ai-design-check is a zero-dependency design-rule checker and agent skill for interfaces built with AI coding agents. It reads web projects including React, Vue, Svelte, Laravel Blade, and HTML, as well as SwiftUI, Jetpack Compose, Flutter, and React Native code, and reports findings with the file and line where a rule is broken. Its rules cover accessibility, responsive layouts, type and spacing scales, component reuse, writing and localization, visual language, motion, and data presentation. The checker can apply mechanical fixes without an agent, while its installed skills can measure a screen, write a design brief, and have an agent repair the remaining issues on a new branch. Additional commands scan a running page in Chrome, inventory component variations, test narrow layouts and long-language content, generate a design system, and create before-and-after videos. It runs through the npm package on Node 20+, supports CI output and pull-request annotations, and is licensed under MIT.
AnyScene is an open-source AI-assisted scene-building process and Blender toolchain that turns reference images or a one-paragraph brief into a dressed, lit 3D scene. An AI agent working through SKILL.md plans the room, creates a prop manifest and reference image for each prop, sends each prop through the Tripo H3.1 and Meshy 7.1 image-to-3D models on fal, selects between the results using same-light comparison renders, and scripts the room shell, layout, dressing, lighting, grime, haze, weather effects, cameras, and Cycles renders. Each stage ends with a visual review gate, and the scene is regenerated from scripts rather than manually modelled. The repository includes example scene data, Blender and FFmpeg pipeline utilities, camera and shot rendering, prop and model-generation clients, a fal GPU render farm, and an optional hosted endpoint. The endpoint accepts one to four reference images and an optional prompt and returns a scene package containing a Blender file, GLB, stills, and panoramas. Running the local workflow requires Blender 5 or newer, Node 20 or newer, a fal API key, FFmpeg, uv, Claude Code, and a GPU with 12–16 GB of memory for Cycles. Code is licensed under MIT; generated assets are outside the repository and have their own licences where published.
Arken-core is the Rust cryptographic core of ArkenForge, a library for passwordless authentication using W3C did:key identities and digital signatures. It generates cryptographically random, one-time challenge nonces with a 60-second TTL, parses did:key identities, and verifies ECDSA secp256k1 signatures. A client signs a server-issued nonce with its private key and retries with an Authorization: Signature header; the core verifies the signature and nonce and returns the client's DID without storing session state. The library compiles to a native shared library for use by higher-level packages through FFI, including the ecosystem's Express and FastAPI middleware and client SDKs. The repository describes it as early-stage and not production-ready: nonces are held in memory, do not work across multiple instances without a shared store, and there is no key-recovery mechanism. It is licensed under Apache 2.0.
blaze is an incremental Rust build tool that replaces selected Cargo commands, including build, check, run, test, and bench. Its daemon keeps the project compiled in memory; after an edit, it compiles only the changed functions and links the program with its own linker, enabling millisecond-scale rebuilds and live edits to a running program. It uses its own Rust toolchain and supports macOS on Apple silicon and Linux. The tool trades memory for speed, requires several gigabytes for a small project, and is distributed under the MIT or Apache-2.0 licenses.
bootgpt is an experimental character-level language model that boots on x86 hardware before an operating system loads. Its 510-byte handwritten 16-bit real-mode assembly runtime uses BIOS calls to load an approximately 11.5 KB int8 model from subsequent disk sectors and generate fictional Linux boot-log text. The model is trained in Python on generated dmesg-style logs using character embeddings, one hidden layer, and a softmax output. Training quantizes the weights with power-of-two scales and simulates the integer arithmetic used by the assembly implementation, including its sampler, so the Python and boot-sector outputs can be compared. The repository provides scripts to build and run a disk image with NASM, QEMU, and NumPy; the image can be written to a USB drive and booted on a BIOS-based machine.
Claude animation skills is a set of six Claude Code skills for generating short animated reels without filmed footage. The skills produce 3D cartoon explainers, stick-figure explainers, minimal line art, 3D map routes, exploded views, and animated line graphs. Claude writes a specification or explainer script, then the skills render animations on the user's computer. The quick skills use Python build scripts, HTML templates, GSAP or three.js, HyperFrames, and headless Chrome via Playwright; they output 1080x1920 MP4 video and a contact sheet with Instagram's safe area marked. The stick-figure and 3D cartoon skills also generate or process local voiceovers and synchronize captions using Kokoro and NVIDIA Parakeet models through sherpa-onnx. The project runs on macOS or Linux and requires ffmpeg, Node.js 18 or later, and Python 3.9 or later. It can be installed as a local repository or as a Claude Code plugin. It is made by Artem Novitckii and licensed under the MIT license.
