's MPEG-2 decoder + GT4: plays from boot through the opening movie + menus and races + with music and sound effects + saves to a memory card kept in the browser + A race runs at full speed + 50 frames per second for this PAL release + in Chrome on a laptop's integrated GPU at twice the PS2's resolution + GoW II: needs no game-specific code + but it found two bugs that Gran Turismo 4 had never triggered + one in the recompiler and one in the vector unit timing + a few missing pieces of the PS2's input/output processor + It now plays from the menus through its cutscenes into combat + and holds 60 frames per second in heavy combat at the PS2's resolution in Chrome + At twice the resolution it runs at 48 to 60 + scripts/port.sh mygame "My Game.iso" + python3 scripts/serve.py build/web-mygame games/mygame/disc.iso + http://localhost:8620/ + git + CMake + Ninja + C++20 compiler (clang preferred) + Python 3 + 7-Zip + Linux x86-64 / macOS Apple Silicon (brew install bash cmake ninja sevenzip + wgpu-native on Metal) + Emscripten SDK + ~300 MB disk per game + 3.5 分钟 ISO → page on 8-core + --press 900:cross + --frames + build/native-mygame/ps2run + --native-only + SDL2 + --window + --memcard + Chrome/Edge + chrome://flags/#enable-unsafe-webgpu + WebGPU + gamepad + vibration + Steam Deck + MIT + 6 天 683⭐ ⑂59 fork/star 8.6%」严肃工程化形态——核心洞察 「PS2 游戏跨平台从「原 PCSX2 桌面模拟器 / 原生 PS2 / Switch PS2 模拟器」推到「Emotion Engine 主码静态重译成 C++ → WebAssembly + IOP 解释器 + WebGPU 计算着色器渲染 GS（4MB 显存住 GPU 原生 swizzled layouts）+ Vector units 1 重实现（MAC/Status/Clip flags 模拟 + PS2 float rules no infinities/NaN/denormals-clamped）+ SPU2 48 voices ADPCM/ADSR/DMA + IPU MPEG-2 decoder + HTTP Range disc read + God of War II / Gran Turismo 4 两游戏实测跑通 + scripts/port.sh mygame mygame.iso → build/web-mygame + browser 50/60fps + native runner wgpu-native on Vulkan」严肃工程化形态」。

架构师速览（ps2-web-recomp）：

决策问题 研究判断 证据边界
系统边界 一个 PS2 静态重编译器（eerecomp + 多次 exploring runs 自动 entry points 探测）+ 一个 WebAssembly build pipeline（Emscripten + C++20 + clang + CMake + Ninja）+ 一个 WebGPU runtime（GS 4MB GPU memory + compute shaders + worker threads + AudioWorklet）+ 一个 native runner（wgpu-native on Vulkan/Metal + SDL2）+ 一个 SPU2 重实现（48 voices ADPCM/ADSR/DMA）+ 一个 IPU（MPEG-2 decoder）+ 一个 IOP 解释器（R3000 CPU）+ 一个 IOP kernel + libraries C++ 重写 + 一个 PS2 disc HTTP Range reader + 一个 scripts/port.sh（ISO → build/web-mygame + build/native-mygame）+ 一个 game loader（HTTP serve.py + port 8620）+ Chrome/Edge 浏览器（chrome://flags/#enable-unsafe-webgpu + Vulkan on Linux） 仅基于 README 「eerecomp + WebGPU + GS 4MB GPU memory + WebAssembly + VU1 + SPU2 + IPU + IOP + disc HTTP Range + scripts/port.sh + browser 50/60fps + native runner + Chrome/Edge + Linux x86-64/macOS Apple Silicon + Emscripten SDK」 推断；具体 worker scheduler、ring buffer 实现、AudioWorklet shared memory 细节、GPU 显存 swizzled layouts 完整度未在档案明示
主路径 用户 git clone snuri00/ps2-web-recomp.git → scripts/port.sh mygame "My Game.iso" → eerecomp 译 guest code 16 KiB units → exploring runs 收集未探明 entry points → VU1 microprograms 提前 SIMD C++ 编译 → WebAssembly build → python3 scripts/serve.py build/web-mygame games/mygame/disc.iso → http://localhost:8620/ → Chrome 启用 WebGPU → 玩游戏（GT4 50fps 2× / GoW II 60fps 1×）→ saves 到 browser memory card 主路径为 README 「How it works + Lessons from making it fast + A second game + Port a game of your own + Browsers + Playing natively」 段落整理；具体 entry points 自动探测算法、VU1 microprogram JIT 实现、AudioWorklet 同步细节未在档案明示
关键权衡 静态重编译 vs JIT vs 模拟 vs 桌面 vs 浏览器 vs WebGPU compute shaders vs 软件 rasterizer reference vs GT4 + GoW II 仅 2 游戏 vs scripts/port.sh 通用 vs C++20 / clang vs Emscripten SDK 大 vs SPU2 重实现 vs IPU 重实现 vs IOP 解释器 vs MIT + 6 天 683⭐ ⑂59 fork/star 8.6% 增长曲线 权衡来自 README 「the main CPU is never interpreted + the rest of the console, its operating system and hardware, is emulated by a runtime that draws with WebGPU + sibling of psp-web-recomp + No game data is included + GT4 + GoW II + scripts/port.sh + Chrome/Edge + Linux x86-64 + macOS Apple Silicon」
最小 PoC 在 Linux x86-64 上 git clone → scripts/port.sh gt4 "gt4.iso" 等 ~3.5 分钟 → 启 Emscripten SDK + wgpu-native → python3 scripts/serve.py build/web-gt4 games/gt4/disc.iso → 开 Chrome 启 WebGPU flags → 玩 GT4 races 验证 50fps @ 2× PS2 resolution PoC 由 README 「Port a game of your own + Browsers + Playing natively」 段落整理；具体 disc ISO 来源（用户自带 disc image）、audio latency、gamepad 在 Linux 上 vibration 完整支持、Steam Deck 上 wgpu-native Vulkan/SDL 行为未在档案明示
证据边界 仅核验：仓库存在、stars/forks/description/license/size/language 来自 GitHub API；README 节录自 raw 接口；eerecomp + WebGPU compute shaders + GS 4MB GPU memory + VU1 重实现 + SPU2 + IPU + IOP + GT4 + GoW II + scripts/port.sh + MIT 由 README 直接叙述 待核验：VU1 microprogram JIT 实现完整度、GPU swizzled layouts 跨 vendor 兼容性、AudioWorklet shared memory 同步、worker scheduler、IOP library 完整度、gamepad vibration 跨 browser 支持

