📦 alibaba-damo-academy/worldolympiad

Is Your AI Model Actually Smart? Meet WorldOlympiad

Think your video-generation AI is a genius? Most models just fake it with pretty pixels, but WorldOlympiad puts them to the ultimate test. It is a new benchmark that treats AI like a top-tier athlete, grading models on three tough tracks: physical rules, geometric consistency, and interaction fidelity. Instead of just checking if a video looks good, it dives deep into mechanics, thermal properties, and 3D structure to see if the world model truly understands reality. By stress-testing these systems with gaming and robotics, it exposes where your AI is actually struggling. Ready to see if your model survives?

🆔 @hackernewsgithubprojects

📦 wdrink/arm

ARM: A Revolutionary Unified Multimodal AI Model

Have you ever wondered if one single AI model could master image understanding, generation, and editing all at once? Meet ARM, a cutting-edge autoregressive model that treats visual data like language. By mapping images into compact discrete tokens, ARM uses next-token prediction to unify these diverse tasks into one efficient framework. It significantly boosts performance through preference optimization, proving that a single model can handle complex visual reasoning and high-quality generation simultaneously. This approach simplifies multimodal AI, showing that autoregressive models can effectively dictate composition and semantics. It is an exciting step forward for unified artificial intelligence research.

🆔 @hackernewsgithubprojects

📦 zlab-princeton/i1

i1: A Fully Open Recipe for Strong Text-to-Image AI

Ever wonder how powerful text-to-image AI is actually built? Most top-tier models keep their training data and recipes secret, making it nearly impossible for the research community to learn from them. The i1 project changes that by offering a fully open, high-performance 3B-parameter diffusion model. By running over three hundred controlled experiments, the team identified the best design choices for both architecture and data. The result is a competitive model that you can fully explore, reproduce, and build upon. Check out their codebase to see how these simple, effective recipes can advance the future of open-source image generation.

🆔 @hackernewsgithubprojects

📦 orchestra-research/ai-research-skills

Supercharge Your AI Agents with This Massive Research Skills Library

Stop wasting time debugging infrastructure and let your AI take the lead on complex research. This repository provides a massive collection of nearly one hundred expert-level skills designed to turn AI models into fully autonomous research agents. Whether you need to handle distributed training with PyTorch, optimize models with vLLM, or write academic papers using provided LaTeX templates, this library covers the entire research lifecycle. By packaging these capabilities, you enable your agent to manage everything from initial ideation and data processing to final evaluation. Unlock the true potential of your AI agents and accelerate your scientific discovery today.

🆔 @hackernewsgithubprojects

📦 s0xdk/ghostty-blackhole

Turn Your Terminal Into A Black Hole Pomodoro Timer

Want to add a touch of cosmic drama to your coding workflow? This project brings a stunning, interactive black hole directly into your Ghostty terminal, acting as a clever visual Pomodoro timer. As you work, the hole slowly grows, warping your terminal text with realistic gravitational lensing, event horizons, and accretion disks. When the hour peaks, it signals that it is time for a break. It is entirely self-contained, using clever shader math to track your focus based on the wall clock and your typing activity. Transform your boring terminal into a dynamic, engaging workspace today.

🆔 @hackernewsgithubprojects

📦 serenakeyitan/awesome-agent-loops

Stop Prompting, Start Looping: Automate Your Coding Agent

Are you still manually prompting your coding agent one step at a time? It is time to level up by using agent loops to get more done with less effort. This repository is a curated collection of powerful, real-world commands for Claude Code and Codex that help you automate repetitive tasks. Whether you need to fix flaky tests until they pass, babysit pull requests, or sync documentation automatically, these commands put your work on autopilot. Simply copy and paste these proven patterns to create your own effective loops. Start building smarter workflows today and let your agent handle the grind.

🆔 @hackernewsgithubprojects

📦 nelson820125/jordium-gantt-vue3

Build Powerful Gantt Charts with Vue 3

Need to manage complex schedules and resource planning in your web app? Check out this powerful Vue 3 Gantt chart library. It simplifies task management by providing a flexible, interactive interface for scheduling tasks and monitoring resource allocation. With built-in features like drag-and-drop, customizable task bars, and even resource conflict detection, it helps you visualize timelines effortlessly. Whether you are building project management tools or resource trackers, this component is designed to handle it all with ease. Dive into this repository to streamline your project scheduling workflow today and bring professional-grade Gantt charts directly into your application.

