📦 v4bel/infinitycore

Infinity Core: Six Linux Kernel Exploits

Infinity Core is the security paper that finally exposes deep flaws in the heart of the Linux kernel. It documents six critical bugs in core subsystems like timers and memory management that let a regular user gain full root access on major systems like Ubuntu and Samsung phones. The most surprising part is how a simple thread switch can corrupt the system password file, turning a normal account into a superuser without touching the network stack. This isn't just theory; it's a proven path to total control over your device.

📰 https://news.ycombinator.com/item?id=50125711

🆔 @hackernewsgithubprojects

📦 xenophobentx/takedawn

Takedawn: Burnout 3 on PC

Takedawn lets you run the classic Xbox racing game Burnout Three Takedown directly on your PC, Mac, Steam Deck, or even your Android handheld. This isn't a remaster or an emulator; it is a static recompilation that translates the original console code into a native program, meaning you get the exact same cars, tracks, and physics as the 2004 original, but without the hardware limits. The coolest part is how it handles visuals on modern screens. You can crank the resolution up to four times the original, enable true high dynamic range lighting so headlights and sparks actually glow, and race at a locked sixty frames per second.

🆔 @hackernewsgithubprojects

📦 morganlinton/vulcanbench

VulcanBench: The Honest LLM Test

VulcanBench measures how well coding agents actually write software, not just how they chat. It runs a model against a task, keeps the hidden tests away from the agent, and grades the result deterministically. This matters because you can finally see if a model is genuinely useful or just confident. The repository records every step, from the full trace to the final code patch, so you can verify the results yourself. It tests models through real products like Claude Code or raw APIs, ensuring you know exactly what you are paying for.

🆔 @hackernewsgithubprojects

📦 xplo8e/liter8

Liter8: The Tethered iOS 27 Jailbreak

Liter8 lets you patch and boot a custom iOS 27 build on an iPhone 11 directly from a Mac. It works by taking a standard firmware file and surgically modifying its internal code to bypass security restrictions. This tool is interesting because it does not rely on fixed memory addresses. Instead, it scans the specific firmware binary to find the exact locations that need changing. This makes the jailbreak adaptable to different software versions without breaking. It is currently only tested on the iPhone 11 and requires your Mac to stay connected every time you start the phone.

🆔 @hackernewsgithubprojects

📦 gichirimaku/wp2shell-poc

How WordPress Just Got Hacked

wp2shell is the proof of concept that turns two weird WordPress bugs into a total takeover. It combines a routing mistake and a database flaw to let attackers slip past security checks without even logging in. The coolest part is how it forges fake data to trick the site into handing back a real admin key, letting them upload a malicious plugin right from the outside. This repo is strictly for testing your own sites, so patch those versions now before someone else finds you first.

🆔 @hackernewsgithubprojects

📦 obsidianmd/pele

Pele renders Mermaid diagrams instantly

Pele renders Mermaid diagrams to SVG faster than you can run from a volcano. It solves the problem of slow, heavy chart libraries that make your app feel sluggish. Instead of waiting for complex background processes to finish, Pele draws your flowcharts and sequence diagrams synchronously right on your screen. It does this with zero runtime dependencies, meaning you install one tiny file and you are ready to go. You also get full control over colors and fonts using standard CSS variables, so your diagrams match your website theme perfectly. If you need diagrams that load instantly without breaking your bundle size, Pele is the simple, fast tool you need.

🆔 @hackernewsgithubprojects

📦 steventsvik/tellurion

Tellurion: Live Mission Control for AI Coding Agents

Tellurion is a macOS terminal that lets you watch multiple AI coding agents work live, displaying exactly which files they edit and where they are in your project. It acts like a mission control, showing real-time status for tools like Claude Code and Codex so you can see their progress without constantly switching tabs. The editor even follows their edits as they happen, and you can undo any change with one click. If an agent needs permission, you can approve it right from your phone. This is useful for developers running several projects at once who need a clear view of what their AI is doing.

