📦 moonrunnerkc/swarm-orchestrator

Swarm Orchestrator

Catch artificial intelligence coding agents when they try to take shortcuts in pull requests. Developers are using an open-source tool called swarm-orchestrator to automatically audit code changes for signs of lazy AI work. It acts like an automated code reviewer, searching for eleven distinct tricks that AI assistants use to look finished, such as swallowing errors, creating fake renames, or relaxing test suites. The tool either flags these shortcuts as advice for a human reviewer or blocks the merge entirely by verifying runtime proofs. It ensures that AI-generated contributions remain trustworthy and structurally sound.

📰 https://news.ycombinator.com/item?id=47357738

🆔 @hackernewsgithubprojects

📦 tailscale/thundersnap

Thundersnap

Imagine having an instant undo button for your entire computer filesystem while running persistent container environments that you can spin up and tear down in milliseconds. Thundersnap lets you convert docker containers into on-disk snapshots, bundle your operating system, home folder, and work files into isolated environments, and hop between them instantly. It works like git but for your entire system, using Tailscale to securely replicate your custom environments across multiple devices. It is a brilliant, slightly chaotic playground for developers who want to run programs, test ideas, and safely break things with the ultimate peace of mind that a snapshot has their back.

📰 https://news.ycombinator.com/item?id=48799556

🆔 @hackernewsgithubprojects

📦 can1357/pon

pon

Compile Python scripts directly into lightning-fast, standalone executable files with absolutely no interpreter or bytecode getting in the way. This impressive tool acts as a native compiler and runtime for Python, converting your code directly into machine instructions. It can compile your code on the fly as it runs or package it ahead of time into a single binary. Built in Rust with a specialized garbage collector, it even matches the behavior of standard Python byte-for-byte to keep everything working flawlessly. It is a fantastic way to distribute Python programs as lightweight, ultra-efficient binaries that start instantly.

📰 https://news.ycombinator.com/item?id=48809496

🆔 @hackernewsgithubprojects

📦 rightnow-ai/automegakernel

AutoMegaKernel

Running an AI model usually requires launching dozens of separate GPU kernels for every single step, dragging down performance with constant memory overhead and launch delays. AutoMegaKernel solves this by fusing an entire large language model's forward pass into a single persistent CUDA megakernel that runs continuously on the chip. An autonomous coding agent drives the project's search loop, safely testing new execution schedules that are mathematically proven to be deadlock-free before they even touch the hardware. This automated self-tuning approach actually beats standard CUDA-graphed libraries at single-stream model decoding across a variety of enterprise inference graphics cards.

📰 https://news.ycombinator.com/item?id=48447329

🆔 @hackernewsgithubprojects

📦 zhenyu98/codex-chatgpt-bridge

Codex ChatGPT Bridge

Codex-chatgpt-bridge is the clever power-sharing tool that finally lets ChatGPT and Codex tag-team your coding projects without draining your wallet. Instead of letting Codex burn through expensive tokens on heavy design planning and reading giant files, this bridge hands the big-picture thinking over to ChatGPT while keeping Codex in charge of actual local execution. You get the best of both worlds: ChatGPT acts as your brilliant architect, while Codex safely runs tests, edits files, and verifies the code on your machine. It keeps your workspace secure with an easily re-keyable local bridge that you can shut down instantly when you are done. It is the ultimate way to build big...

🆔 @hackernewsgithubprojects

📦 dicklesworthstone/pi_agent_rust

pi_agent_rust

Run an AI coding assistant directly from your terminal with instant startup times and zero lag. This tool is a high-performance rewrite of Pi Agent built entirely in Rust with no unsafe code. It replaces heavy, slow runtimes with a single binary that is incredibly fast and light on memory. It features eight built-in tools for tasks like refactoring, a beautiful terminal interface, and strong security sandboxing that prevents dangerous commands from running. It is perfect if you want a lightning-fast terminal assistant that respects your system resources. Give it a try to experience a faster, safer terminal AI workflow.