Co-op Anything is an open-source MCP server and playbook for directing AI coding agents to add cooperative multiplayer to single-player PC games. The MCP server searches a directory of more than 230 co-op mods, detects a game's engine from its installation folder, provides research and reverse-engineering guidance, and validates contributions. An agent either revives an existing mod by auditing and fixing its code or builds co-op from scratch by hooking into the game, synchronizing players, enemies, worlds, story, level changes, and saves. Builds are tested with bots driving real game clients, then packaged for installation and Steam-relay play without port forwarding. The repository includes agent skills, engine-specific guidance, networking and split-screen playbooks, automated QA procedures, game-specific projects, and contribution rules. It is licensed under the MIT License.
Droy is an open-source Android AI assistant for conversation, research, and phone tasks, maintained by moonoops. Its Tasks agent plans actions, uses Android Use to inspect screens and operate apps by tapping, swiping, and entering text, then checks the result; sign-ins, verification codes, and other steps requiring user input are handed back to the user. It also provides companion chats with separate memories, voice interaction, notifications, on-screen pets, and extensions through skills, plugins, MCP, and automation via DeepSeek Harness (DSH). The runtime runs on the phone after setup and supports user-selected model services; cloud models require an internet connection. Droy uses a Kotlin/Compose interface, supports Chinese and English, requires an ARM64 Android 11-or-later device, and its own code is licensed under AGPL-3.0-only.
dsh-superagent is a research-prototype proactive coding agent for a single GitHub repository, packaged for DeepSeek Harness. It listens to GitHub and optionally Feishu; when an authorized person requests work, it creates a work item, runs the task in its own agent session and directory, and replies in the original conversation, potentially with a pull request. A web panel shows items requiring human attention, while authorization rules limit when it acts and what it communicates. The repository provides documentation covering its philosophy, domain model, authorization, execution, prompts, panel, deployment, architecture, and adding sources. It is research code rather than an official DeepSeek product or production-ready release, and is licensed under MIT.
DuoDesk is a free, open-source workspace application by 0xCodez that connects macOS, Windows, Linux and Android devices so they can share input, displays, applications, files, clipboard data, cameras, sound, notifications, calls and texts. It lets a mouse and keyboard move across computers and phones, uses a tablet or phone as a second screen, exposes another computer's applications in windows, and provides phone controls such as calls, messages, photos, files, camera, remote control and gamepad input. Each device runs a DuoDesk client; computer clients use Electron with native helpers for input injection, screen capture, virtual displays, window capture and drag-and-drop, while the Android client uses Kotlin and can use wireless debugging with adb and scrcpy. Devices pair with a code and communicate directly through an encrypted TLS session over the local network, USB, Bluetooth LE, Wi-Fi Direct or a phone hotspot. DuoDesk can also connect multiple devices with DuoDesk Mesh, sharing the pointer, clipboard and notifications without a central hub. The application requires no account, cloud relay or telemetry and is distributed for macOS, Windows, Linux, NixOS and Android. It is licensed under the GNU GPL v3.0 and is based on TwinDesk.
fm1-sampler is an Android sampler and eight-track audio looper that uses the M-VAVE FM-1 as its hardware control surface. The phone performs the sound processing, while the FM-1 supplies keys, buttons, seven encoders, a 240×240 secondary display driven by SysEx commands, and USB audio output. It accepts microphone recordings, imported audio files, and bundled samples. Samples can be played melodically, as one-shots or loops, as transient-detected slices, with granular synthesis, or with Paulstretch; the app also provides per-track envelopes, LFOs, an arpeggiator, tape-style effects, reverb, delay, sidechain compression, pitch and time stretching, and shared loop recording with overdub. Its DSP is implemented in the repository's engine/dsp code, with KissFFT used for Paulstretch mode. The repository provides APK and matching FM-1 firmware downloads in its releases. The README reports that the project has been host-tested, including its DSP and simulated interfaces, but had not yet been run on a physical phone or FM-1 after the deck redesign.
Ghostman is a keyboard-driven terminal user-interface (TUI) API client and webhook inspector developed as a local-first alternative to cloud-based API clients. It composes and sends GET, POST, PUT, DELETE, and PATCH requests, displays HTTP status, payload size, DNS and time-to-first-byte latency, and formats JSON responses with syntax highlighting. Requests and collections are stored as plain JSON or YAML files in the local .ghostman directory. A built-in webhook server can listen on port 9090; captured callbacks can be inspected in a live stream and loaded into the request editor. Keyboard shortcuts control sending, saving, creating requests, and toggling the webhook listener. The project is distributed through pip or from source and is released under the MIT License.