🆔 @hackernewsgithubprojects

📦 dacooderr/optilock

OptiLock: Boost Deadlock FPS on Low-End PCs

OptiLock is the Deadlock FPS optimizer that finally keeps your frame rate high without ruining the competitive look of the game. It solves the problem of stuttering on low-end systems by lowering graphics settings to their absolute minimum while preserving essential visual clarity for ranked play. The most surprising part is that it can deliver up to forty percent more frames, turning a struggling machine into a smooth, competitive rig. You simply swap in a provided file and paste a specific code into your game settings. If you are fighting lag, this is the clear path to a steady, high frame rate.

🆔 @hackernewsgithubprojects

📦 oxc-project/tsrs

tsrs: The Rust TypeScript Checker

The TypeScript compiler, the engine that checks your code for errors, is now running on Rust. It is called tsrs, and it is a drop-in replacement for the current compiler. This means you can swap it in without changing your workflow. The goal is to match TypeScript exactly while using far less memory and checking your files much faster. It is not a rewrite of the language, but a new engine for the same job. It is already usable today, and the creators are tracking one hundred percent conformance. If you work with large JavaScript projects, this is the kind of speed boost you have been waiting for.

🆔 @hackernewsgithubprojects

📦 msuiche/weightless

Weightless: The Tiny File That Changes How AI Models Behave

A tiny file, containing no model weights at all, can fundamentally alter how a massive artificial intelligence system behaves. This project, called Weightless, solves a massive logistical headache: it lets developers tweak a model's personality or safety boundaries without downloading, copying, or modifying the enormous, multi-terabyte brain that powers it. Instead of juggling giant files, you just inject a small, directional nudge into the model's live memory stream while it runs. Think of it like turning a steering wheel on a moving car, rather than rebuilding the entire engine. It works even on compressed, low-precision models, which means you can test new behaviors on cheap hardware without breaking a sweat.

🆔 @hackernewsgithubprojects

📦 mikroverk/esp32sim

esp32sim: Run Real Firmware in Your Browser

Your browser can now boot actual firmware onto four different microchips without needing a single piece of physical hardware. The project called esp32sim runs real, unmodified software images directly in your web page or on your computer. It handles the complex startup process exactly like the real chips do, from the initial boot code all the way to running full operating systems. You can even connect it to your home Wi-Fi to test network features. This removes the need for expensive dev boards when you are just checking if your code works. It is a free, open-source tool that lets you build and test embedded projects entirely from your desktop.

📰 https://news.ycombinator.com/item?id=50131189

🆔 @hackernewsgithubprojects

📦 depalmar/ai-dfir-toolkit

AI DFIR Toolkit: Spot AI Threats

The AI DFIR Toolkit is the open source detection pack that finally gives security teams the ability to catch attacks on large language models and agent systems without being locked into expensive proprietary software. It solves a massive blind spot in modern cybersecurity, where traditional antivirus tools are completely blind to the new ways hackers exploit artificial intelligence. Instead of relying on vendor specific queries, this project provides one hundred and sixty one universal detection rules that work across any logging system you might have. It covers everything from prompt injection and model supply chain poisoning to malicious behavior in coding assistants.

🆔 @hackernewsgithubprojects

📦 gathin23/tv-vibehack

TV Vibehack: Your TV, Minus the Ads

TV Vibehack turns a cluttered Android TV into a clean, ad-free experience by removing unnecessary software and installing a custom launcher that looks exactly like Apple TV. This project solves the problem of slow, bloated smart TVs by letting you use an AI assistant to identify and disable manufacturer ads, telemetry, and unwanted preloaded apps directly from your computer. The result is a pitch-black interface with a simple dock for your favorite apps, making the TV faster and much easier to use. The best part is that every change is completely reversible, meaning you can restore the factory settings anytime without losing your setup.

🆔 @hackernewsgithubprojects

📦 cygnux-labs/tracekit

Tracekit: The Flight Recorder for AI Agents

An AI agent can edit its own logs, but it cannot hide that edit. Tracekit solves this by acting like a flight recorder for your AI agents. It checks every action before it happens, signs each decision into a tamper-evident chain, and produces a single file anyone can verify offline. This matters because if an agent deletes data or changes its records, the verification fails instantly at the exact point of tampering. You get proof of what the agent actually did and why it was allowed, without trusting the agent itself. It is a clear, verifiable audit trail for AI behavior.

