日报 · 2026-10-26
GitHub 新星 · 汇总昨日新收录项目
当日新增收录 40 个项目
覆盖日期：2026-10-26 · 北京时间
发布时间：2026/10/27 08:00:30
本期文章生成失败，后台将在稍后重试。下方为已保存的项目资料。
本期项目资料
收录数按本期首次达标收录的项目统计。
- JoasASantos/REA-PromptsBatchfile · 2,721 Stars
逆向工程提示词、软件分析
Stars 统计更新于 2026/10/27 07:55:32
查看原始资料
作者描述
REAP - Prompts for you to use REA (Reverse Engineer Anything), get the most out of it, and clone or break down any program
编报时的 AI 中文解读
为REA逆向工程工具集提供一套精选提示词，帮助AI编码代理分析软件工作原理，涵盖从原生二进制、固件、内核到游戏、Electron应用、APK和网站等多种目标。
编报时的 README 依据
# REA Investigation Prompts (REAP) — Reverse Engineering with your Coding Agent A curated collection of prompts for [REA (Reverse Engineer Anything)](https://rea.tools/) — the toolset that lets your AI coding agent inspect how software works, from native binaries, firmware and kernels to games, Electron apps, APKs and websites. Prompts are ready to paste into your agent. Replace everything in `<angle brackets>` and keep each request focused on **one feature** — narrow questions get evidence-backed answers. ## Star History
## Install as agent skills (recommended) `setup.sh` (macOS/Linux) or `setup.bat` (Windows) copies every prompt section into your coding agent's skills folder (`~/.hermes/skills`, `~/.claude/skills` or `~/.codex/skills`): ```bash ./setup.sh # auto-detect installed agents ./setup.sh hermes # or: claude-code | codex | all ``` ```bat setup.bat :: Windows ``` What you get:
Every prompt asks the agent to separate *confirmed facts* from *hypotheses* and to cite evidence — treat that as non-negotiable. --- ## 0. Install & verify ```text Install REA and connect it to this agent using: npx rea-agents@latest setup First, show me the installation plan and the configuration changes for my approval. After the install, verify the connection and list the available tools and workflows. ``` --- ## 1. Getting started ### First exercise — trace a CSV export ```text Download https://rea.tools/examples/notes-example.zip into a new folder and extract the contents.
| — | [Wildcard prompt](#wildcard-prompt-start-here) · [Prompt-writing tips](#prompt-writing-tips-why-these-work) · [Next steps](#next-steps-see-what-people-built-with-rea) | > ⚠️ **Use responsibly.** Only analyze software you own or are explicitly authorized to test. Anti-cheat, DMA, DRM, kernel and vulnerability prompts below are for **security research, interoperability, CTFs, bug bounties and testing the protection of your own software**. Bypassing protection of software you don't own, cheating in online games, or weaponizing a found vulnerability may violate law and terms of service.
- t3dotgg/tstackPython · 1,528 Stars
PR审查、PR合并、PR现代化
Stars 统计更新于 2026/10/27 07:35:49
查看原始资料
作者描述
The agent skills I use every day. Works with Claude Code and Codex.
编报时的 AI 中文解读
这是一套日常使用的代理技能集合，适用于Claude Code和Codex，可帮助处理PR审查、合并、现代化等任务。
编报时的 README 依据
cp -R tstack/skills/* ~/.claude/skills/ # Claude Code cp -R tstack/skills/* ~/.codex/skills/ # Codex ``` ## Requirements - An authenticated [GitHub CLI](https://cli.github.com) (`gh`). - `python3` for `babysit-pr`'s `pr-wait` script. - `council` and `sol-loop` need subagents on other providers' models. T3 Code does this with `delegate_task`. Otherwise you need a harness that can run those models. ## Make them yours These are starting points. Edit them. Put personal settings, like which models sit on your council, in your `CLAUDE.md` or `AGENTS.md`, so you can update the
- lyramakesmusic/neuron-explorerPython · 1,242 Stars
模型探索、实时推理、Claude集成
Stars 统计更新于 2026/10/27 07:35:49
查看原始资料
作者描述
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
编报时的 AI 中文解读
面向Qwen3.5-4B模型的神经元探索工具，通过推动每个MLP神经元并保留所有rollout，以3D视图展示模型内部结构，支持交互式浏览、实时推理和Claude集成。适用于模型可解释性研究者和对LLM内部机制感兴趣的技术人员。
编报时的 README 依据
# neuron explorer Steering-first neuron explorer for Qwen3.5-4B. Every MLP neuron in a layer is pushed up and down on a few prompts, every rollout is kept, and the explorer lays the results out over a 3D view of the model.  ## Explorer uv sync uv run python explorer/server.py # http://localhost:8791 Reads sweep runs from `runs/` and per-layer stats from `stats/`, both produced by the sweep below; the caches in `explorer/cache/` ship with the repo. Loads Qwen3.5-4B locally (MPS, CUDA or CPU) for live
tools and a shell in your home folder. Its prompt is `explorer/agent_prompt.md` plus `explorer/agent_methodology.md`, and `explorer/agent_prompt.local.md` when present (untracked, for your own additions); `EXPLORER_CLAUDE_MODEL` picks the model. The server answers only pages and tools on this machine. ## Sweep Runs on a CUDA 13 GPU pod with the steering fork of vLLM ([RhizoNymph/vllm](https://github.com/RhizoNymph/vllm)). bash sweep/setup_pod.sh # fork, deps, model; writes /workspace/env.sh
python sweep/neuron_stats.py --layers 22 --tokens 2000000 --out stats/q35-4b python sweep/steer_sweep.py --stats stats/q35-4b_L22.npz --design main_rel --dose-unit rel \ --neurons all --random-controls 256 --run q35-4b-L22 sweep/run_depth.sh "4 9 13 17" 384 64 # random-subset sweep across several layers `main_rel` doses are fractions of the median residual norm leaving the layer: {who, snake, orange} × {−0.7, +0.5, +0.8} and {free} × {+3}, plus one unsteered baseline per prompt. | script | output | |---|---|