Git koi is an open-source tool that generates a pixel-art koi from a GitHub account’s public activity. The account’s programming languages determine the koi’s colors, contributions determine its size, streaks mark it, and the day’s activity determines how it swims. A nightly GitHub Actions workflow renders the koi as a looping GIF in a profile README; the project also provides a command-line tool and an underlying Rust library published on crates.io. It is released under the MIT license.
godraw is a local, native whiteboard application written in Go and drawn with MyGo's native UI. It has no web view or JavaScript and reads and writes tldraw .tldr files as an independent reimplementation. The editor provides selection, freehand drawing, erasing, arrows, text, notes, images, lines, highlighter, laser, frames, and geometric shapes, along with multi-selection, snapping, grouping, alignment, multiple pages, undo and redo, autosave, crash recovery, and document windows. It exports selections or pages as PNG and SVG and supports panning, zooming, a minimap, dark mode, and focus mode. Unknown tldraw record types are preserved and written back, but rich text is simplified when edited; multiplayer, video playback, image cropping, and several other tldraw features are not implemented.
HAS_Prepper_Disk is an offline knowledge-archive setup for Raspberry Pi 5. It combines Internet-in-a-Box and Kiwix with configuration files, installation and helper scripts, a content-download script for Kiwix ZIM files, and a custom homepage. The Pi serves Wikipedia, maps, medical references, repair guides, and other survival resources over its own Wi-Fi hotspot, allowing phones and laptops to browse the material without internet access. The project mounts an SSD at `/library`, installs Internet-in-a-Box, downloads content selected in `config/content.conf`, and provides an administrative interface for choosing map regions. Its default content configuration requires approximately 150 GB of storage. The repository's scripts and homepage are released under the MIT License; Internet-in-a-Box, Kiwix, and the downloaded content retain their own licenses.
hop is a Windows application for installing Linux alongside Windows without a USB stick. It checks the PC and selected drive, downloads and verifies a Linux ISO from official sources or accepts an existing ISO, creates a FAT32 installer partition on an internal disk, copies and checks the ISO files, adds a one-time startup entry using the ISO's signed bootloader, and restarts into the Linux installer. It can make space by using existing free space or shrinking a Windows drive, and can undo an installation or remove another Linux installation. Supported systems include Ubuntu and its flavours, Linux Mint, Pop!_OS, Zorin OS, Debian, Fedora Workstation, Arch Linux, Omarchy and others, with system-specific limitations such as Secure Boot or BitLocker requirements. It requires Windows 10 version 1803 or newer or Windows 11 on a UEFI PC with a GPT disk; Linux must be installed on a disk inside the PC. The project is distributed as hop.exe and is licensed under the MIT License.
Initialis is an interactive browser-based explorer and JavaScript library for generating ornamental drop caps, illuminated lettering, and related typographic designs. Each initial is rendered in the browser from a letter, frame, ground, ornament, texture, and layered inks rather than being a pre-made image. Users can wander through generated variations, bookmark designs, preview them as the opening of a printed page, share settings, and download transparent and paper-backed images, animations, videos, and JSON settings. The site can be run as a dependency-free static site; its library can render still or growing initials in other webpages from settings or share links. Rendering runs in workers using WebGL2, with WebGPU used for the aged-page effect where supported. The code is MIT-licensed, while bundled fonts and third-party libraries retain their own terms.
Lasm is a compiler and CLI that compiles Lean 4 programs ahead of time to WebAssembly for JavaScript and WebAssembly runtimes. It supports Node.js, Bun, Deno, browsers, serverless and edge platforms, WASI, Spin, and Fastly, with targets that produce deployment-specific output such as Node launchers, browser sites, function packages, or WASI modules. Applications can use Lasm libraries for HTTP handling, browser interfaces, typed client-server APIs, web APIs, key-value storage, and JavaScript/npm interoperability. Lean proofs, tactics, macros, elaboration, and evaluation run during compilation; deployed programs cannot elaborate or compile new Lean code or load Lean compiler data at runtime. The project includes templates for command-line programs, REST APIs, serverless functions, edge functions, browser applications, and full-stack applications, and is distributed under the MIT license through the @lasm/cli npm package.
