📦 homeassistant-ai/skills

Skills

Teach your AI coding agents to configure Home Assistant correctly without guessing or making up messy custom code. The skills repository is a collection of portable markdown knowledge packs built specifically to guide AI assistants through smart home setup. Instead of letting an AI generate fragile templates that fail silently, this tool equips your agent with direct, official best practices for choosing native helpers, structuring automations, and handling safe refactorings. It works directly with tools like Claude Code and Cursor. You simply install the package, and your AI assistant instantly knows how to design reliable smart home configurations.

🆔 @hackernewsgithubprojects

📦 evilsocket/nyx

Nyx

Nyx is the self-contained script that thoroughly scrubs digital footprints across Linux, macOS, and Windows. Instead of manually digging through obscure directories to clean up your system, this clever tool targets everything from shell histories and system logs to browser caches and application traces with a single command. The absolute coolest thing about it is how deeply it understands each platform, scrubbing deep-seated artifacts like Windows event logs, macOS unified logs, and Linux package manager histories. It is basically a digital leaf blower for your operating system, perfect for when you need a completely fresh, trace-free state without reinstalling everything from scratch.

🆔 @hackernewsgithubprojects

📦 nickustinov/itsyhome-macos

itsyhome-macos

You can now control your entire smart home directly from your Mac's menu bar without ever opening a bulky app. itsyhome-macos is a native menu bar application that bridges the gap between your Mac, Apple HomeKit, and Home Assistant. It puts your lights, thermostats, and security systems just a click away, but the coolest part is that it includes a built-in webhook server. This means you can trigger smart home scenes and toggle devices using simple terminal commands, keyboard shortcuts, or even your Stream Deck. It is the ultimate tool for turning your Mac into a powerful smart home command center.

🆔 @hackernewsgithubprojects

📦 huangchihhungleo/claude-real-video

Claude Real Video

Give your favorite AI model actual eyes to watch videos by running a clever script right on your own computer. Instead of blindly guessing or over-sampling a static screen every single second, claude-real-video detects real scene changes and uses a sliding pixel comparison to drop near-duplicate cuts. It packages these key visual frames with a synchronized text transcript, meaning you feed the AI fewer, highly meaningful images for much cheaper context costs. Just point it at a local file or web link, drop the output folder into Claude, and start asking questions about what actually happened on screen.

📰 https://news.ycombinator.com/item?id=48766005

🆔 @hackernewsgithubprojects

📦 thenetworkchuck/frigate-nvr-guide

frigate-nvr-guide

Your wireless home security cameras might actually be choking your Wi-Fi network to death, even if your internet bandwidth looks completely fine. The culprit is airtime saturation, where dozens of video streams constantly collide and force endless packet retries. The frigate-nvr-guide is a complete, hands-on blueprint for building a private, local AI surveillance system that solves this. It walks you through setting up Frigate NVR using a Raspberry Pi or an old desktop, configuring smart object detection that processes everything locally, and optimizing your network settings to prevent Wi-Fi meltdowns. Grab your hardware, follow the steps, and take back control of your home privacy today.

🆔 @hackernewsgithubprojects

📦 hkust-nlp/toolathlon

Toolathlon

Put your AI agents through a brutal, multi-day digital decathlon to see if they can actually survive real-world office chaos. While most AI benchmarks test models on simple, single-turn questions, this repository challenges agents with complex, long-horizon tasks across real software tools like Canvas, Slack, and email systems. It forces models to navigate messy, multi-step scenarios, like grading programming homework or organizing a university exam schedule without breaking. If you want to build AI that does actual, multi-day work instead of just answering trivia, this is the ultimate playground to test their limits. See how your agent stacks up today.

🆔 @hackernewsgithubprojects

📦 sanyuan0704/sanyuan-skills

sanyuan-skills

You can now transform your terminal-based AI agents into specialized powerhouses with a curated library of production-ready skills. The sanyuan-skills repository provides a collection of installable tools designed for command-line assistants like Claude Code. Instead of dealing with generic AI responses, these specialized skills add structured capabilities like rigorous code reviews, automated wiki building, and even an interactive reading coach that uses spaced repetition and Socratic testing. You can easily add these tools to your terminal workflow to instantly handle complex software audits, study technical books systematically, or build custom knowledge bases. It is the perfect way to make command-line AI agents genuinely practical.