- TecnaGamer/BeamNG.EarthPython · 1,164 Stars
全球驾驶、真实天气模拟、多人联机
Stars 统计更新于 2026/10/27 07:46:21
查看原始资料
作者描述
A 1:1 scale streamed Earth for BeamNG.drive: real elevation and satellite imagery, floating origin, LOD streaming from a companion server
编报时的 AI 中文解读
为BeamNG.drive提供1:1比例、基于真实高程和卫星影像的流式地球地图模组，支持全球驾驶、真实时间天气、季节、重力、AI交通和多人联机。
编报时的 README 依据
# BeamNG.Earth A 1:1 scale, streamed Earth for [BeamNG.drive](https://www.beamng.com/): drive anywhere on the planet on real elevation data and satellite imagery, with distance-based level of detail and the real curvature of the Earth. Status: beta (see VERSION). Tested on BeamNG.drive 0.40.3 on Linux (Vulkan) and the Steam Deck. See [docs/ROADMAP.md](docs/ROADMAP.md) for what is coming. **Community:** join the [Discord server](https://discord.gg/tgf) and head to the #beamng-earth channel to talk about
- **Traffic**: stock BeamNG traffic and police can be switched on, and parked cars line the streets and fill parking lots (stock BeamNG parking). **Multiplayer (BeamMP, in testing)** - Play the real world together on a BeamMP server (every player runs the BeamNG.Earth app, which starts BeamMP with the game): BeamNG.Earth servers have their own **BeamNG.Earth** tab in BeamMP's server browser, players who join are sent to the others, cars show in the right place and with their crash damage in everyone's world, and other players show on the world map (see "Multiplayer" below).
- Easy setup: one download per system (Windows .exe, Linux AppImage) that installs the mod and starts the server and the game; from the source, `install.py` installs everything and `start.py` keeps the mod in sync. - Runs on a Steam Deck (aim: 30 fps on the Handheld graphics preset). - Uses your graphics card for some of the world server's work (colour matching of the photos, compressing the distant textures) when it can, so the server needs less of your CPU; without a usable one the CPU does it as before. - Offline play from the download cache (with a size cap and "clear cache").
- **Linux:** `BeamNG.Earth-<version>-x86_64.AppImage`. Make it executable (right-click > Properties > "Allow executing", or `chmod +x`). 2. **Run it** (with Steam running). A console window opens (like BeamMP's launcher); it copies the mod into the game, starts the world server and starts BeamNG.drive. Closing the game stops the server. - **Linux, native game:** started with the grass patch (`tools/gcpatch`) already applied, no Steam launch options needed. - **Linux, Windows build under Proton:** the AppImage sees the Proton setting and starts that build through
**You need:** BeamNG.drive (tested with 0.40.x) on Steam, free disk space for the map download cache (it grows as you visit places, up to 20 GB by default; settings page > "Download cache size" sets 5-50 GB), and an internet connection (the map is downloaded while you play). 1. **Download** the file for your system from the [Releases](https://github.com/TecnaGamer/BeamNG.Earth/releases) page: - **Windows:** `BeamNG.Earth-<version>.exe`. Windows may warn about an unknown app: **More info > Run anyway**.
- HazyResearch/ncmPython · 1,091 Stars
并行文本生成
Stars 统计更新于 2026/10/27 07:46:22
查看原始资料
作者描述
NCM: Next Concept Modeling
编报时的 AI 中文解读
Next-Concept Modeling (NCM) 是一个用于并行文本生成的两阶段框架，通过多尺度残差VQ分词器将序列表示为离散码的层次结构，并由Transformer从粗到细预测该层次，实现快速、并行的文本生成。适用于研究并行生成方法、评估模型性能的研究人员。
编报时的 README 依据
git clone https://github.com/HazyResearch/ncm.git cd ncm pip install -r requirements.txt ``` Python 3.10 or newer and a CUDA-capable PyTorch installation are recommended. ## Evaluation Download the checkpoints and evaluation inputs: ```bash hf download hazyresearch/ncm-checkpoints --local-dir artifacts hf download hazyresearch/ncm-datasets --repo-type dataset --local-dir data ``` Then run: ```bash python evaluate.py --config config/eval/<task>.yaml --checkpoint-dir artifacts/<task> --data-dir data/<task> # <task> is one of xsum, wmt14, sudoku-easy, sudoku-medium or sudoku-hard
python evaluate.py --config config/eval/<dataset>.yaml --checkpoint-dir artifacts/<dataset> # <dataset> is tinystories, owt or lm1b-<size> with <size> one of small, medium or large ``` This runs one generation shard per visible GPU, merges them, and prints ROUGE-1/2/L and generative perplexity for XSum; cased and lowercased sacreBLEU, chrF and length ratio for WMT14; solved, cell and valid-grid rates for Sudoku; and generative perplexity under GPT-2 large with unigram entropy, over 5 seeds of 256 unconditional samples, for TinyStories, OpenWebText and LM1B, the latter under the FLM scoring
- zhuyansen/awesome-claude-video-skills未标注语言 · 1,037 Stars
视频生成、视频剪辑、技能资源导航
Stars 统计更新于 2026/10/27 07:56:28
查看原始资料
作者描述
Open-source skills and toolkits that let Claude Code, Codex and other coding agents make video. 180 repos by type, each security-graded. English / 中文.