Looseleaf is a React, Tailwind v4, and Radix component library with a shadcn registry, making accessible interface components look hand-drawn on paper. It provides 123 components, 61 layout blocks, coded icons and illustrations, stickers and stamps, a sketchpad, a chat kit, and the Kevin Sans typeface. Components retain shadcn/ui parts, props, and data-slot names, and can be added to a project with the shadcn CLI; the generated source is copied into the application rather than installed as a dependency. Its drawing layer places an SVG behind the real DOM label. Seeded TypeScript recipes generate the artwork with CSS-variable colors, while pre-drawn layers controlled by CSS handle hover, press, and focus states. The library includes accessibility features such as keyboard support, visible focus, reduced motion, forced colors, RTL, text contrast, target sizing, and multiple density options. The code, artwork, and Kevin Sans are licensed under MIT; IBM Plex Mono is licensed under SIL Open Font License 1.1. The project is developed and distributed through its GitHub repository, documentation site, and shadcn registry.
Nanolight is a native Zig application launcher for macOS, Windows, and Linux, drawn with nanoui and distributed without Electron. It opens from a configurable hotkey and searches applications by full names, word starts, initials, ordered letters, and typos; frequently opened apps rise in the ranking. It also provides an inline calculator, Google suggestions with a privacy guard for address-like queries, Google results in an embedded panel, Grok and X search, Raycast extensions, and Raycast script-command support. Raycast extensions run through Nanolight's implementation of @raycast/api and require Node.js 20 or later, while script commands run directly in Zig without Node.js. On macOS it runs from the menu bar without a Dock icon. Optional synchronization through Codegraff shares the launcher key, Google-suggestions setting, and app-opening counts while excluding queries, history, and files. The project is licensed under the MIT license.
NectarLink is a free, open-source application that connects an Android phone with a Windows PC, Linux computer, or preview macOS client. It provides notifications, messaging, calls, clipboard and file transfer, photo access, screen mirroring, phone apps in desktop windows, remote phone and PC controls, phone-as-webcam support, and audio features. Devices pair by QR code or a six-digit code and communicate directly over a local network or USB, without an account or cloud service. Connections use QUIC with TLS 1.3 and device keys; each feature has its own per-device permission switch. The shared Rust core handles pairing, sessions, storage, protocol, and capability negotiation, while the Windows desktop app uses Qt Quick and Rust and the Android app uses Kotlin and Jetpack Compose. The project distributes Windows and Android builds through GitHub Releases and documents Linux packages and a macOS preview. The applications and command-line client use GPL-3.0-or-later, core libraries use MPL-2.0, and the protocol documentation uses CC BY 4.0.
neorex is a Rex-native frontend for Neovim, written in Zig. It runs one Neovim instance behind multiple Rex panes, sharing buffers, registers, undo history, and language servers; Neovim splits map to Rex panes, while focusable floating windows become floating panes and other UI elements are drawn inside their associated pane. It follows Rex light/dark themes and palette colors, and includes Neovim colorscheme and lualine themes. The project targets macOS, requires Rex and a recent Neovim, and is built with Zig. It is described as very early and incomplete: Neovim window move, swap, and resize commands do not control Rex panes, only one tab page is supported, and some plugin layouts and command-line messages are limited. neorex is released under the MIT license.
OmniDiff is a Rust command-line diffing tool by Marko Ivankovic that compares files according to their content. It uses tree-sitter grammars to structurally diff 25 programming and markup languages, showing changes in terms of functions, classes, structs, and other syntax nodes rather than only line-by-line text; unknown extensions fall back to plain-text comparison. It provides an interactive terminal viewer, plain-text headless output for scripts and CI, JSON output, and integrations with Git and Jujutsu. It also compares non-text content, including pictures pixel by pixel, archives member by member, fonts glyph by glyph, message catalogs message by message, and PDFs page by page. It can be installed from source with Cargo, through prebuilt releases and several package managers, or used through VS Code and Neovim integrations. The project is distributed under the GNU Affero General Public License version 3, with a separate commercial license available through GitHub Sponsors.
Open 3D Skills is a collection of eight agent skills for building 3D and animated websites with three.js, scroll-scrubbed films, transparent video, cursor-following characters, exploded-view explainers, AI-generated motion assets, browser checks, and frame-exact video recording. Its skills include code-built real-time scenes, generated-clip pipelines with start and end frames, per-frame transparent-video processing for browser playback, pose-grid cursor rigs, staged machine views with labels and picking, and automated Chromium and WebKit checks for frame time, draw calls, errors, blank canvases, overflow, and reduced-motion behavior. The frame-capture skill records pages at 60 fps using a virtual clock and can create comparisons. The repository installs selected or all skills through the `npx skills` CLI, Claude Code, Codex, Cursor, Gemini CLI, OpenCode, or a shell installer; each skill is a plain `SKILL.md` folder that compatible agents can load. The repository's scripts require Node 22.20 or newer, Python 3.9 or newer with Playwright and media-processing libraries, and ffmpeg; some skills also require an image or video generation service. It is licensed under the MIT License.
pyWolf3D is a Wolfenstein 3D game clone and engine written from scratch in Python with Pygame. It uses a raycaster with the DDA algorithm to render 64×64 tile maps, handling doors and pushwalls explicitly, while sprites and projectiles use perspective projection. The game includes actor behavior, map parsing, directional and attenuated sound, and runtime music synthesis through an OPL emulator using the original game's music data files. The repository includes the shareware first episode; the remaining original episodes require game files from a purchased installation. It can be installed from the repository's requirements file and run as a Python module.
