📦 pngwn/twinkleplop

Twinkleplop

Twinkleplop is the lightweight syntax highlighter that finally makes code presentation fast and modular. Instead of forcing you to load giant, heavy bundles just to highlight a few lines of code, this clever toolkit lets you install only the exact languages your project needs. It is built to generate clean, lightweight HTML or raw tokens that you can style however you like using standard CSS. It includes custom grammars for popular languages like JavaScript, Rust, and Go, plus a built-in playground to test everything. Check out Twinkleplop to keep your site fast and your code looking sharp.

🆔 @hackernewsgithubprojects

📦 sail-sg/precision-rl

Precision RL: The Simple Switch for Better AI

Switching large language models to a different type of math during training makes them significantly better at reasoning, and the project called Precision RL proves it. When AI models learn from their own successes and failures, a tiny mismatch between how they calculate numbers during practice versus during real use often causes them to stumble. By simply changing the internal math to a more precise format, researchers found that the training process stabilizes, the model stops drifting off course, and it actually performs better on complex tasks. This isn't about adding new features, it's about fixing a hidden flaw in how we teach machines to think.

🆔 @hackernewsgithubprojects

📦 leter/zh-tech-writing

zh-tech-writing

Turn any AI-generated Chinese technical documentation into clear, human-readable prose by following the precise style guide established by the renowned developer阮一峰. This specialized agent skill eliminates the characteristic robotic tone often found in machine-written text, enforcing short sentences and plain language that actual readers can easily understand. It also enforces strict typographical standards for numbers, punctuation, and English terms to ensure the final document looks professionally published. The project solves the common frustration of editing awkward AI output by providing a ready-made set of rules that guarantee your technical writing feels natural, polished, and trustworthy every single time.

🆔 @hackernewsgithubprojects

📦 hazyresearch/hipkittens

HipKittens: The Fast AMD Kernel Toolkit

HipKittens is the low-level toolkit that makes AMD AI kernels genuinely fast. It gives developers the exact building blocks they need to write high-performance code for AMD hardware, stripping away the usual complexity. The real magic is how it handles memory and computation at once, letting the chip work on data while it is still loading the next batch. This keeps the hardware busy instead of waiting around. It is not just a library; it is a philosophy of building from the hardware up. If you care about making AI run faster on AMD, this is your starting point.

🆔 @hackernewsgithubprojects

📦 re1nhard-1/deconv-preflight

Deconv Preflight

Deconv Preflight will catch the exact moment your bulk RNA-seq table and single-cell reference stop agreeing, saving you from a silent mathematical disaster. It is a lightweight Python tool that acts as a strict gatekeeper, checking that every gene and cell ID matches perfectly before you run expensive deconvolution analysis. The surprising part is that it finds sneaky errors, like samples that technically exist but have zero expression on shared genes, which would otherwise ruin your results. Since it runs locally with no external dependencies, your sensitive data never leaves your machine. It does not guess or fix anything; it simply tells you exactly what is broken.

🆔 @hackernewsgithubprojects

📦 jasonwong1991/wonderbox

Wonderbox macOS System Utility

Wonderbox is the system utility that finally tells you the truth about your Mac's performance. While other cleaners just show a giant green checkmark and hope you do not look too closely, this tool uses honest metrics you can easily verify right in your Activity Monitor. Instead of just deleting temporary files, it triggers the actual system memory pressure broadcast to force background apps to release their compressed memory. It safely targets gigabytes of hidden browser and developer caches rather than useless megabytes, and even gives you direct fan control and keep awake toggles from your menu bar.

🆔 @hackernewsgithubprojects

📦 ipenywis/laya-ultrafast

Laya Ultrafast

Laya Ultrafast is the local browser agent that finally lets you run automated web tasks entirely on your own Mac without paying per-step API costs. Built specifically for Apple Silicon, this clever tool ports the popular browser-use project to run completely offline using a compact, open-weight decision model. Instead of sending every single page click and form fill to a expensive hosted API, it uses local hardware to make rapid-fire decisions in just thirty-three milliseconds. It is perfect for developers who want to automate flight searches or scrape websites privately, giving you full browser control without the constant cloud subscription fees.