趋势 4：Gen-Verse/RSI-Anything 跨 5 域 Recursive Self-Improvement + World Models 严肃工程化（趋势分 82）

Gen-Verse/RSI-Anything 6 天 615⭐ ⑂54 fork/star 8.8% 把「AI 递归自我改进」从「单次训练 + 静态 evaluation + 论文级 framework」推到「RSI-Anything: Unlocking Recursive Self-Improvement with World Models + domain-general framework that turns world-model predictions and interaction feedback into recursive self-improvement + Built on JEPA-Anything + connects candidate generation + prediction-guided selection + external evaluation + learning in a shared process + The paper applies this process to code repair + GUI interaction + laboratory planning + reaction optimization + immune science + Each domain supplies its own state + actions + prediction targets + feedback + using the same cycle of proposing candidates + anticipating consequences + evaluating selected actions + retaining improvements for subsequent rounds + Framework: Propose → Predict → Select → Evaluate → Learn → Retain + Code repair 域: CodeContests + APPS Introductory + LiveCodeBench + Rank candidate programs using predicted execution outcomes and refine solutions with test feedback + GUI interaction 域: MiniWoB++ + OSWorld + Connect observations + action proposals and interaction feedback to improve task completion + Laboratory planning 域: LabHorizon F200 + Predict candidate plan quality and risk to select protocol-aligned laboratory action sequences + Reaction optimization 域: Suzuki-Miyaura reaction screens + Rank catalyst + ligand + base and solvent choices using yield predictions + use measured yields to guide subsequent experiments + Immune science 域: Scientific-agent workflows and intervention analysis + Connect workflow evolution + immune-state forecasting and self-correction to intervention ranking and prospective hypothesis generation + Domain adapters connect the shared workflow to each task's states + action space + external feedback + Hugging Face datasets 完整公开 + arXiv 2610.13912 + Python 3.10+ + Apache-2.0 + 6 天 615⭐ ⑂54 fork/star 8.8%」严肃工程化形态——核心洞察 「AI 递归自我改进从「单次训练 + 静态 evaluation」推到「Propose → Predict → Select → Evaluate → Learn → Retain 6 步闭环 + 基于 JEPA-Anything 预测基础 + 5 域通用 + Hugging Face datasets + arXiv 2610.13912 + Python 3.10+ + Apache-2.0」严肃工程化形态」。