Shim is a local-first, terminal-based AI harness for connecting model APIs, MCP servers, and Agent Skills. It runs as a single static binary with no runtime dependencies, desktop UI, daemon, telemetry, or bundled model weights; it is a harness rather than an inference engine. It provides both a scriptable CLI and natural-language control. An optional local model routes management requests to an allowlisted command, while mutating commands such as file edits, shell commands, and MCP calls require confirmation by default. Shim supports OpenAI-compatible and Anthropic model adapters, MCP over stdio and Streamable HTTP, and on-demand loading of SKILL.md files from configured skill directories. Configuration is stored as JSON and sessions as append-only JSONL, with credentials supplied through environment variables. The project is written using only the Go standard library and is released under the MIT license.
Skein is an open-source execution control plane for local AI-agent swarms, developed by Paritet and written in Go. It runs explicit JSON task graphs with dependencies, bounded concurrency, reusable multi-pass blocks, per-pass timeouts, and commands represented as argument arrays rather than passed through an implicit shell. The static binary requires no daemon, database, or runtime dependencies. Its fail-closed stop mechanism uses a persistent epoch fence: an emergency stop invalidates queued work and rejects late completions from the previous epoch, while tracked local process trees can be terminated on Linux, macOS, and Windows. Skein is an execution control plane rather than an agent framework or model runtime. The project identifies sandboxing arbitrary commands, authentication, semantic output validation, exactly-once external side effects, distributed consensus, remote work, and automatic crash recovery as outside its coverage. It is distributed under the MIT license and is described as an early release.
An agent skill by caru-ini that turns a natural-language description of a technology stack into an architecture diagram with official logos. It works with Claude Code, Codex, Cursor, and other agents supporting skills, and can depict application infrastructure, development tooling, AI applications, and cloud environments including AWS, Google Cloud, and Azure. The skill fetches logos at build time, routes connector lines around labels, and renders light and dark versions. It produces a self-contained HTML file, PNG, dependency-free React component, or Artifact; PNG export uses Playwright with Chromium. It requires Python 3.9 or later and network access for logo retrieval. The code and documentation are MIT-licensed, while bundled fonts use the SIL Open Font License 1.1.
T3 Code is an open-source agent-harness control surface for coding agents running on a developer's computer. It provides iOS and Android mobile apps, a web app, and an Electron-based desktop app for controlling locally configured Codex, Claude Code, Cursor, Grok Build, and OpenCode agents while using the user's existing provider subscriptions. The project can be launched with `npx t3@latest`, which starts its backend and local web interface on the machine. It also distributes desktop builds for Windows, macOS, and Arch Linux, and supports remote access from a phone or another machine. The repository describes the project as early-stage and warns that bugs are expected.
Teachable Figures is an agent skill for Claude Code and Codex that turns a lesson claim into an interactive teaching figure. It structures each lesson around a predict–try–explain loop: the learner predicts an outcome, changes a cause, observes the effect, and explains a new case. The skill can produce standalone HTML figures or integrate with a host application's lesson schema, using 2D SVG, isometric or mechanical SVG, turning 3D, and optional WebGL renderers. The workflow captures the audience, claim, misconception, representation, fidelity, and controls, then builds from a shared model that drives motion, readouts, accessible summaries, and explanations. It includes rendering kits, reviewed examples, browser checks, and a learner workspace using LESSONS.md, LEARNER.md, and LEARNING-PLAN.md. The repository provides fifteen complete examples, including gear ratios, transformer data flow, seasons, Bayes and base rates, natural selection, and dynamic equilibrium. The skill and standalone pages have no package runtime dependencies; Node.js is used for Anatomy builders and Playwright for browser checks. Its own code is MIT-licensed and its original lesson content is CC BY 4.0, while bundled upstream skills and third-party assets retain their stated licenses. The figures are teaching designs rather than evidence of learning effectiveness, and the repository notes that they have not been evaluated with novice learners or a screen reader in the cited review.