编报时的 AI 中文解读
这是一个精选的资源汇总列表，收录了能让Claude Code、Codex等编码代理生成视频的开源技能与工具包，共347个仓库，涵盖框架、宣传片、解说、剪辑、短视频、虚拟形象、动画等类型，并提供了安全评级和实测对比。
编报时的 README 依据
## Which one to install We ran 16 of these end to end (16 gave a result). This is what we would pick; the [full test](#tested) is below. - 🥇 **For a finished film fast: [explainroo](https://github.com/vincentsch/explainroo)** Two minutes from brief to a film with sound, the quickest of the sixteen, and it does not look like the default. - 🥈 **For a film with one idea running through it: [onetake](https://github.com/feitangyuan/onetake)**
<td align="center" valign="top" width="33%"><b>📣 Promo & demos</b><br><sub>68 repos</sub><br><br><a href="https://github.com/howseen-ai/claude-motion-design"><img src="assets/previews/howseen-ai__claude-motion-design.gif" width="260" alt="howseen-ai/claude-motion-design"></a><br><sub>Product launch films, ads and demo videos.</sub><br><a href="#type-promo"><b>View the list →</b></a></td>
# Awesome Claude Video Skills [中文](README.zh-CN.md) Open-source skills and toolkits that let **Claude Code, Codex and other coding agents make video**: HyperFrames, Remotion, motion graphics, editing, explainers, avatars. 347 repos, each one read and security-graded by [Agent Skills Hub (agentskillshub.top)](https://agentskillshub.top?utm_source=github&utm_medium=awesome-list). Live page with filters: **[https://agentskillshub.top/best/claude-video-skills/](https://agentskillshub.top/best/claude-video-skills/?utm_source=github&utm_medium=awesome-list)** · refreshed every 8 hours
2. An agent operates it: a skill, a plugin, an MCP server, or a toolkit written for the agent. 3. It has a README. Without one it cannot be graded. 4. At 50 stars or more it is listed on topic alone. Under 50 it must also clear a README quality bar (shows the result, one-command start, a concrete outcome, complete docs), and have 5 stars unless it names Claude Opus 5.5. The questions are answered by a decision model reading each README, not by hand. A repo near a cut-off can land on either side; open an issue if one is misfiled. <a id="tested"></a> ## 🧪 Tested end to end
**Know before you pick:** hyperframes (the most-starred renderer here took 23.5 minutes on a 15-second brief); motion-graphics (built for talking-head footage, which this brief did not have). Whichever you pick: you may not need one for a simple promo. Claude Code with no skill made a film that passed every check in 4.4 minutes. A skill earns its place with a look, a workflow, or footage to edit.
- impacifist/Windows-GPU-UNLOCKC · 758 Stars
实验性UEFI GPU解锁与Windows直接启动交接工具，包含针对特定硬件的配置与构建脚本，以及详细的启动和恢复指南。适用于研究或实验目的的技术用户。
Stars 统计更新于 2026/10/27 07:35:49
查看原始资料
作者描述
Windows GPU UNLOCK: RTX3090 UEFI port, Korean no-reset Windows boot guide, verified source and benchmark evidence
编报时的 AI 中文解读
实验性UEFI GPU解锁与Windows直接启动交接工具，包含针对特定硬件的配置与构建脚本，以及详细的启动和恢复指南。适用于研究或实验目的的技术用户。
编报时的 README 依据
| **Internal headless** | B650M/3090 FE or B660M/3070 Ti profile; monitors on RTX; display temporarily removed; no T/U/Enter | [Internal headless guide](docs/HEADLESS.en.md) | | **Headless USB** | X870/3090 Ti contributor-tested route; profile-specific wrapper and child on a FAT32 USB stick | [Headless USB guide](docs/HEADLESS-USB.en.md) | | Internal AMD-display manual / Auto | Original B650M profile, motherboard/iGPU display preserved; manual variant requires U | [Existing internal guide](docs/INTERNAL-BOOT.en.md) |
.\internal-boot\Configure.ps1 -UnlockDriveLetter U ` -WindowsEspGuid 'YOUR-WINDOWS-ESP-GUID' .\research\headless\build-unlock.ps1 ` -Clang 'C:\TOOLS\LLVM20\bin\clang.exe' ` -LldLink 'C:\TOOLS\LLVM20\bin\lld-link.exe' # Stage and append a separate F8 test entry; retain existing boot priority. .\internal-boot\Headless.ps1 -Assets '.\assets' -Register ``` Without an action switch, `Headless.ps1 -Assets '.\assets'` only stages files. Configuration and build do not register a boot entry. 1. Save work and fully power off. On the tested ASUS firmware, select
**F8 → GPU Headless UNLOCK - EXPERIMENTAL**. 2. **Wait without T, U or Enter.** Initial Esc within five seconds skips release and unlock and boots Windows directly. 3. The screen becomes black while the RTX GOP is removed. The child attempts once and the wrapper starts Windows directly. Do not reset between unlock and Windows. Confirm display recovery and [measure](docs/BENCHMARK.en.md). 4. Only after that F8 test succeeds, enable default automatic boot: ```powershell .\internal-boot\Headless.ps1 -EnableAutomatic -ConfirmedSuccess ```
| `b650m-3090fe` (default) | RTX 3090 FE / ASUS TUF GAMING B650M-PLUS WIFI / 3057 / Ryzen 7 7800X3D | Internal headless; [original maintainer's hardware evidence](evidence/headless-handoff/README.md) | | `b660m-3070ti` | MSI RTX 3070 Ti / ASUS TUF GAMING B660M-E D4 / 4003 / Core i7-12700F **without iGPU** | Internal headless, aggregate console; [contributor's report and implementation, PR #5](https://github.com/impacifist/Windows-GPU-UNLOCK/pull/5) |
select `-HardwareProfile b660m-3070ti` in both configuration and build as described in [compatibility](docs/COMPATIBILITY.en.md). X870 uses the separate USB route. Read the [full guide and recovery procedure](docs/HEADLESS.en.md) before changing boot settings. Use the exact supported hardware, Secure Boot disabled, a fully decrypted Windows C: with BitLocker protection off, and a connected keyboard. The tools do not disable encryption. Preserve the **existing Windows ESP and recovery partition**. Prepare a separate
GPT FAT32 **EFI System Partition of at least 1 GiB**, with a letter such as U:. Tools do not create, shrink, delete or format partitions. An ordinary FAT32 data partition alone is not the required ESP. You need Python 3.9+, clang/lld-link **20.1.8**, the pinned repository dependencies and five locally obtained [firmware assets](docs/ASSETS.en.md). Host integration tests additionally need MSVC/Windows SDK and private fixtures. From an administrator PowerShell in the repository root: ```powershell # Replace the GUID and tool paths with your own values.