📰 https://news.ycombinator.com/item?id=47437651

🆔 @hackernewsgithubprojects

📦 anthropics/claude-agent-sdk-python

Build Powerful AI Agents with Claude Agent SDK

Unlock the power of building autonomous AI agents directly within your Python projects using the Claude Agent SDK. This powerful tool allows you to seamlessly integrate Claude Code into your applications, offering features like bidirectional conversations and highly customizable agent behavior. You can define custom tools as in-process servers for better performance and type safety, or use hooks to precisely control agent actions. Whether you are automating complex workflows or creating intelligent assistants, this SDK provides the robust infrastructure needed to get started quickly. Dive into this repository to start building your own advanced AI-driven tools today.

🆔 @hackernewsgithubprojects

📦 larksuite/cli

Master Lark with the Ultimate Official CLI

Unlock the full potential of your workspace with the official Lark CLI tool designed for both humans and AI agents. This open source powerhouse covers eighteen business domains, offering over two hundred curated commands to manage your calendar, messenger, docs, sheets, and tasks seamlessly. It is specifically engineered to be agent-native, featuring twenty-six built-in skills that allow AI to operate your workspace with zero extra setup. Whether you need to automate meeting schedules or manage complex data, this tool provides a secure, three-layer architecture to streamline your workflow. Install it today to bring professional-grade automation directly to your terminal.

🆔 @hackernewsgithubprojects

📦 fluxerapp/fluxer

Build Your Own Messaging Platform with Fluxer

Tired of locked-down messaging apps? Discover Fluxer, a free and open source platform built for friends and communities that puts you back in control. It offers a complete suite for real-time chat, including threaded replies, custom emojis, and voice and video calling powered by LiveKit. The best part is its focus on self-hosting, allowing you to run your own instance and own your data entirely without vendor lock-in. Whether you want to customize your server or just enjoy a private community space, Fluxer provides the tools to get started today. Take control of your digital home and start building.

🆔 @hackernewsgithubprojects

📦 always-further/nono

Stop AI Agents from Taking Over Your Machine

Are you worried about AI agents running wild on your computer with full access to your sensitive files and credentials? Meet nono, a powerful, kernel-level sandboxing system built specifically for AI agents. Instead of trusting agents with your entire machine, nono uses OS-level primitives to restrict them to only the specific paths, network access, and environment variables they actually need. It works by enforcing these policies at the kernel level, making security rules irreversible and impossible to bypass. Secure your workflow, protect your private data, and keep your agents in check by defining explicit, audit-ready capabilities today.

🆔 @hackernewsgithubprojects

📦 npc-worldwide/npcsh

Build AI Teams Right From Your Terminal with npcsh

Want to supercharge your workflow with AI? Meet the agentic shell that lets you build and run entire teams of AI agents directly from your command line. Instead of juggling separate tools, you can use simple slash commands to schedule tasks, design custom execution templates, and even host multi-agent discussions to solve complex problems. It integrates seamlessly with major LLM providers and supports flexible agent definitions, allowing you to easily orchestrate specialized AI personas. Whether you are automating workflows or diving into deep research, this project gives you full control. Start building your own autonomous agent team today.

🆔 @hackernewsgithubprojects

📦 agentlas-ai/hephaestus

Build AI Agents from One Sentence with Hephaestus

Ever wanted to turn a simple idea into a functional AI agent or a full multi-agent team? Meet Hephaestus, the no-code engine that does exactly that. Instead of manually coding complex workflows, you just provide a sentence describing your goal, and Hephaestus generates a complete, installable repository. It works as an alternative to existing frameworks like CrewAI or LangGraph and integrates seamlessly with tools like Claude Code, Codex, and Cursor. Whether you need a single specialist or a coordinated team, Hephaestus builds the architecture for you. Stop building from scratch and start generating production-ready agent structures instantly.