🆔 @hackernewsgithubprojects

📦 k-dense-ai/science-superpowers

Science Superpowers

Science Superpowers is the agentic framework that finally brings real scientific discipline to your AI research assistants. Instead of letting your AI blindly run code and hunt for patterns in your data, this clever toolkit forces it to act like a rigorous scientist. The moment you ask a research question, the agent pauses to frame a falsifiable hypothesis and locks in its predictions before it ever looks at the data. By pre-registering these decision rules first, it completely prevents common statistical traps like p-hacking. It then executes the analysis in a clean, reproducible sandbox, rigorously investigates anomalies, and even unleashes a virtual red-team agent to shred its own conclusions before...

🆔 @hackernewsgithubprojects

📦 3sk1nt4n/cybersentinel-ai

cybersentinel-ai

Most security chatbots just suggest terminal commands, but cybersentinel-ai actually launches thirty-three real security tools inside an isolated sandbox container to verify threats in real time. This local security platform lets you swap between multiple artificial intelligence models mid-conversation while executing live vulnerability scans, mapping attack paths on a visual graph, and analyzing real event logs. It solves the massive problem of AI hallucination in cybersecurity by grounding every conversation in real tool outputs, compliance standards, and live intelligence feeds. If you want a secure environment to run scans and hunt threats completely offline without cloud subscriptions, this project lets you deploy a full security orchestration stack on your own...

🆔 @hackernewsgithubprojects

📦 416rehman/deepzero

DeepZero

You can now hunt for security flaws in Windows kernel drivers while you sleep. A smart project called deepzero lets you build automated pipelines using simple text configuration files to handle the heavy lifting. Instead of manually inspecting files, you can chain together steps to automatically parse files, decompile complex binary code, and run deep language model assessments over thousands of drivers at once. It runs tasks in parallel and saves its state safely to your drive, meaning you can stop the process anytime and resume without losing progress. It is a brilliant way to let smart AI agents do the tedious work of finding exploitable code.

🆔 @hackernewsgithubprojects

📦 talkcody/talkcody

TalkCody

This open-source coding agent lets you control your entire software development workflow remotely using chat apps like Telegram and Feishu. Instead of staying glued to a heavy desktop editor, talkcody connects directly to your local workspace, allowing you to trigger complex programming tasks, execute terminal commands, and review code diffs straight from your phone. It runs as a local client to keep your files secure while integrating with multiple language models and custom tools. By bridging the gap between local codebases and remote chat interfaces, it turns mobile messaging into a fully functional development environment for managing code on the go.

🆔 @hackernewsgithubprojects

📦 norma-core/norma-core

NormaCore

You can now run entire physical robotic setups, from data collection to AI model execution, using a single unified toolkit. norma-core is an open source platform that provides everything you need to build and manage hardware ecosystems, featuring its own 3D-printed robotic arm and custom jaw gripper. Instead of wrestling with fragmented drivers, it unifies high-performance data pipelines, a web interface, and an SDK to let you record structured datasets for imitation learning and even fine-tune AI control policies. It is a game changer for anyone wanting to jump straight into training real robots without the massive enterprise price tag.

🆔 @hackernewsgithubprojects

📦 mrt4ntr4/ntwarden

NtWarden

You can now inspect hidden processes and detect memory hollowing directly from a clean, neon-themed desktop interface. ntwarden is a Windows system analysis and research toolkit that combines a polished user interface with deep kernel-level diagnostics. It helps developers and security researchers examine active processes, verify driver signatures, and analyze running memory for anomalies like unbacked shellcode or direct syscalls. By pairing a custom kernel driver with an intuitive GUI, it uncovers hidden activities that standard task managers completely miss. Grab ntwarden to get an incredibly clear, deep-dive view into the absolute lowest levels of your Windows operating system.

🆔 @hackernewsgithubprojects

📦 tylergraydev/claude-code-tool-manager

Claude Code Tool Manager

Managing configuration files for artificial intelligence coding assistants usually means manually editing scattered JSON files across your system. The desktop app claude-code-tool-manager changes that by providing a single, clean visual interface to orchestrate your entire local setup. It automatically syncs your Model Context Protocol tools, custom slash commands, and specialized agent skills across multiple development tools simultaneously, including Cursor and Claude Code. You can visually toggle tools on a per-project basis, run instant connection tests to verify new servers, and even let your AI assistant configure its own tools programmatically. It is a brilliant way to bring order to your local development environment.