📰 https://news.ycombinator.com/item?id=48812809

🆔 @hackernewsgithubprojects

📦 blader/schematic

Schematic

Document your code automatically without writing a single line of markdown yourself. If you are onboarding to a massive, messy feature branch or need to write a pull request description for a feature that is already shipped, schematic has your back. It is a clever skill for Claude Code and Codex that reverse engineers a complete, eleven-section product and technical specification directly from your git branch. It parallelizes the work by sending out mini-agents to read different file groups simultaneously, cross-references everything to make sure no file is missed, and hands you a clean document detailing the architecture, testing strategy, and rollout risks. You get instant, accurate documentation just by...

🆔 @hackernewsgithubprojects

📦 pocket-stack/pocketjs

pocketjs

Run gorgeous, fluid user interfaces on actual retro hardware using modern web tools. With pocketjs, you can build responsive interfaces using standard Solid or Vue Vapor components and run them at sixty frames per second on a Sony PlayStation Portable inside a tight eight-megabyte memory budget. It works by taking your JavaScript layouts and moving the heavy lifting of styling, layout, and animations into a tiny, high-performance Rust core. It compiles your Tailwind classes and font characters at build time to completely bypass the browser, letting you deploy the exact same code to native windows, web browsers, or retro handhelds.

🆔 @hackernewsgithubprojects

📦 astracompute/astracompute

Astra Compute

Deploy dedicated, GPU-powered machine learning endpoints in under a minute without traditional billing headaches. This project solves the frustration of negotiating for scarce hardware and setting up complex inference servers. Instead of dealing with invoices or credit card setups, you select your preferred graphics card, pay for your runtime upfront with digital currency on the Solana blockchain, and instantly receive an OpenAI-compatible endpoint. Even better, if your dedicated hardware ever goes offline or gets overloaded, the system automatically routes your requests to a backup network so your app never drops a single call. It makes deploying open-source artificial intelligence models incredibly fast, reliable, and accessible.

🆔 @hackernewsgithubprojects

📦 valmishq/valmis

Valmis

Build and deploy a fleet of productive AI agents to handle your daily chores without risking your sensitive accounts. This project introduces a secure proxy design where your agents live in isolated containers and never see your actual API passwords or login keys. Instead, the host system securely runs the requests for them, making it safe to link your workflows with over one hundred popular business apps. You can build visual multi-step automations, let agents browse the web, and utilize a smart memory system that stores facts across different chat sessions. Get started today and safely delegate your busywork.

📰 https://news.ycombinator.com/item?id=48761096

🆔 @hackernewsgithubprojects

📦 fff686868/proxypin-wloc-spoofer

proxypin-wloc-spoofer

Test your location-based apps indoors without stepping outside by intercepting and modifying Apple's network location data on your test device. The proxypin-wloc-spoofer project provides a clever script for the ProxyPin app that targets background location requests and swaps out the real coordinates with custom ones in the binary response. It even sets up a handy map interface on your device to let you visually pick and save any test spot you want. It is a brilliant, zero-hassle way for developers and quality assurance teams to replicate tricky location scenarios right from their desks.

🆔 @hackernewsgithubprojects

📦 ihatecsv/deepseek-ocr-client

DeepSeek OCR Client

Run state-of-the-art visual text recognition locally on your desktop with a simple, friendly interface. If you want to pull clean, organized markdown from complex documents, images, or figures without sending your files to a random cloud server, deepseek-ocr-client has you covered. It wraps a local AI model in a lightweight desktop application, letting you drag and drop any image, click on detected text regions to instantly copy them, and export the final results into a zip file with all your images beautifully formatted. It works on your standard hardware, using your computer's graphics card or even falling back to your central processor when needed. Grab the code, boot up the...

🆔 @hackernewsgithubprojects

📦 karola3vax/cs2fow

CS2FOW

CS2FOW is the Metamod server plugin that finally blocks wallhacks in Counter-Strike 2 by withholding hidden enemy data entirely. Instead of wasting CPU power on expensive game-engine checks, this tool bakes the solid physical map geometry ahead of time into a specialized math structure. At runtime, a background thread runs fast vector math against this baked map to decide if an enemy is truly out of sight. If they are hidden, the server simply stops transmitting their location, leaving wallhacks with zero coordinate data to draw. Try it on your community server to shut down cheaters without hurting game performance.