| `x870-3090ti` | ASUS RTX 3090 Ti / Gigabyte X870 AORUS ELITE WIFI7 / F13d / Ryzen 7 9800X3D | **Headless USB only**; [contributor's PR #10](https://github.com/impacifist/Windows-GPU-UNLOCK/pull/10), [sanitized evidence](evidence/x870-3090ti-usb/README.md) | These results concern the exact builds and systems in their reports. The current cold-precheck hardening has host/build verification, **not a new physical boot or recovery test** on those machines. Contributor reports are identified separately from the original maintainer's run; they do not establish support for every card
of the same model or every BIOS. The Ryzen CPUs in the table have iGPUs, but their reported headless routes used no active AMD GOP. The latest tagged release, **v0.3.1, is a source-only pre-release**. This source tree has changes beyond that tag, including the USB route and X870 profile; those are not a newly published release. Build locally with your own partition GUIDs. Firmware assets and private, partition-specific EFI binaries are not bundled. ## Choose the correct boot path | Path | Display and input | Instructions | |---|---|---|
- andyhuo520/digital-human-kouboTypeScript · 664 Stars
数字人视频生成、视频剪辑、视频包装
Stars 统计更新于 2026/10/27 07:35:52
查看原始资料
作者描述
AI digital-human talking-head videos from one photo: Renoise MCP generation + 5 editing styles + 48 creator-grade packaging components. Agent skill for Claude Code / Codex / OpenClaw / WorkBuddy …
编报时的 AI 中文解读
这是一个AI数字人口播视频生成工具，通过一张照片和一句话即可自动生成完整的口播视频，包括脚本、唇形同步、配音、字幕和包装，适用于直播带货、产品推荐、教程讲解等场景。
编报时的 README 依据
The skill is a folder with a `SKILL.md` (the open Agent Skills format) plus Python / shell / Remotion scripts. To run the **whole** pipeline an agent needs three things: 1. **Load Agent Skills** (a skills folder or skill upload), 2. **Connect a remote MCP server** (Renoise — streamable HTTP, browser login, no API key), 3. **Run commands on your computer** (ffmpeg, Python, Node — for downloading, QA and editing). | Agent | Skills | Renoise MCP | Local commands | Status | |---|---|---|---|---|
cd skills && zip -r ../digital-human-skills.zip digital-human-koubo finance-explainer && cd .. ``` ### 3 · Connect Renoise MCP (AI video / image generation) Create an account at **[renoise.ai](https://renoise.ai)** first. Login is in the browser — you never paste an API key. | Agent | How | |---|---| | Claude Code | `/plugin marketplace add ArcoCodes/renoise-plugins-official` then `/plugin install renoise@renoise-plugins-official` | | Codex | `codex mcp add renoise --url https://renoise.ai/api/public/mcp` then `codex mcp login renoise` |
Latest official instructions: **[renoise.ai/plugin](https://renoise.ai/plugin)**. Check it worked by asking your agent: *"查一下我的 Renoise 余额 / check my Renoise balance."* ### 4 · Local tools (for downloading, QA and editing) | Tool | macOS | Windows | Linux | |---|---|---|---| | ffmpeg | `brew install ffmpeg` | `winget install Gyan.FFmpeg` | `sudo apt install ffmpeg` | | Python 3 + numpy + Pillow | `brew install python` then `python3 -m pip install numpy pillow` | `winget install Python.Python.3.12` | `sudo apt install python3-pip` |
bash ~/.claude/skills/digital-human-koubo/scripts/doctor.sh ``` The **text-behind-presenter** effect uses Apple Vision and needs **macOS**; everything else is cross-platform. ### 5 · Sound effects & placeholder music The repo ships **no third-party audio**. Generate an original set of 30 sound effects and a placeholder background track (pure code, no copyright): ```bash python3 ~/.claude/skills/digital-human-koubo/scripts/make_assets.py ```
reference/ launch-169 · tutorial-169 · multihost-169 assets/ style previews · packaging gallery · avatar reference finance-explainer/ Style A engine (episode.json beat sheet, research helpers) docs/ README images, GIF previews, posters ``` ## Known pitfalls (all handled by the skill) - Grok occasionally **burns subtitles** into landscape clips even when told not to → frame check + re-shoot that segment.
