📦 ksimback/looper

Design Better Agent Loops Before You Run Them

Looper is a design tool for Claude Code that prevents bad automation by requiring you to build and verify a loop structure before execution begins. Instead of letting an agent run unchecked, this skill interviews you to define clear goals, programmatic checks, and review gates, often involving a second model to provide an objective critique. It generates a portable specification, including a structured plan and termination guards, which can be run immediately in your session or later via an external runner. This approach helps ensure your agents have a clear definition of done and reliable error handling.

🆔 @hackernewsgithubprojects

📦 ahmaddarkhalil/hoi-detr

HOI-DETR: Detecting Hand-Object Interactions with AI

HOI-DETR is the transformer-based framework that accurately detects hands, held objects, and their complex interactions in a single forward pass. By building on the powerful Co-DETR architecture, it introduces a specialized interaction module capable of identifying both primary objects touched by a hand and secondary tools acted upon through those objects. This system significantly improves detection benchmarks, offering a robust solution for understanding intricate manual tasks. It even includes refined annotations to correct common dataset errors, making it a reliable tool for researchers looking to advance computer vision. Explore this repository to master advanced interaction detection for your projects.

🆔 @hackernewsgithubprojects

📦 ahmetvural79/camerarealtimestyle

Control Your Camera Styles With Hand Gestures

Your webcam feed instantly shifts into an artistic style like anime or claymation simply by pointing your index finger left or right. This project uses browser-based hand tracking with Google MediaPipe to recognize your gestures and trigger real-time transformations via the flux-2-klein model. By detecting whether you point or hold an open hand to reset, it provides a completely hands-free way to cycle through visual filters. It is a clever demonstration of integrating low-latency AI streaming with gesture control to create interactive video effects. Grab the code to customize your own creative webcam experience today.

🆔 @hackernewsgithubprojects

📦 mohitsoni48/turbollm

Run Local LLMs Fast with Zero Hassle

Optimize and run any local LLM engine directly on your hardware without the overhead of heavy software like Electron or Python. This tool provides a polished web interface and an API that is fully compatible with OpenAI and Anthropic, allowing you to run advanced tools like Claude Code on your own machine. It automatically benchmarks your GPU to apply the fastest settings, ensures your VRAM is utilized effectively, and supports community-built engine forks for maximum performance. Whether you are managing multiple models or sharing them across your network, this lightweight, offline-first application puts complete control over your local AI infrastructure.

🆔 @hackernewsgithubprojects

📦 jacob-bd/the-ai-counsel

Stop Trusting Just One AI: Build Your Own Deliberation Council

The AI Counsel replaces single-model responses with a multi-model system that uses structured deliberation and debate to deliver higher-quality answers. Instead of relying on one AI, this tool orchestrates several models to independently solve problems, peer-review each other, and synthesize a final, vetted conclusion. You can choose between a council mode for objective synthesis or an advisor mode where named personas, like a strategist or a skeptic, debate your questions in rounds to uncover risks and trade-offs. This project supports various cloud and local model providers, web search integration, and custom persona configurations for specialized decision-making, helping you arrive at better results.

🆔 @hackernewsgithubprojects

📦 dweep-desai/facegate-mac

Secure Your Mac Apps With Face Recognition

FaceGate acts as a native app-locker for macOS that lets you secure individual applications using face recognition, Touch ID, or a custom password. This utility intercepts app launches in real-time and requires authentication before granting access, effectively bridging a gap in native macOS security. The entire recognition pipeline runs locally on your device, leveraging the Apple Neural Engine to process face embeddings without any data leaving your machine. With features like liveness detection to prevent spoofing and per-app session timers, it offers a practical layer of privacy. Install it to add a modern authentication barrier to your sensitive applications.

🆔 @hackernewsgithubprojects

📦 imlixinyang/flashworld

Generate 3D Scenes in Seconds with FlashWorld

A detailed 3D environment materializes on screen almost instantly from a single text prompt or image input. FlashWorld is a powerful research project that enables high-quality 3D scene generation in mere seconds by leveraging advanced neural rendering techniques. This tool is designed to solve the challenge of slow scene creation by providing a fast, efficient workflow for developers and creators. It offers both a local web interface and a command line tool, allowing users to export results in versatile formats like PLY and SPZ. Explore this repository to accelerate your 3D generation tasks and streamline your creative pipeline today.