🆔 @hackernewsgithubprojects

📦 alexanderwanyoike/the0

the0

You can write your algorithmic trading bots in almost any major language, deploy them to an isolated self-hosted engine, and automatically generate custom React dashboards to monitor them. That is the magic of the0, an open-source platform that breaks down the walls of quant trading. Instead of forcing you into a single ecosystem, it supports everything from Python and TypeScript to Rust, C++, and Haskell. It runs your strategies in clean, isolated environments while giving you full control over your private financial data. It even includes a built-in assistant integration so you can deploy bots and inspect logs using simple, everyday natural language.

🆔 @hackernewsgithubprojects

📦 ddalcu/mlx-serve

mlx-serve

Run any open-source artificial intelligence model directly on your Mac at blisteringly fast speeds without ever needing Python or the cloud. This project is mlx-serve, a native engine built specifically for Apple Silicon that serves both MLX and GGUF models. It is significantly faster than standard tools and behaves exactly like OpenAI, Anthropic, or Ollama APIs, making it a drop-in replacement for your favorite developer apps. The most incredible part is its sandboxed agent mode, which lets an AI agent run terminal commands inside an isolated, secure Linux virtual machine that boots in under a second. Take control of your local AI workflow with mlx-serve.

🆔 @hackernewsgithubprojects

📦 govctl-org/govctl

Govctl

Enforce strict quality control on AI-generated code by putting a local governance control plane directly in your repository. The govctl tool helps you manage AI-assisted development by translating goals into structured, trackable files like RFCs, decisions, and work items. Instead of letting AI agents randomly write code and drift away from your design, you can set up executable verification gates that force the AI to prove its work actually passes your standards before anything gets merged. It is a fantastic way to keep your codebase clean and structured even when bots are writing the code. Give it a spin and keep your AI helpers on a short, reliable leash.

🆔 @hackernewsgithubprojects

📦 eigent-ai/toolathlon_gym

toolathlon_gym

AI agents struggle with complex real-world tasks, but toolathlon_gym lets you stress-test them across over five hundred automated scenarios using only local databases instead of brittle, live APIs. It packages twenty-five different tool servers into isolated Docker environments, forcing your agent to perform multi-step jobs like pulling mock database records, writing spreadsheets, and sending calendar invites. Because it runs entirely offline on a local database, your tests won't break when external APIs change. It is basically a local playground to see if your AI can actually handle messy, multi-tool workflows. Grab the code and see if your agent survives.

🆔 @hackernewsgithubprojects

📦 rexglue/rexglue-sdk

rexglue-sdk

The Xbox 360 recompilation toolkit that translates classic console games into native PC code. Instead of running games through a slow, resource-heavy emulator, rexglue-sdk analyzes the original PowerPC instructions and translates them directly into highly efficient code your PC can run natively. It builds a bridge between two completely different eras of hardware by handling the console's unique audio decoders, custom graphics processors, and complex file systems right inside a modern runtime. This means classic titles can finally run with incredible performance and native precision. If you love deep retro-engineering, this is the ultimate way to bring legendary console games into the modern PC era.

🆔 @hackernewsgithubprojects

📦 sgharlow/claude-code-recipes

claude-code-recipes

You can now handle complex office operations and technical documentation using a pre-configured library of one hundred step-by-step prompts. The open-source project claude-code-recipes provides a massive collection of targeted templates designed specifically for daily business tasks. Instead of guessing how to prompt your AI assistant, you get structured guides categorized into ten distinct professional areas like strategic analysis, human resources, and project management. It simplifies everything from turning messy meeting notes into clear action items to writing detailed technical specifications. It is a highly practical reference book for anyone looking to make their everyday digital workload much lighter.

🆔 @hackernewsgithubprojects

📦 sdc17/swireasoning

SwiReasoning

Stop your reasoning AI models from overthinking and wasting computing power on simple questions. It turns out you do not need to retrain a massive model just to make its thought process more efficient. Instead, this project introduces a clever, training-free method that lets AI dynamically switch between thinking silently in the background and spelling out its steps explicitly. By limiting the number of times the model switches back and forth, you get the same accurate answers in a fraction of the time. Grab the code, plug in your favorite model, and enjoy much faster, smarter reasoning without the endless waiting.

🆔 @hackernewsgithubprojects