UNCUT is a headless coding-agent harness written in Go for shell pipelines, CI jobs, containers, and agent experiments. It sends the complete conversation, file contents, and command output to an OpenAI-compatible Chat Completions endpoint through one agent loop with four tools: read_file, write_file, replace, and run_command. The tool does not summarize, prune, compact, or silently truncate context; provider-side response truncation is reported as an error instead of being presented as a successful partial answer. It accepts piped UTF-8 input, can emit structured NDJSON events, and optionally stores complete resumable sessions in append-only JSONL files. UNCUT is distributed as a statically linked Go binary for Linux, macOS, and Windows, supports local OpenAI-compatible endpoints, and is released under the MIT license. It is not a sandbox: shell commands run with the operating system user's permissions.
Verdict is a local semantic verifier for AI-agent tool calls and command-line runs. It reads the command, exit code, stdout, and stderr, then classifies the observed outcome as success, failure, blocked, or unknown. Its zero-dependency inference model uses hashed character 3–5-grams and word 1–3-grams with multiclass logistic regression; the released weights are quantized int8 and run on the CPU without network calls or telemetry. The repository provides a Python API and CLI, with bounded head-and-tail clipping for long output streams and an optional browser demo using the same quantized weights. Verdict is an advisory classifier rather than authoritative proof of task completion, code correctness, safety, or truthfulness. The project describes its benchmark as based on synthetic, English-biased examples and recommends deterministic checks and authoritative state verification for consequential actions. The code and model weights are Apache-2.0 licensed, while the synthetic dataset is CC0-1.0.
Who Are You is an open-source privacy self-check that runs on Cloudflare Workers and shows what a website can infer from a visit. It runs browser and network sniffers without permission prompts, combines their readings into a portrait covering location, language, time zone, device, browser, graphics hardware, network, preferences, and fingerprinting, and compares those readings to identify contradictions that may reveal a VPN or proxy. Twelve environment checks receive an A–F grade, including time-zone and language mismatches, DNS and WebRTC leaks, split routing, proxy-like round-trip differences, data-center networks, user-agent inconsistencies, tampering, fingerprint stability, and automation. The site provides per-check explanations and system-specific DNS or WebRTC fixes, supports comparisons with previous runs and sharing of masked results, and keeps most readings in the browser; its Cloudflare Worker handles the request data and anonymous aggregate counters without retaining visit logs. It can be self-hosted with Wrangler and is released under the MIT license.
Searchable transcript of GitHub Trending Weekly #57: hop, coop-anything, nectarlink, omnidiff, t3code, blaze, DuoDesk, shim — Github Awesome (15:17). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by Github Awesome. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 Welcome back to GitHub awesome. This is GitHub trending weekly number 57. 35 trending open- source projects on GitHub right now. Let's go. Hop is a Windows app that installs Linux right next to Windows with no USB stick needed built-in C. It checks your PC first, then makes a small installer partition on your own drive and copies the Linux ISO onto it using the ISO's own signed bootloadader.
00:26 You restart once and you're in the Linux installer with a list of the choices you'll need to make. It downloads Ubuntu, Linux Mint from their official servers and checks each download. Co-op anything gives AI coding agents a playbook for turning single player PC games into co-op games you can play with friends. An MCP server reads a game's install folder to work out its engine and checks a shared directory of more than 230 co-op mods.
00:53 Then the agent either finishes a mod someone already started or builds co-op from scratch. Nectar Link connects your Android phone to your Windows PC so you can stop picking up your phone every time a notification arrives. Reply to messages, move files, share your clipboard, or open phone apps in their own windows on your computer. The devices connect directly over your network or USB with no account required.
01:21 Some features need wireless debugging set up on your phone first. Omnidiff is a code diffing tool written in Rust that understands the structure of your code, not just its lines. It parses each file with tree sitter, so it can show that you renamed a function or moved a block instead of a wall of red and green lines. It handles 24 languages and falls back to a normal line diff for anything else.
01:46 You can browse changes in an interactive terminal view, print plain output for scripts, or get JSON for your editor. T3 Code is Theo's open-source control panel for the AI coding agents running on your own machine. Sign into the agents you already use, like Codeex, Claude Code, and run one command to start a local server. Then you manage all of them from one place, whether that's the web app, a desktop app, or the iOS and Android apps.
02:15 It handles permission modes, multiple accounts, and source control, and it can run in the background. DSH Super Agent is a prototype of a proactive coding agent for a single GitHub repository built for Deep Seek Caress. When someone on your team asks for work on GitHub or on FishU if you connect it, it opens a work item and handles it in its own agent session and folder.