🆔 @hackernewsgithubprojects

📦 luwudynamics/xgoduck_hardware

The Duck That Codes

XGO Duck is the two-legged robot that puts real coding power into a tiny, waddling duck. It’s not just a toy; it’s a full build kit featuring fifteen servos and a custom expansion board stacked on an Arduino Uno Q. This means the duck doesn’t just follow pre-programmed moves; it can learn to walk, fall, and pick things up thanks to a companion software project. The repo gives you the printable body, the circuit board, and a clear assembly guide, turning a simple microcontroller into a serious robotics lab.

🆔 @hackernewsgithubprojects

📦 mistralai/mistral-vibe

Mistral Vibe: Your New Terminal Buddy

Mistral Vibe lets you talk to your codebase like a person, not a compiler. It is a command line tool that listens to your natural language requests and actually does the work for you. It reads your files, fixes bugs, and runs tests while you watch. The coolest part is that it asks you questions when it is unsure, so you stay in control. It is not a magic fix, but it is a smart assistant that lives in your terminal. If you code, this is your new partner.

🆔 @hackernewsgithubprojects

📦 orneryd/nornicdb

NornicDB: The All-in-One AI Database

Query your data at lightning speed with NornicDB, a database that handles graph connections and vector searches in the same breath. Instead of juggling separate tools for structure and semantic meaning, this single engine processes both at once, keeping your data coherent and incredibly fast. It even remembers the history of your changes, so you can look back at exactly how your information evolved over time. This makes it perfect for building smart, memory-driven systems where context is everything. If you need a database that actually thinks about relationships, NornicDB is worth your time.

🆔 @hackernewsgithubprojects

📦 microsoft/debugmcp

DebugMCP

Give your AI coding assistant the power to actually run and debug your code instead of just guessing what went wrong. DebugMCP connects smart agents like GitHub Copilot, Claude Code, and Cursor directly to a real debugger inside VS Code or straight from your terminal. This means your AI can autonomously set breakpoints, step through your execution line by line, pause running programs, and inspect variables safely. It runs completely locally with built-in secret redaction to keep your credentials safe. Just install it and let your favorite AI agent find the actual root cause of your bugs in seconds.

🆔 @hackernewsgithubprojects

📦 nvidia/osmo

How OSMO Makes AI Clusters Simple

Orchestrate massive AI workloads across wildly different hardware without drowning in infrastructure code. OSMO, built by NVIDIA, takes the chaos of mixing training GPUs, simulation clusters, and edge devices and wraps it in a single, readable YAML file. You define your compute needs in plain text, and the platform handles the rest, spinning up resources exactly where you need them and tearing them down when you are done. It is like having a personal chef who knows exactly which pan to grab for every dish in your recipe. No more wrestling with cluster scripts or manual scheduling.

🆔 @hackernewsgithubprojects

📦 isas1/skills

Skills

You can now control exactly how Claude packages its summaries by installing two lightweight local skills that route entirely on what you want to end up holding. Instead of typing out long instructions or drowning in walls of text, these skills package plain-English shortcuts to give you exactly what you need. If you just want to understand something briefly, one skill delivers a short, plain-language answer directly in your chat. If you need a shareable page, the other skill instantly generates a single, self-contained HTML file with zero build steps or dependencies. It even lets you trigger visual themes like an accessible mode, an artistic design, or smooth animations.

🆔 @hackernewsgithubprojects

📦 ryana/jevify

Jevify

Jevify is the prompt that turns your coding agent into a ruthless auditor of your app's hidden costs. You paste it into your AI assistant while it works on your project, and it stops guessing to start hunting for where you are wasting money on slow, expensive language calls. It digs into your code to find places where you could swap brittle rules for fast, cheap semantic checks, unlocking features that were previously too costly to build. It forces a deep look at whether you are overpaying for basic decisions, helping you design a smarter, leaner system from the ground up.