3. Only clone **your own** voice or a voice you have explicit permission to use. > Note: Grok generates lip movement from its own voice. Swapping in a different voice afterwards leaves ~100 ms lip offset (fine in picture-in-picture, visible in full-screen close-ups). For your own recording with exact lip-sync, use an audio-driven model such as OmniHuman (documented in the skill). --- ## Quick start Talk to your agent in plain language: ```text 用我的数字人 Skill 讲这段：……（paste your content）
- surendrapandar/minilyticsTypeScript · 619 Stars
网站分析、收入归因
Stars 统计更新于 2026/10/27 07:35:52
查看原始资料
作者描述
Opensource Revenue-first analytics for indie hackers & solo founders
编报时的 AI 中文解读
Minilytics 是一个面向独立开发者和个人站长的开源网站分析工具，重点追踪访客来源、行为以及哪些渠道带来了收入。它部署在 Cloudflare 免费套餐上，适合自托管分析。
编报时的 README 依据
# Minilytics Revenue-first web analytics for your own websites, inspired by DataFast. It shows where visitors came from (which tweet, article, Google page or AI assistant), what they did, and **which source made you money**. It runs on Cloudflare's free plan.  ## Sponsors <table> <tr> <td align="center" width="33%"> <a href="https://diskmate.app/?ref=minilytics-github"><img src="docs/sponsors/diskmate.svg" width="64" height="64" alt="DiskMate"></a><br> <a href="https://diskmate.app/?ref=minilytics-github"><b>DiskMate</b></a><br>
- **What they did**: pages, entry pages, exit links (outbound clicks), button clicks and custom goals, funnels - **Who paid**: a Journey list of every customer with their source, amount, time to convert, and a full visitor timeline (visit → pages → clicks → payment) - Click any row to filter the whole dashboard (for example "only visitors from the launch tweet") - Multiple websites, one login ## Deploy for free (about 10 minutes) Every person deploys **their own copy** on their own free Cloudflare account. Your data stays on your account, and you
npx wrangler secret put SESSION_SECRET # any long random text, e.g. the output of: openssl rand -hex 32 ``` Typing is hidden; that's normal. If asked *"There doesn't seem to be a Worker called minilytics… create it?"*, answer **Yes**. **4. Deploy** ```bash npm run deploy ``` This builds the dashboard, creates the database tables and deploys. It prints your link, e.g. `https://minilytics.<you>.workers.dev`. Open it, log in, add a website and copy its script tag. Each site's Settings
get your own dashboard with its own password. You need a free [Cloudflare account](https://dash.cloudflare.com/sign-up), [Node.js 20+](https://nodejs.org) and Git. Run these commands in a terminal on your computer. You only do this once; afterwards the app runs on Cloudflare. **1. Download the code and log in to Cloudflare** ```bash git clone https://github.com/surendrapandar/minilytics.git cd minilytics npm install npx wrangler login # opens your browser; click "Allow" ``` **2. Create the database** ```bash npx wrangler d1 create minilytics ```
| 100,000 Worker requests/day | About 100k tracked events per day across all sites. The script and dashboard files are static and don't count. | | 100,000 D1 rows written/day | About 10k pageviews per day | | 5M D1 rows read/day | A dashboard load reads roughly 10 times the sessions in the range, so with 10k visitors/month that's hundreds of loads a day | | 5 GB D1 storage | Years of data at side-project traffic | At 5–10k visitors a month you'll use a few percent of this. If you outgrow it, Cloudflare's Workers Paid plan is $5/month. ## Install on a website ```html
-d '{"transaction_id":"txn_1","amount":29.99,"currency":"USD","visitor_id":"<ml_vid cookie>","email":"jane@acme.com"}' ``` ## Accuracy notes No analytics tool is 100% accurate. These are the main gaps and how this project handles them: - **Ad blockers** block some analytics domains. Proxy the script through your own domain (Settings shows a Next.js rewrite). Payments are always recorded by the webhook, even when the visitor's browser blocked tracking. - **Safari** keeps cookies set by JavaScript for at most 7 days. A Safari visitor who returns after more than a week counts as new. The
- DrWhax/unfairplayGo · 464 Stars
IPA解密
Stars 统计更新于 2026/10/27 07:35:52
查看原始资料
作者描述
Decrypt IPA's on MacOS
编报时的 AI 中文解读
UnfairPlay 是一个独立的 Go 程序，用于在 macOS 上恢复 FairPlay 模型并解密个性化的 iOS IPA 文件。它不登录 Apple、不购买应用或下载 IPA，需要用户自行准备加密的 IPA 和已安装的应用。
编报时的 README 依据
# UnfairPlay UnfairPlay recovers a FairPlay model from an enrolled Mac and uses it to decrypt personalized iOS IPAs. It is a standalone Go program. It does not log in to Apple, buy apps, or download IPAs. You can use ipatool or another downloader for that part. ## Build ```sh go build -o unfairplay ./cmd/unfairplay ``` ## Usage First, sign in to the native Mac App Store. Install an iPhone or iPad app and open it once. Download an encrypted IPA with the same Apple Account. Collect the Mac state and matching IPA: ```sh sudo ./unfairplay collect \ --app /Applications/App.app \
--ipa App.ipa \ --output App.capture ``` Recover the model and validate it: ```sh ./unfairplay recover \ --capture App.capture \ --output App.recovery ``` Decrypt another IPA from the same enrollment: ```sh ./unfairplay decrypt \ --input AnotherApp.ipa \ --model App.recovery/fairplay-enrollment.bin ``` Test a model without keeping the decrypted IPA: ```sh ./unfairplay verify \ --input AnotherApp.ipa \ --model App.recovery/fairplay-enrollment.bin ``` Use `unfairplay doctor` to check a capture, IPA, or model. Use `unfairplay profile` to inspect a fairplayd binary.
