OPEN SOURCE OBSERVER
从项目热度走向工程判断：追踪开源工具、模型与开发者生态。
01 / CODE
| 排名 | 项目 | 语言 | 今日新增 | 总 Star |
|---|---|---|---|---|
| 01 | morluto/reaReverse engineer anything with agents, from app behavior down to native binaries. | TypeScript | +4,666 | 14,423 |
| 02 | sponsors/mattpocockSkills for Real Engineers. Straight from my .agents directory. | Shell | +1,406 | 279,449 |
| 03 | boykopovar/AnyPS5Tool for automatic PS5 executables porting to Linux and Windows | C++ | +2,725 | 10,108 |
| 04 | ayghri/i-have-adhdA skill to stop your coding agent from burying the answer. ADHD-friendly output. | Python | +620 | 55,032 |
| 05 | sponsors/cathrynlaveryEditorial diagram design for Claude Code, Codex, GitHub Copilot, Factory Droid, and Pi. 42 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop. | HTML | +828 | 44,880 |
| 06 | addyosmani/agent-skillsProduction-grade engineering skills for AI coding agents. | JavaScript | +693 | 102,718 |
| 07 | EpicGames/raddebuggerA native, user-mode, multi-process, graphical debugger. | C | +82 | 7,834 |
| 08 | sponsors/thedotmackPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More | TypeScript | +578 | 97,643 |
| 09 | manaflow-ai/cmuxOpen source Ghostty-based macOS terminal with vertical tabs and notifications for AI coding agents. Built for multitasking, organization, and programmability. | Swift | +96 | 27,808 |
| 10 | sponsors/trycuaScale computer-use 2.0 with open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation. | Rust | +229 | 28,703 |
02 / MODELS
| 排名 | 模型 | 任务 | 收藏 | 下载 |
|---|---|---|---|---|
| 01 | Z-Image-Turbo | 文本生成图片 | 2,570 | 838,732 |
| 02 | MiniMax-H3 | — | 2,060 | 197,909 |
| 03 | Qwen3.8-27B | 视觉多模态理解 | 1,568 | 616,940 |
| 04 | DeepSeek-R1 | — | 1,398 | 927,458 |
| 05 | Qwen3.8-Flash-Next | 视觉多模态理解 | 1,362 | 79,624 |
| 06 | Qwen-Image-2.1 | 文本生成图片 | 1,229 | 43,524 |
| 07 | Qwen3-Coder-480B-A35B-Instruct | 文本生成 | 1,100 | 138,539 |
| 08 | IndexTTS-2 | 语音合成 | 998 | 212,518 |
| 09 | chatglm3-6b | — | 832 | 1,591,159 |
| 10 | Qwen3.8-2.4T-A95B | 文本生成 | 781 | 2,156 |
DEVELOPER VIEW
如果你在搭 Agent 工作流，mattpocock 和 addyosmani 的技能集可以直接抄结构——它们代表的是「什么样的话术能让 Agent 稳定产出可合并的代码」。i-have-adhd 值得单独试，成本极低但能立刻改善日常交互体验。
thedotmack 的持久上下文方案解决的是真实痛点：每天重新解释项目背景。它支持多宿主这一点尤其关键，避免被单一工具锁定。trycua 的价值在于它是基础设施而非玩具，如果你要做 computer-use 的评测或数据生成，这是少数提供跨系统机群能力的开源选项。
模型侧，Qwen3-Coder-480B-A35B 对需要本地或可控代码模型的团队是重要备选；IndexTTS-2 在语音克隆与合成上保持了可玩性和可用性。
WORTH A STAR
morluto/rea — 首日近 4700⭐ 不是偶然。逆向工程长期缺乏好的自动化路径，Agent 驱动的行为分析到二进制分析是真正的新范式，且目前几乎没有成熟竞品。
thedotmack 的持久上下文项目 — 跨会话记忆是每个重度 Agent 用户都会撞上的墙，它把这层做成了可插拔组件，适配面广，落地阻力小。
sponsors/cathrynlavery 的图表设计技能 — 42 种图表、自包含 HTML+SVG、明确拒绝 Mermaid 的粗糙输出，这种「审美即功能」的技能正在成为 Agent 生态里被低估的一类资产。
SOURCES & METHOD
GitHub Trending（项目热度、语言、Star） · ModelScope（模型任务、收藏与下载）
抓取日期：2026-10-08。排名是当日快照，指标可能随平台更新而变化。