🆔 @hackernewsgithubprojects

📦 ffroliva/gflow-cli

Automate Google Flow Video Generation from Your Terminal

Control Google Flow directly from your terminal to generate high-quality AI images and videos without ever opening the web interface. This unofficial tool allows Ultra and Pro subscribers to batch process prompts efficiently, reuse characters consistently across different shots, and build automated content pipelines. By using a persistent Playwright session to handle authentication and reCAPTCHA, it streamlines your workflow while maintaining the full power of the underlying models. It is an excellent way to turn repetitive generation tasks into simple scripts. Install this command-line utility today to start integrating professional AI generation directly into your own automation stack.

🆔 @hackernewsgithubprojects

📦 cassie07/agentskill_survey

Understanding Agent Skill Evaluation and Evolution

Navigate the landscape of artificial intelligence agent development by exploring systematic frameworks for how agents create, use, and evolve their internal skills. This repository serves as a comprehensive survey that organizes modern research into clear paradigms like execution feedback, trajectory distillation, and reinforcement learning. By centralizing references for various skill-centric benchmarks and general-domain evaluation methods, it highlights structural gaps in current testing metrics to help researchers improve agent reliability and safety. Use this resource to better understand the lifecycle of agentic systems and identify key open challenges for your next project in this evolving field.

🆔 @hackernewsgithubprojects

📦 samasante/liquid-glass

Real Apple-Style Liquid Glass for the Web

Create immersive glass effects by refracting your live web content directly within React. This library provides a headless, zero-dependency lens that uses SVG displacement filters to bend, tint, and frost elements while keeping text selectable and links functional. Unlike static snapshots, it works seamlessly across Chrome, Safari, and Firefox. You can easily define custom lens geometry or use built-in motion utilities to create interactive UI components like sliders and switches that feel smooth and native. By handling the complex cross-browser physics for you, this tool helps you bring high-quality, liquid optics to your own web projects.

🆔 @hackernewsgithubprojects

📦 am-will/codex-app

Getting the OpenAI Codex Desktop App Running on Linux

You see a Linux terminal running a complex packaging process that unpacks a proprietary installer into an active, functional desktop environment. This repository serves as a specialized reconstruction workspace and Electron starter designed to bring the OpenAI Codex desktop application to Linux users. It tracks the necessary packaging pipeline and provides a controlled environment to port patches, manage protocol registrations, and verify builds for platforms like Ubuntu. By automating the extraction and compatibility logic, it enables developers to run this otherwise macOS-centric tool on their preferred Linux distributions. Explore this project to see how to enable cross-platform agentic coding.

🆔 @hackernewsgithubprojects

📦 off-grid-ai/mobile

Run AI Entirely Offline on Your Phone

Run advanced artificial intelligence models directly on your smartphone without needing an internet connection or cloud access. This project functions as a comprehensive mobile suite that keeps all your data private by processing everything locally on your device hardware. You can engage in text conversations, generate images using models like Stable Diffusion, perform real-time speech-to-text transcription, and analyze documents or camera feeds with vision capabilities. By keeping your information strictly on your phone, it ensures total privacy while providing tools for document retrieval and task automation. Take complete control of your personal AI experience today.

🆔 @hackernewsgithubprojects

📦 cashmeredev/kitty-graphics.el

Bring Images and Video Directly Into Terminal Emacs

View images, play videos, and see scaled text directly inside terminal Emacs using the Kitty graphics protocol or Sixel. This project automatically detects your terminal's capabilities at startup to provide seamless multimedia support without extra configuration. It works across many different terminals and even persists inside tmux. Beyond standard images, you can render LaTeX and Typst math fragments, view PDFs in doc-view, or even experiment with an inline web browser. It turns a standard terminal interface into a rich, visual environment, making it much easier to read documentation or browse files with visual previews right in your editor.