## Support Verified profiles: `25G83`, `26A428`/`26A434`, and `26B118` compatibility mode. Unknown arm64e builds are accepted only if model recovery and full IPA validation succeed. macOS arm64 and Linux amd64/arm64 are supported. Windows is compile-tested but cannot yet enforce the required private-file permissions. The IPA, installed app, Mac state, hardware identity, and model must belong to the same enrollment. Account or selector mismatches are rejected. Keep capture and model files private. Do not share SIDB, SINF, hardware data,
- hevnsnt/CVEnginePython · 376 Stars
攻击面管理、漏洞验证、资产发现
Stars 统计更新于 2026/10/27 07:35:53
查看原始资料
作者描述
AI-first external attack surface assessment with continuous discovery, evidence-based vulnerability validation, and developer-agent MCP integration.
编报时的 AI 中文解读
cvENGINE 是一款免费开源的外部攻击面管理工具，能够自动发现并映射互联网资产，识别服务与版本，匹配已知漏洞，并通过只读检查验证风险，提供持续监控和AI辅助研判。适用于安全团队、渗透测试人员和漏洞赏金猎人。
编报时的 README 依据
<img src="docs/images/cvengine-tour.gif" alt="A tour of cvENGINE: the live dashboard, hosts ranked by score, a validated CVE match, the change timeline with screenshot diffs, and the asset graph" width="900"> </p> ## Why cvENGINE Attack surface tools find assets. Vulnerability scanners find flaws. Threat feeds find CVEs. cvENGINE runs all three as one continuous loop and hands you a verdict instead of a pile of alerts. It answers the one question that matters: **which of today's vulnerabilities are live on assets I own, and can I reproduce them?**
**Operate.** A live dashboard, an AI activity log, and a Changes page with side-by-side screenshot diffs. Rescan, pin, exclude, or archive any host, and record its owner and criticality. Organizations, projects, roles (admin, user, audit), an All projects view, two-step sign-in, an IP allowlist, and an audit log. A REST API, an MCP server, and an OpenAPI spec so agents work under the same permissions. A [bug bounty workbench](docs/BUG-BOUNTY.md) for hunters: browse HackerOne, Bugcrowd, Intigriti, and YesWeHack programs and scope, point the scanner at what is in scope, and draft reports from
- **Verdicts, not noise.** AI triages your findings, read-only checks reproduce what they can, and fixed rules mean the model can never wave away hard evidence. Every check records exactly what it ran. - **Nothing changes unseen.** New hosts, opened ports, version bumps, new certificates, login pages that appear, and front pages that change all land on a timeline. - **Scanning in minutes.** One Docker Compose stack with every scanner built in. No agents to install, no cloud account to open.
You need Linux with [Docker Engine](https://docs.docker.com/engine/install/) and the Compose plugin, plus `git`, `make`, and `python3`. 8 GB of RAM and 20 GB of disk are recommended. ```bash git clone https://github.com/hevnsnt/CVEngine.git cd CVEngine cp .env.example .env # set DOMAIN_NAME and replace every change-me make certs # HTTPS certificate for the built-in proxy make up # build and start (the first build takes about 30 minutes) make username # create your admin account ```
- Loopback, link-local, and cloud metadata addresses are never scanned, private ranges stay off until you allow them, and addresses on shared platforms such as CDNs are never port-scanned. - Injection attempts are blocked, and at the strike threshold the account is disabled and the source IP is blocked for 24 hours. - Each project's data is isolated, and stored secrets are encrypted. - Two-step sign-in, an IP allowlist, and an audit log protect access. Only scan systems you own or are authorized to test. ## Documentation | Doc | What it covers | | --- | --- |
its findings. Off until an admin enables it. Slack, Discord, Telegram, and Lark alerts if you don't want to monitor a dashboard. One switch pauses all scanning. Memory-aware scheduling keeps small servers healthy, and updates drain running work first or deploy automatically on every new commit. **Bring your own AI.** Anthropic, OpenAI, OpenRouter, Ollama, or vLLM, with keys stored encrypted. Without a model, cvENGINE still discovers, scans, matches CVEs, and validates findings. The AI adds triage, targeted scans, and template drafts. ## Built on the best
- agharsallah/esp-labs-webTypeScript · 344 Stars
ESP32学习、在线实验模拟、代码阅读
Stars 统计更新于 2026/10/27 07:35:53
查看原始资料
作者描述
The ESP32-S3 bench guide as a React app: live pinout, lessons with interactive simulators, a code reader for the esp-labs firmware, and the bench log. In 11 languages.