🆔 @hackernewsgithubprojects

📦 yu1745/wetype-ime-linux

WeType for Linux

Run the WeType Android input engine directly on your Linux desktop. This project ports the popular mobile keyboard engine to the WeChat Input Method for Linux, letting you type using the same pinyin logic and dictionary data as your phone. It bridges the gap between mobile typing habits and desktop workflows by running the core engine through a compatible layer inside the Linux terminal. You get consistent candidate words and predictive text on your computer without switching apps. It is a clever way to keep your favorite typing flow on Linux.

🆔 @hackernewsgithubprojects

📦 sudharsanselvaraj/token-print

Token Print: See Your AI Model Think

Watch your language model think in real time. Token Print renders the actual math happening inside an AI as a live, three dimensional scene, letting you see attention heads and data streams move as the model generates words. This isn't a static diagram or a fake demo; it pulls live numbers from a real model to show you exactly what is happening. It solves the problem of AI being a black box by making the invisible process visible and tangible. If you want to truly understand how these models work, this tool turns abstract code into a concrete visual experience you can actually see.

🆔 @hackernewsgithubprojects

📦 jamespolyakov9829/captcha-api

captcha-api

Solve tough browser verification hurdles automatically using a lightweight Python client that runs with zero external dependencies. This project handles both Cloudflare Turnstile widgets and verification challenges inside a single interface. It stands out by matching your request with a precise browser identity block, including custom header sequences and fingerprint emulation, so your target website actually accepts the solved token instead of throwing it away. You can estimate your exact API costs beforehand, track spending in real-time, and run multi-threaded batch operations without writing complex concurrency logic. It is a clean, standard-library-only tool built to keep automated tasks running smoothly.

🆔 @hackernewsgithubprojects

📦 michael-jiahao-zhang/game-the-llm-reviewer

Game the LLM Reviewer

Peer reviewers powered by artificial intelligence often score the exact same scientific paragraph differently depending on how it is phrased. The tool called Game the LLM Reviewer solves this by making tiny, meaning-preserving edits to a finished paper. It rearranges sentences to highlight key contributions and frames data as clear gains, all without changing the actual science. This defensive rewrite helps ensure your research is judged on its merit, not its wording. It is a final polish for any author wanting to protect their work from algorithmic bias. Use this simple skill to make your paper impossible to misread before you submit it.

🆔 @hackernewsgithubprojects

📦 hirotomasato/yowes

Yowes

This lightweight tool can instantly generate high-resolution, realistic teacher verification documents like ID cards, employment letters, and teaching licenses for thirteen different countries. Known as yowes, it functions as a headless server using the Model Context Protocol, allowing AI assistants to directly create these files during a conversation. Instead of relying on generic placeholders, yowes features real school databases with local street addresses, consistent profile photos, and built-in fonts so that everything looks completely authentic right out of the box. It is an incredibly clever way to integrate automated, localized document generation straight into your favorite AI developer tools.

🆔 @hackernewsgithubprojects

📦 livecontext-ai/livecontext-ce

LiveContext: The Self-Hosted AI Automation Platform

LiveContext lets you describe a job in plain chat and instantly generates a working, self-hosted automation complete with a visual workflow, scoped AI agents, and a custom app interface. This open-source platform acts as a self-contained alternative to tools like Zapier or N8N, but with a major twist: it builds the entire system for you. Instead of stitching together separate chatbots and automation tools, you get a single canvas where the chat builds the logic, the agents execute the tasks, and the app displays the results. It is designed to be transparent, so you can see exactly what the AI is doing and how much it costs.

🆔 @hackernewsgithubprojects

📦 scenario-labs/skills

Skills: The Creative Toolset for AI Agents

One prompt can now spin up a fully consistent character, generate matching 3D meshes, and cut a finished music video, all without leaving your terminal. This repository, simply called Skills, turns coding assistants into professional creative studios. It teaches agents like Claude or Copilot to speak directly to a specialized media server, handling everything from pixel art and audio to complex video edits. The magic lies in consistency, where the same character looks identical across a hundred different scenes. If you build games or videos, this toolkit turns your code editor into a production line. It is the bridge between writing code and shipping actual creative work.

🆔 @hackernewsgithubprojects