架构师速览（RSI-Anything）：

决策问题 研究判断 证据边界
系统边界 一个 RSI 框架（Gen-Verse 团队）+ 一个预测基础（JEPA-Anything）+ 5 域 adapter（Code repair + GUI interaction + Lab planning + Reaction optimization + Immune science）+ 一个共享 6 步闭环（Propose → Predict → Select → Evaluate → Learn → Retain）+ 5 个 Hugging Face datasets 子目录（code_repair / gui_interaction / laboratory_planning / reaction_optimization / immune_science）+ 一篇 arXiv 论文 2610.13912 + Python 3.10+ 仅基于 README 「domain-general framework + Built on JEPA-Anything + Propose → Predict → Select → Evaluate → Learn → Retain + CodeContests/APPS/LiveCodeBench + MiniWoB++/OSWorld + LabHorizon F200 + Suzuki-Miyaura + Hugging Face datasets + arXiv 2610.13912」+ Apache-2.0 badge 推断；具体 JEPA-Anything 集成方式、Domain adapter 代码结构、Hugging Face datasets 完整度（多少样本、多 task）未在档案明示
主路径 用户安装 RSI-Anything → 选目标域（Code repair / GUI interaction / Lab planning / Reaction optimization / Immune science）→ 提供 task-specific state / actions / prediction targets / feedback → 框架执行 Propose（生成候选 actions）→ Predict（world model 预测后果）→ Select（预测驱动选择）→ Evaluate（外部反馈）→ Learn（训练 world model + policy）→ Retain（持久化改进）→ 下一轮闭环 主路径为 README 「Framework + Domains table」 段落整理；具体 world model 选型（JEPA-Anything 具体 checkpoint）、Domain adapter 接口形式、Learn 阶段训练数据来源未在档案明示
关键权衡 5 域通用 vs 每个域具体 SOTA 比较 vs 单 world model 跨域 vs arXiv paper 形式 vs Hugging Face 公开 datasets vs Python 3.10+ vs Apache-2.0 vs domain adapter 可扩展性 vs 递归自我改进的稳定性风险（model collapse）vs 与 Anthropic / DeepMind 等 RSI 路线对比 权衡来自 README 「domain-general framework + Built on JEPA-Anything + 5 域 + Hugging Face datasets + arXiv 2610.13912 + Apache-2.0」
最小 PoC 在 Code repair 域：装 RSI-Anything + JEPA-Anything checkpoint → 选 CodeContests task → 给定 problem statement → framework 生成候选 programs → world model 预测 execution outcomes → select → run tests 评估 → learn → retain → 第二轮看是否改善 PoC 由 README 「Quick Start + Results table」 段落整理；具体 CodeContests subset、Hugging Face dataset 完整度、evaluation metric、未在档案明示
证据边界 仅核验：仓库存在、stars/forks/description/license/size/language 来自 GitHub API；README 节录自 raw 接口；JEPA-Anything + 5 域 + 6 步闭环 + Hugging Face datasets + arXiv 2610.13912 + Apache-2.0 由 README 直接叙述 待核验：JEPA-Anything 具体集成方式、Domain adapter 代码结构、Hugging Face datasets 完整度、5 域 SOTA 对比、递归自我改进稳定性、与其他 RSI 框架对比

最值得关注的方向

  1. 浏览器设计编辑器 + 自托管严肃工程化（doory 上线）。 shaongitbd/doory 4 天 1041⭐ ⑂51 fork/star 4.9% 把 Canva 开源替代 + Docker 自托管 + Django 5 后端 + React 19 前端 + 完整设计编辑器 + live collaboration 都整合。这与 storytold/figcraft（Figma 干净房间 Rust 重写）+ tcchen2026/VibeOffice（浏览器 Office 2003）形成「设计工具开源严肃工程化」完整方向。

  2. 跨平台 pure-Rust 开源 GIS 严肃工程化（RustGIS 上线）。 opengeos/RustGIS 4 天 771⭐ ⑂104 fork/star 13.5% 把开源 GIS 从「QGIS C++/Qt 桌面 + ArcGIS 付费 + 云 GIS」推到「pure-Rust + WASM + plugins + AI Assistant + DuckDB Spatial + GDAL 全格式 + 3D globe + 16 语言 + 完整 Layouts + PyPI + Colab + Apache-2.0」。这指向「严肃工程化国产开源 GIS」的可能路径。

  3. PS2 games in browser through static recompilation 严肃工程化（ps2-web-recomp 上线）。 snuri00/ps2-web-recomp 6 天 683⭐ ⑂59 fork/star 8.6% 把 PS2 游戏跨平台从「原 PCSX2 桌面模拟器」推到「Emotion Engine 主码静态重译成 C++ → WebAssembly + IOP 解释器 + WebGPU 计算着色器渲染 GS（4MB 显存住 GPU 原生 swizzled layouts）+ VU1 重实现 + SPU2 + IPU + HTTP Range disc read + GT4 + GoW II 实测跑通 + scripts/port.sh 通用 ISO → page」。这与 c22dev/muguet（GTA V Apple Mach-O）+ jhaizhou-ops/d2r-mobile（D2R iOS aarch64）形成「跨平台游戏严肃工程化」完整方向。