编报时的 AI 中文解读
ESP32-S3 开发板的交互式学习指南，提供32个动手实验、实时模拟器、3D板图、引脚图和代码阅读器，支持11种语言。适合想要通过实践学习ESP32-S3的开发者、学生和爱好者。
编报时的 README 依据
<div align="center"> # ESP32 Labs: the bench guide **Wire it, flash it, watch it, then read the code.**<br> An interactive guide to the ESP32-S3: 32 hands-on labs with live simulators, a 3D board, a pinout, a reader that explains the firmware line by line, and the chip's datasheet made readable. [](https://esp-labs-web.vercel.app) [](https://react.dev)
pnpm dev # http://localhost:5180 pnpm check # typecheck + lint + tests pnpm i18n:extract # regenerate the English content catalogues after editing lessons pnpm build # static bundle in dist/ (relative paths: serve it from any folder) pnpm preview # serve dist/ locally (browsers will not run module scripts from file://) ``` ## Configuration Copy `.env.example` to `.env` to override: | Variable | Default | Meaning |
You need **Node 24.14+** and **pnpm 11** (`corepack enable` gives you the pinned version). The code reader shows the firmware from a sibling checkout of `esp-labs`, so clone both repos side by side: ```sh git clone https://github.com/agharsallah/esp-labs.git git clone https://github.com/agharsallah/esp-labs-web.git cd esp-labs-web pnpm install pnpm dev # http://localhost:5180 ```
Without `../esp-labs`, everything else still works: the lessons, the simulators and the bench log. Only the code reader has nothing to show. Point `FIRMWARE_DIR` elsewhere if your checkout lives somewhere else ([Configuration](#configuration)). ## How it is built React 19 with React Compiler, React Router 8 (data mode, hash URLs), TanStack Query 5, zod 4, i18next, three.js, CSS Modules and design tokens, built by Vite 8. Lessons are data, simulators are lazily loaded widgets, and the bench log is JSON fetched at run time. ```mermaid flowchart LR
- yaojingang/GEOInflowTypeScript · 336 Stars
AI爬虫分析、网站分析、内容覆盖分析
查看原始资料
作者描述
Self-hosted web and AI crawler analytics for Generative Engine Optimization (GEO). Track visitors, AI search referrals, crawl logs, content coverage, and conversions. Jointly published by 猎看科技 · 北京日报社智源GEO. MIT licensed.
编报时的 AI 中文解读
GEOInflow 是一个自托管的网站和 AI 爬虫分析工具，用于生成引擎优化（GEO）。它帮助团队了解哪些 AI 爬虫访问了他们的网站、请求了哪些页面、哪些重要页面没有被爬取、请求失败的位置，以及 AI 搜索如何带来访问者和转化。 该系统专为公司网站、内容网站、媒体发布者、开发者文档以及管理多个网站的 GEO 团队设计。操作员可以在网站和页面级别工作，而维护者可以检查数据摄取、队列、来源质量和计算方法。
- th0mk/nfsu2-webC · 334 Stars
游戏运行
查看原始资料
编报时的 AI 中文解读
这是一个粉丝项目，将PC游戏《极品飞车：地下车会2》移植到浏览器中运行，需要用户提供自己的游戏副本。适用于希望在现代浏览器中重温该游戏的玩家。
- JMS1717/ming-voicePython · 332 Stars
语音助手、家庭自动化控制、本地模型部署
查看原始资料
作者描述
Low-latency local speech-to-speech voice assistant with Ming-3.0-audio-tiny on AMD and Nvidia GPUs. Easy setup on Windows and Linux
编报时的 AI 中文解读
Ming Voice 是一个完全本地化、低延迟的语音助手，适用于 AMD 和部分 NVIDIA GPU，在 Linux 或 Windows 上运行。用户可以通过语音与它对话，它能执行天气查询、网页搜索、计时器、购物清单等任务，并能控制家庭自动化设备。
- quanhoangindex/ascii-revealTypeScript · 315 Stars
加载动画、骨架屏
查看原始资料
- atbfun/ATBGamePortPython · 307 Stars
游戏移植、HTML5转换
查看原始资料
作者描述
Transplant all the games to the web platform
编报时的 AI 中文解读
该项目提供一套工业级的Agent Skills，用于将经典桌面游戏逆向工程并移植为现代、零依赖的HTML5 Canvas与Web Audio网页游戏，适用于希望将老游戏迁移到Web平台并保持高质量体验的开发者。
- ZiChuanLan/agent-purgeJavaScript · 285 Stars
系统提示词管理
查看原始资料
作者描述
One persona, many coding agents. Replace or inject system prompts across codex, claude, opencode, grok, omp, hermes, pi, zcode - with a local web console.
编报时的 AI 中文解读
这是一个命令行工具，用于在多个编码代理之间统一注入、替换或管理系统提示词，支持替换、追加和清除模式，并提供本地网页控制台进行热切换和状态查看。
- mochiiwuff/capri-desktopJavaScript · 236 Stars
桌面宠物、Claude状态播报
查看原始资料
作者描述
Capri, the Project Capricorn construct from Cyberpunk: Edgerunners 2, living on your desktop and narrating what Claude Code is doing.
编报时的 AI 中文解读
Capri 是一个桌面宠物应用，让《赛博朋克：边缘行者》中的角色 Capri 出现在桌面上，实时播报 Claude Code 的工作状态，并在需要用户授权时提醒用户。
- kuraeow/ultratearsC · 235 Stars
游戏移植
查看原始资料
作者描述
Unofficial native Windows port of Tears of the Kingdom 1.4.3 using static recompilation. Builds locally from the player's own extracted game dump; no game files included.