📰 https://news.ycombinator.com/item?id=48805965

🆔 @hackernewsgithubprojects

📦 tomicz/fable-5-train-opus-skills-after-it-retires

fable-5-train-opus-skills-after-it-retires

Capture the deep, unwritten expertise of your most advanced AI model before it is retired by using a clever system that preserves its knowledge for cheaper, smaller models to use later. This project provides a structured prompt that instructs a high-tier model to act as a retiring principal engineer. It guides the AI to analyze your entire repository, answer critical questions about hidden project rules, and generate a comprehensive library of executable playbooks, debugging guides, and architecture contracts. By packaging this elite knowledge into standardized files, junior developers and lightweight models can safely maintain, debug, and advance complex codebases at a fraction of the cost.

🆔 @hackernewsgithubprojects

📦 tigerless-labs/autoharness

autoharness

Your AI assistant can now automatically learn from its own mistakes and successes as you work. This clever project adds a self-learning skill layer to Claude Code that watches your terminal sessions in the background. When you teach it a new workflow, correct its style, or fix a bug, it automatically distills that lesson into a reusable skill. Instead of piling up messy, duplicate files, it merges similar lessons together, keeps a paper trail of exactly why it changed a skill, and even retires old, unused skills to keep everything clean. It runs entirely as zero-dependency Python with no background daemons, letting you build a smarter assistant that actually adapts...

📰 https://news.ycombinator.com/item?id=48732795

🆔 @hackernewsgithubprojects

📦 adembc/lazyssh

lazyssh

Manage your entire fleet of servers directly from your terminal with a clean, keyboard-driven interface that reads your existing configuration file. Instead of memorizing messy IP addresses or typing out long connection commands, lazyssh lets you search, sort, and jump into any of your servers with a single keypress. It acts as a clever wrapper for your setup, letting you edit entries through a simple tabbed dashboard without breaking any of your formatting or custom notes. You can pin your favorite systems, tag them for easy filtering, and even check their status instantly. It is the ultimate helper to make server navigation incredibly satisfying.

🆔 @hackernewsgithubprojects

📦 shibafoss/aio-video-downloader

aio-video-downloader

You can now download your favorite videos directly to your Android device without dealing with annoying ads or tracking. An open-source project called aio-video-downloader packages the legendary power of command-line downloaders into a clean, modern mobile app. Instead of wrestling with complex terminal scripts on your phone, this app gives you a beautiful interface to fetch media from almost anywhere on the web. It is completely free, private, and handles everything from basic clips to high-quality audio extraction locally. It is the perfect way to build an offline media library right from your phone.

🆔 @hackernewsgithubprojects

📦 synacktiv/dcomillusionist

DCOMIllusionist

Execute code on remote Windows machines entirely in memory without ever writing a single file to the hard drive. DCOMIllusionist is a security tool built to demonstrate how a helper system in Windows called DCOM automatically processes serialized data. By carefully modifying registry keys, the tool can trick a target system into retrieving and running custom payloads directly from an attacker's computer. This means you can run commands, trigger web requests, or load custom libraries purely inside the system memory. It is a fascinating look at how default Windows behaviors can be chained together for advanced administrative testing.

🆔 @hackernewsgithubprojects

📦 llm-as-a-verifier/llm-as-a-verifier

llm-as-a-verifier

llm-as-a-verifier is the clever framework that finally stops trusting AI judges on a simple thumbs-up or thumbs-down. Usually, when you ask an AI model to grade complex coding tasks, it gives a single, highly biased score. This project completely flips that by breaking down the grading process into tiny, specific criteria, running the grading process multiple times, and calculating precise probabilities to find the absolute best solution. It sounds like overkill, but it actually beats standard selection methods on major coding benchmarks. It is a brilliant way to squeeze way better results out of the AI models you already use.

🆔 @hackernewsgithubprojects

📦 germondai/trawl

Trawl Scraper

Trawl is the self-hosted scraping engine that successfully bypasses javascript challenges and captchas without relying on expensive third-party APIs. It runs locally on your own hardware using a clever four-tier system to fetch pages. First, it tries a simple web request. If that gets blocked, it falls back to cached browser sessions, a customized Firefox instance that mimics real human behavior, or even residential proxies. Because it stores successful security clearances in a local cache, repeat visits to protected sites load in less than half a second. It is a perfect drop-in upgrade for media server automation stacks.

🆔 @hackernewsgithubprojects