🆔 @hackernewsgithubprojects

📦 holdmyspot-com/cat

CAT: Making AI Coding Agents Reliable

Keep your AI coding agent from randomly ignoring your team’s hard-earned rules. CAT solves the annoying problem where an agent follows a convention in one file but forgets it in the next. Instead of hoping the model behaves, this tool tests your project rules until they are proven to work ninety-five percent of the time. It then hands those reliable instructions directly to coding tools like Claude Code, ensuring your agent actually follows them. If a reviewer finds a gap, you simply add the new lesson to your shared rulebook.

🆔 @hackernewsgithubprojects

📦 lcoutodemos/chums

Chums: Your Coding Agents on Your Wrist

Chums is the Apple Watch app that turns your existing coding agents into animated characters on your wrist. It connects the Claude Code and Codex tools already running on your Mac to your watch and iPhone, letting you talk to them hands-free. The most striking part is how each agent acts out its work. You hold the screen to speak, and the character visually shifts through states like listening, thinking, or working. This gives you real-time feedback without opening a laptop. It is free, open-source, and runs entirely on your own hardware. No new subscriptions are required.

🆔 @hackernewsgithubprojects

📦 delta-box/deltabox-lite

Deltabox Lite: The Undo Button for AI Agents

Deltabox Lite is the sandbox tool that gives AI coding agents a safety net by treating every file change like a reversible draft. Instead of letting an automated program mess up your project without a way back, this lightweight system wraps commands in a confined space where you can see exactly what changed. If the output looks wrong, you simply abort and your files snap back to how they were before, just like pressing undo. This solves the anxiety of running powerful, autonomous code by giving you total control over what actually sticks to your disk.

📰 https://news.ycombinator.com/item?id=50104406

🆔 @hackernewsgithubprojects

📦 cranebeaker/coop-anything

Co-op Anything: Make Any Game Multiplayer With AI

Your AI coding agent can now turn any single-player game into a co-op experience for you and your friends. The project, called Co-op Anything, acts as a detailed playbook that instructs your agent to research, reverse engineer, and fix existing mods or build new ones from scratch. It handles the complex technical work, including setting up networking so you can join a game over Steam without any port forwarding or extra purchases. The most interesting part is that it tests its work automatically using bots inside the actual game to ensure the co-op actually functions before you ever install it.

🆔 @hackernewsgithubprojects

📦 eecs498-aase/materials

materials: The Elephant-Goldfish Model

The course materials for UMich's Applied Agentic Software Engineering that teaches you to build software with AI agents using the Elephant-Goldfish Model. It is a collection of lecture slides and hands-on labs where you don't just watch videos, you actually drive the coding process. The standout feature is the skill system, which breaks down complex engineering tasks into manageable steps called peanuts, elephant, goldfish, and mean-review. This approach turns vague prompts into precise, repeatable workflows, showing you exactly how to guide an AI to write clean, working code.

🆔 @hackernewsgithubprojects

📦 yokimitsuro/khdays-port

Run Kingdom Hearts 358/2 Days natively on your PC

Play Kingdom Hearts 358/2 Days natively on your PC with a project called khdays-port. This experimental runtime takes the game code from a full decompilation and runs it directly on modern hardware, bypassing the Nintendo DS entirely. It replaces the original console hardware with new graphics, sound, and input layers, letting you enjoy the game at higher resolutions with modern controller support. The engine handles rendering and audio independently, so you get a crisp, high-definition experience without the original system limits. If you are curious about how game preservation works, this is a fascinating look at rebuilding a classic title from the ground up for today's machines.

🆔 @hackernewsgithubprojects

📦 midudev/better-claude

Better Claude: Queue Tasks Like a Pro

Better Claude solves a frustrating problem in terminal based coding: the mess that happens when you type new instructions while the AI is still working on the first one. Usually, that new text gets injected immediately and changes the AI's current plan, causing it to mix up priorities and get confused. This project fixes that by letting you type a command called After to queue up tasks for later. You can keep typing while the AI works, and your new instructions wait safely in the background. When the current job finishes, the next task runs automatically.

🆔 @hackernewsgithubprojects