编报时的 AI 中文解读
这是一个非官方的《塞尔达传说：王国之泪》Windows 原生移植项目，通过静态重编译将 Switch 游戏代码转换为 PC 可运行版本。用户需自行提供从自己主机提取的游戏文件，在本地构建后游玩。适合希望以原生方式在 Windows 上体验该游戏的玩家。
- NyxFoundation/nyxisLean · 233 Stars
授权安全、后端框架、形式化验证
查看原始资料
编报时的 AI 中文解读
Nyxis 是一个 Lean 后端框架，为 AI 生成的 Web 应用提供经过形式化验证的授权安全保障。开发者只需编写和审批 Cedar 模式与策略，即可确保任何 HTTP 请求序列都无法越权访问数据。 适用于希望通过 AI 代理快速开发 Web 应用、同时确保授权安全性的独立开发者。开发者无需编写 Lean 代码，仅需通过 Lake 集成 Nyxis 并配置策略。
- Bodila51/reverse-engineer-Wi-Fi-routerPython · 158 Stars
路由器隐私检查
查看原始资料
- piermark/omarchytvJava · 156 Stars
电视桌面启动器、主题切换、AI助手
查看原始资料
作者描述
Omarchy-styled tiling launcher ecosystem for Android TV, Google TV, LG webOS, Android Mobile & Linux/macOS/Omarchy PCs
编报时的 AI 中文解读
OmarchyTV 是一个面向 Android TV、Google TV、LG webOS 等设备的 Omarchy 风格桌面启动器，提供原生 TUI 助手、26 个 MIT 主题、4K 壁纸、MCP 服务器和手机无线安装器。适合希望将 Omarchy 桌面体验带到电视上的用户。
- DiaoOrg/awesome-supermimic未标注语言 · 144 Stars
研究资料、项目索引
查看原始资料
作者描述
A curated collection of SuperMimic implementations, extensions, demos, and resources.
编报时的 AI 中文解读
这是一个精选的SuperMimic实现、扩展、演示和资源合集，主要面向研究社区，提供社区项目索引和原始论文链接。
- lemiur/Game-Decompilation-Skill-For-Claude未标注语言 · 139 Stars
游戏反编译、代码匹配、项目协作
查看原始资料
作者描述
Skill for creating a bytematching decompilation of a game with Claude
编报时的 AI 中文解读
为Claude Code提供的一组技能，用于指导游戏字节匹配反编译项目，包括重构、诊断、排列、审查和合并等方法。
- Motionsites-AI/claude-skillPython · 136 Stars
网页构建、提示词生成
查看原始资料
作者描述
Free Claude skill: make Claude build award-level, cinematic landing pages. By MotionSites.
编报时的 AI 中文解读
这是一个Claude技能，帮助Claude构建电影级落地页，包含示例提示词、媒体素材、技术配方和自动检查修复循环。
- mxhm11cr/LinuxRunnerC · 119 Stars
运行Linux应用、Linux桌面环境
查看原始资料
作者描述
Run ARM64 Linux applications and desktop environments on iPhone.
编报时的 AI 中文解读
LinuxRunner 是一个在 iPhone 上运行 ARM64 Linux 应用和桌面环境的工具，无需越狱、虚拟机或 JIT。它通过自定义解释器和兼容层实现，适合希望在 iOS 设备上体验 Linux 桌面和开发工具的用户。
- boyuchengz/canvas-mcpJavaScript · 115 Stars
课程信息查询、作业提醒、成绩查看
查看原始资料
作者描述
Read-only MCP server that connects Claude, ChatGPT and other AI assistants to Canvas LMS.
编报时的 AI 中文解读
这是一个只读的MCP服务器，用于将Claude、ChatGPT等AI助手连接到Canvas LMS，让用户可以用自然语言查询课程信息、作业、成绩和公告等。
- Aaryan-Kapoor/english-rsRust · 101 Stars
语法检查、类型系统演示
查看原始资料
作者描述
English, rewritten in Rust: grammar as trait bounds, pronouns as borrows, and englishc, a compiler with rustc-style diagnostics for prose
编报时的 AI 中文解读
这是一个用Rust重写英语的项目，包含一个将英语语法编码为类型系统的库和一个检查普通散文语法错误的编译器，以rustc风格输出诊断信息。适合对Rust类型系统感兴趣的程序员和语言爱好者。
- jacke-pp/ai-wallet-binanceJavaScript · 96 Stars
加密货币交易、狙击新币、资产转账
查看原始资料
作者描述
让任何 AI 用你自己的币安 AI 钱包交易：秒买秒卖、狙击 Agent 专属发射台（Four、Flap、GANG）、AI 专属 Mint、本机交易页面。币安签名，不碰私钥。Let any AI trade with your own Binance AI wallet.
编报时的 AI 中文解读
这是一个命令行工具，让AI通过币安AI钱包进行加密货币交易，支持买卖、狙击、Mint和资产转账，并在本地提供可视化交易界面。
- ELU-Labs/dark-factory未标注语言 · 94 Stars
自动化开发、模型服务接入
查看原始资料
作者描述
One prompt that sets up a dark factory for your repository with Grok Bots and ELU.
编报时的 AI 中文解读
通过单个提示词在GitHub仓库中搭建自动化软件交付流程，借助Grok Bots和ELU完成从需求到发布的全部开发环节，无需人工编写或审查代码。适合拥有Web应用仓库并希望实现全自动化交付的开发者。 工厂需要特定基础设施和账户支持，包括GitHub仓库、CI系统和ELU账户，并推荐使用Grok之外的模型作为验证器以避免盲区。 提供改进渠道，用户可通过GitHub issue或PR对提示词本身进行反馈和修正。
- OrchardLinux/omarchy-aQML · 83 Stars
桌面环境
查看原始资料
作者描述
Experimental Omarchy desktop for Apple A-series devices.
编报时的 AI 中文解读
为Apple A系列设备（如iPad）提供实验性桌面环境，基于Arch Linux ARM，重新设计Omarchy界面以适配触控操作。适合希望在iPad上运行Linux桌面的技术爱好者。
其余 8 个项目见收录列表。