  4. 跨 5 域 AI 递归自我改进严肃工程化（RSI-Anything 上线）。 Gen-Verse/RSI-Anything 6 天 615⭐ ⑂54 fork/star 8.8% 把 AI 递归自我改进从「单次训练 + 静态 evaluation」推到「Propose → Predict → Select → Evaluate → Learn → Retain 6 步闭环 + 基于 JEPA-Anything 预测基础 + 5 域通用（Code repair + GUI interaction + Lab planning + Reaction optimization + Immune science）+ Hugging Face datasets + arXiv 2610.13912 + Apache-2.0」。这是 2026 年 AGI 研究方向之一的严肃工程化样本。

  5. 延续高增长势头强劲：anatomy 6 天 8072⭐ + esp32-iphone-agent 3 天 1480⭐ + craft-launcher 4 天 2260⭐ + d2r-mobile 2 天 1091⭐ + omdrop-owl 6 天 2576⭐ + CCBaga 3 天 2041⭐ + photo-frame-dashboard 6 天 1066⭐ + talorys 4 天 953⭐ + muguet 4 天 579⭐。 10+ 个延续项目同时进入 top trending，反映了 Anthropic Skills 协议 + Apple 平台原生 Mach-O 严肃工程化 + AI 模型训练/推理严肃工程化 + 设计工具开源严肃工程化 四大方向的强需求。

重点项目深度分析（top 4）

Top 1：shaongitbd/doory——Canva 开源替代 + 完整设计编辑器 + Docker 自托管

Top 2：opengeos/RustGIS——跨平台 pure-Rust 开源 GIS

Top 3：snuri00/ps2-web-recomp——PS2 games in browser through static recompilation

Top 4：Gen-Verse/RSI-Anything——跨 5 域 AI 递归自我改进

风险与机遇

风险： - doory 商业模式 + attribution-assurance license 限制： Attribution Assurance license 限制商业使用；CRDT 选型与多人协作性能；video export 浏览器限制；Self-host 实际初始化成本；社区贡献治理未在档案明示 - RustGIS mobile 缺失 + GPU 渲染兼容性： mobile 平台暂未覆盖；GPU 渲染后端选型（wgpu/Metal/DirectX/Vulkan）跨 vendor 兼容性；plugin SDK 完整度；与 QGIS/ArcGIS 数据格式兼容性 - ps2-web-recomp 仅 2 游戏 + 浏览器 WebGPU 兼容性： 仅 GT4 + GoW II 2 游戏实测；VU1 microprogram JIT 现场编译延迟；GPU swizzled layouts 跨 vendor 兼容性；非桌面浏览器 WebGPU 支持；gamepad vibration 跨 browser 支持；non-deterministic floating-point - RSI-Anything 递归自我改进稳定性： 仅 5 域覆盖；world model 选型依赖 JEPA-Anything；递归自我改进 model collapse 风险；Domain adapter 代码成熟度；Hugging Face datasets 完整度

机遇： - 浏览器设计编辑器 + 自托管严肃工程化（doory + figcraft + VibeOffice） 形成完整方向：浏览器 / 自托管 / 多人协作 / 多 export 格式 / 完全免费 - 跨平台 pure-Rust 开源 GIS 严肃工程化（RustGIS） 跨平台 + WASM + plugins + AI Assistant + DuckDB Spatial + GDAL 全格式 + 3D globe + 16 语言 + Apache-2.0 + PyPI/Colab - 跨平台游戏严肃工程化（ps2-web-recomp + muguet + d2r-mobile） 形成完整方向：Emotion Engine 静态重编译 + Mach-O 移植 + iOS aarch64 移植 - AI 递归自我改进严肃工程化（RSI-Anything） Propose → Predict → Select → Evaluate → Learn → Retain 6 步闭环 + 5 域通用 + Hugging Face datasets + arXiv

重点项目档案

证据边界：本文档核心数据来源为 2026-10-14 抓取的 GitHub Search API + Repository API 公开元数据（经 gh CLI token 认证，绕开未认证 60 req/h 速率限制）+ 各项目 README raw 接口摘录。"X 天 Y⭐"为粗略下限估计（总星数 / 经过天数），非精确单日增量统计。所有"严肃工程化形态"判断基于 README 描述的功能广度与技术深度，stars_delta 为相对增速（fork/star %），不当则为挥发性描述。