📰 https://news.ycombinator.com/item?id=47332741

🆔 @hackernewsgithubprojects

📦 mkubicek/xtap

xTap: Silent Tweet Archiving for Power Users

xTap is the browser extension that quietly archives your X timeline into structured daily files without any manual scraping. Instead of making extra network requests, it transparently intercepts the live GraphQL API responses your browser already receives while you surf. It processes these payloads into clean JSONL objects, capturing full text, media, and article content with zero footprint. By offloading data writing to a local background daemon, it operates entirely in the background, ensuring your browsing habits remain private and efficient. This tool provides a seamless way to maintain a searchable, permanent archive of everything you encounter online.

🆔 @hackernewsgithubprojects

📦 alpacalabsllc/skills-for-architects

Automate Architecture Workflows with Claude Code

You see a terminal command effortlessly pulling zoning data or generating a detailed workplace space program in seconds, turning complex architectural tasks into simple slash commands. This repository provides a comprehensive collection of specialized skills designed for architects, real estate professionals, and workplace strategists working within Claude Code. It solves the problem of manual, time-consuming data collection and analysis by offering automated tools for site planning, zoning analysis, sustainability reporting, and material specifications. These skills enable you to accelerate your design process and focus on high-level decision-making. Try it out to streamline your next project and work faster.

🆔 @hackernewsgithubprojects

📦 endo-yusuke/humation

Humation: Deterministic Kawaii Avatars Without AI

Humation is a lightweight, deterministic avatar engine that generates hand-drawn, kawaii-style illustrations for applications without relying on artificial intelligence or external API calls. By using a simple seed string as input, the system produces consistent visual results, ensuring the same seed always yields the exact same avatar. The project provides flexible packages for React and web components, allowing developers to manually select specific character parts or colors via CSS variables. It is an efficient, privacy-focused solution for developers who want to integrate custom, scalable, and stable character illustrations into their software while maintaining full control over the visual assets.

🆔 @hackernewsgithubprojects

📦 xuefanfu/3dg-vln

Autonomous UAV Navigation: The 3DG-VLN Framework

Unmanned aerial vehicles can now perform precise target-oriented navigation using high-resolution visual data and dynamic spatial updates. This framework enables drones to navigate toward specific targets by processing both front-view and downward-facing camera feeds to maintain accurate orientation. By integrating coarse 3D directional cues and continuously updating target positions during flight, the system effectively minimizes spatial drift in complex environments. Researchers can use this implementation to experiment with advanced navigation benchmarks, as it includes the necessary code, pretrained models, and evaluation tools to support robust research in aerial vision-language navigation for autonomous drone operations.

🆔 @hackernewsgithubprojects

📦 jhdev2006/super-mario-bros.-remastered-public

Super Mario Bros. Remastered Overview

Mario traverses a meticulously recreated level stage while responding to smooth, modern gameplay physics. This project serves as a full ground-up remake of classic titles like the original Super Mario Bros. and The Lost Levels, built using the Godot engine. It offers fans an expanded experience featuring custom characters, support for unique resource packs to modify visuals and audio, and a built-in level editor for creating personal challenges. By providing an open-source framework, it invites players to customize and build upon the iconic platforming foundation. Dive into this modernized take on a timeless classic and start building today.

🆔 @hackernewsgithubprojects

📦 bradautomates/claude-video

Give Claude the Power to Watch Any Video

Claude processes a video by extracting key frames and generating a timestamped transcript, allowing you to ask questions about the visual content directly. This repository provides a skill that bridges the gap between static text and video by using tools like ffmpeg and yt-dlp to download or access local files, capture specific frames based on video length, and leverage Whisper for audio transcription. It solves the problem of needing to watch long videos yourself to find specific details, making it easy to analyze hooks, diagnose visual bugs, or summarize content quickly by interacting with both audio and images.

🆔 @hackernewsgithubprojects

📦 rhyssullivan/executor

The Missing Integration Layer for AI Agents

Empower your AI agents to interact with any external service by centralizing your tools into one secure, searchable catalog. This project functions as an integration layer that allows you to connect OpenAPI, GraphQL, and MCP tools, making them instantly available across all your favorite agents. Whether you are using a command line interface or connecting via a web dashboard, you can define sources, manage authentication, and call functions with full type safety. By providing a unified runtime for tool discovery and execution, it bridges the gap between your agents and the real-world data they need. Start building smarter workflows today.

🆔 @hackernewsgithubprojects