02:40 Then it replies right where you asked, often with a pull request. It only acts when asked or when a team rule allows it. Stack diagram is an agent skill for clawed codecs and cursor that turns a one-s sentence description of your text stack into a clean architecture diagram. Say something like next.js on Versel talking to Postgress and Stripe and the agent builds the diagram, renders it and checks it over before handing it back.
03:10 It pulls each tool's official logo at build time and never redraws or recolors it. This repo holds Benji Taylor's design skills for coding agents, starting with stage show, which builds realistic product UI mockups in React. Ask for a still shot, a feature card, a cursor demo, or a bento grid, and the agent composes it using stage shows guidance on layout and motion.
03:33 It comes with a starter kit and worked examples. Then it records what it built so the agent can watch its own mockup and fix what looks off. Rejit in the Blaze repository makes rebuilding a Rust project after an edit take milliseconds instead of seconds. A background process keeps your whole project analyzed and compiled in memory using Rust Analyzer and the Cranalifty code generator.
03:59 When you change a function, only that function gets recompiled and it's patched straight into the program that's already built with no full relink. Has prepper disk turns a Raspberry Pi into an offline library you can reach from your phone or laptop. It uses Internet in a Box and Kwix to serve downloaded Wikipedia articles, maps, and repair guides over the Pi's own Wi-Fi hotspot.
04:23 Once the content is downloaded, you can browse it without an internet connection. You'll need a Raspberry Pi 5 and an SSD with room for the archive. Who are you is a privacy selfch check that shows what a website can figure out about you just from your visit. Running on Cloudflare workers. Open the page and scripts in your browser read your settings and fingerprints without asking for any permission.
04:49 It writes a little portrait of you with your likely location, device, and browser. Then it grades 12 checks from A to F, flagging leaks like a time zone that doesn't match your IP with tips to fix each one. Duesk connects your computers and Android devices so they feel like one workspace. Move your mouse between a Mac and a Windows PC, copy text across devices, or use a tablet as a second screen.
05:15 It can also bring your phone's calls, messages, and camera to your computer. devices connect directly without a cloud account and you choose which features to turn on. Uncut is a tiny headless coding agent written in Go and shipped as one static binary with no dependencies made for shell pipelines, CI jobs and containers. It works with any OpenAI compatible endpoint, local models through Olama included.
05:42 There's no interface, no telemetry, and no plugins. Just one agent loop with four tools. read a file, write a file, replace text, and run a command. Goddraw is a native whiteboard app written in Go by Egoist, built in the style of TLRAW and able to open and save the same TLDDR files. There's no web view and no JavaScript, just a native interface built on his MYO framework.
06:07 You get shapes, arrows, frames, and a highlighter with grouping, alignment, and full undo. Boards can have several pages, save automatically, and export to PNG or SVG. Initialist generates ornate drop caps that look like they belong at the start of an old manuscript. Drag around to explore designs, pick a letter you like, and it fills the screen with variations.
06:30 You can bookmark your favorites, preview one on a printed page, or download it as an image or animation. Each design is drawn in your browser and there's a library for adding one to your own website. Open 3D skills gives your coding agent a set of playbooks for building animated websites. It covers everything from 3JS scenes and scroll driven films to interactive diagrams that pull machines apart.
06:56 The package also helps the agent check layouts in real browsers and record the finished page as a video. Some effects need an external image or video generation service. The codebuilt 3D scenes don't. Shim is a small terminal based AI agent that connects your models, MCP tools, and agent skills in one place. You can use a hosted provider or a local model, then ask it to work on your code or manage its setup in plain language.
07:23 It asks before editing files or running commands by default. The app itself is one binary, but you'll need to connect a model separately. Neorex lets Neovim live inside the Rex terminal as real Rex PES written in Zigg. One Neoim runs quietly in the background and powers every pane you open. So your buffers, registers, undo history and language servers are all shared.
07:47 When you split a window in Neovim, it opens as a proper Rex pane and closing that pane closes the window. Teachable figures is an agent skill for Claude Code and Codeex that builds interactive figures for teaching where students learn by doing instead of just looking. Each lesson asks the learner to predict what will happen, change a cause, watch the effect, then explain a brand new case.
08:12 The agent starts by writing down the idea, the common misconception, and the real behavior. Then picks a style from simple SVG to full 3D. Loose Leaf is Shaden UI drawn by hand, a React component library where every button, card, and input looks sketched on paper. It installs through the Shaden command line. So the components get copied straight into your project as code you own.
08:37 There are 123 components, 61 ready-made blocks, plus handdrawn icons, doodles, stickers, and its own handwritten font. Gitcoy turns your GitHub activity into a pixel art koiish that swims across your profile page. built in Rust. Every night, a GitHub action draws a fresh looping GIF for your profile readme. The fish's colors come from the languages you code in, and it grows bigger as your contributions add up.
09:04 Your streaks show up as markings on its body, and yesterday's activity decides how lively it swims. Ghost brings API testing into your terminal with keyboard shortcuts for sending requests and saving them into collections. responses show up with formatted JSON, status codes, and timing details while your saved requests stay in local files. It also has a built-in web hook listener so you can catch a call back and inspect its headers and payload without leaving the terminal.
09:38 FM1 sampler turns an MVA FM1, a small music controller, into a sampler and eighttrack looper powered by your Android phone. The FM1 gives you the keys, knobs, and its little screen while the phone does all the sound processing with audio code written from scratch. Record from the mic, import a file, or pick from the built-in library, then play it as melodic notes, chopped up slices, or granular textures.
10:05 LSM compiles lean for programs ahead of time into web assembly, bringing a language built for mathematical proofs to the JavaScript world. Write your code in lean. Run one build command and you get a web assembly file plus a small JavaScript launcher. The same code can run as a node server, a serverless function on AWS Lambda or Cloudflare workers or an app in the browser.
10:29 Any scene builds photorealistic Blender scenes from a single paragraph with AI agents doing every step. The agent starts by generating reference images, then turns each prop into a 3D model through two different services and keeps whichever came out better. It scripts the room layout, sets up the lighting, and adds wear and grime so nothing looks too clean.
10:52 Boot GPT is a tiny language model that fits in a computer's 512 boot sector and runs before any operating system loads. The boot code is handwritten assembly running in 16- bit real mode, the way a 1980s PC starts up. It loads the model's weights from the disk sectors right after it, then does every calculation with plain integer math. The model was trained in Python on fake Linux startup logs.
11:20 Arkin O is a library for passwordless sign-in on web servers with a Rust cryptography core that plugs into Express and Fast API. Instead of a password, each user has a private key and a public identity called a DID. When someone asks for a protected page, the server sends a one-time challenge. The user's app signs it with their key and sends it back, and the server checks that the signature matches the identity.
11:47 The Copar LTE modem is an open hardware circuit board that gives your computer its own phone line through a single USBC cable. It's built around a Quctel LTE module, so your computer gets mobile data, text messages, and real voice calls. A small RP2040 chip on the board shows up as a USB sound card, so call audio goes through like an ordinary microphone and speaker.
12:11 Claude animation skills helps you create short animated videos without filming anything. Each skill makes a different style. a 3D cartoon whatif explainer, a stick figure explainer, minimal black and white line art, a 3D route on a map, an exploded view of an object, and an animated line graph. Under the hood, they use 3JS and Gap with a local voice model reading the script and timing the animation to match.
12:38 Droid is an open-source AI assistant that runs on your Android phone built on Deep Seek harness with a Cotlin interface. Give it a task and an agent plans the steps, reads what's on screen, taps, swipes, and types its way through your apps, then checks that it worked. It hands the phone back to you whenever a signin or verification code comes up, and you can stop it anytime.
13:02 Pywolf 3D is a remake of the classic Wolfenstein 3D, written from scratch in Python with Pygame, and the author wrote every line by hand with no AI help. It draws its fake 3D world the way the 1992 original did with a raycaster that casts one ray per column of the screen to find how far away each wall is. Sprites are scaled to look right at a distance and the music plays through an emulation of the old OPPL sound chip generated live.
13:32 Nanolite is a spotlight style launcher for Mac OS written in Zigg with no Electron and the whole app is about 1 megabyte. Press a hotkey and start typing. It finds your apps even from initials or a typo. And the ones you open most rise to the top. It also does quick math, shows Google suggestions and opens results right inside its panel. You can jump straight to a result with command and a number.
13:58 Verdict is a tiny classifier that reads what happened when an AI agent ran a command and decides whether it actually worked. It looks at the command, the exit code, and the output. Then labels the run as success, failure, blocked, or unknown. That matters because a command can exit cleanly and still fail to do its job. The whole model weighs about 200 kilob, answering in about half a millisecond with no network calls.
14:23 AI design- check is a set of 74 design rules, plus a checker for interfaces built by AI coding agents. Run it on your React view, spelt, or HTML files, and it points to the exact line where something breaks a rule, like a font size off your scale, an icon button with no label, or an emoji used as an icon. Other commands count duplicate button and card styles, and test your layout at phone width.
14:52 SCA is a control plane for running a swarm of local AI agents safely, written in Go as one static binary with no Damon and no database. You describe tasks in JSON with which ones depend on which, how many can run at once, and how long each one gets. Commands run directly, never through a hidden shell. Every batch of work belongs to an epoch.