🆔 @hackernewsgithubprojects

📦 jianhuiwei7/videoweaver

VideoWeaver: Evolving AI Agents for Long Video Generation

Ever wonder if AI agents can handle complex, long-form video production? Meet VideoWeaver, a research project designed to test and evolve how agent frameworks perform in long-horizon video generation. Instead of rigid pipelines, it provides a library of reusable foundation skills—like audio generation, video trimming, and metadata analysis—that allow agents to create and refine their own custom workflows. By evaluating execution traces and output quality, the system progressively optimizes these skills through feedback. It is a fascinating way to study how agents move from simple tasks to complex, multi-step creative orchestration. Check it out to see the future of automated video production.

🆔 @hackernewsgithubprojects

📦 sustech-nlp/awesome-text-world-models

Mastering Text World Models for AI Agents

Ever wonder how AI agents learn to navigate complex digital worlds? This repository is a goldmine for anyone researching Text World Models, which are transition models that predict outcomes for agents by analyzing textual states like terminal outputs or API responses. It solves the agent-world gap by providing a massive, curated collection of papers organized by how these models are built, trained, and used for inference or evaluation. Whether you are into programmatic environment synthesis or deep tree search planning, this resource maps out the entire lifecycle. Dive into these curated insights to level up your agent research today!

🆔 @hackernewsgithubprojects

📦 louieworth/trd

Stop Prefix Failure! Meet Trajectory-Refined Distillation

Are your language models struggling with fragmented training? Prefix failure is a common hidden hurdle in on-policy distillation where standard token-level fixes just do not cut it. Trajectory-Refined Distillation changes the game by correcting entire student rollouts under teacher guidance before training even begins. This approach mitigates issues at the source, improves exploration, and leads to more stable, accurate reasoning. This repository provides the complete implementation, including training recipes and evaluation tools to help you reproduce these results on your own models. Dive into this method to boost your model's performance and see smarter, more consistent reasoning today.

🆔 @hackernewsgithubprojects

📦 yizheng-sun/distract-bench

Is Your AI Easily Distracted? Meet Distract-Bench

Are multimodal AI models truly reliable or just easily distracted by visual noise? Distract-Bench is a specialized benchmark designed to test if vision-language models can stay focused on core reasoning tasks when faced with intentional semantic distractions. This repository provides critical metadata, pre-computed model outputs, and custom metrics like the Harmful Reference Rate to measure how often a model gets led astray. By tracking whether a model notices and follows these visual traps, researchers can better understand model robustness. Explore these tools to identify weaknesses in your AI models and build smarter, more resilient reasoning systems today.

🆔 @hackernewsgithubprojects

📦 rindig/interpreted-context-methdology

Stop Using Complex AI Frameworks: Use Folder Structures Instead!

Are you tired of managing complex AI agent frameworks that feel like a black box? Meet the Interpreted Context Methodology, a brilliant approach that replaces code-heavy orchestration with a simple, intuitive filesystem structure. By organizing your workflow into numbered folders, each representing a clear stage, you allow an AI agent to read exactly the right prompts and context at the right time. This makes every step observable, editable, and incredibly easy to debug since your entire project state is just plain text files on your computer. Stop fighting abstractions and start building transparent, human-in-the-loop pipelines where you remain in total control.

🆔 @hackernewsgithubprojects

📦 zubair-trabzada/ai-marketing-claude

Automate Your Marketing Agency with Claude Code

Want to supercharge your marketing workflows using artificial intelligence directly from your terminal? This repository is a comprehensive marketing suite built specifically for Claude Code that lets you instantly audit websites, generate copy, build email sequences, and create content calendars. By leveraging fifteen specialized skills and five parallel subagents, it tackles everything from competitive intelligence to professional PDF reporting. Whether you are an agency builder or a solopreneur, this tool automates complex research and content creation to help you scale your services. Start using these AI-driven marketing commands today to turn manual analysis into actionable insights.

🆔 @hackernewsgithubprojects

📦 hesamsheikh/awesome-openclaw-usecases

Stop Guessing: Automate Your Life with OpenClaw Use Cases

Ever feel like AI tools are powerful but you just don't know how to use them to actually save time? This project changes the game by collecting real-world, community-tested ways to integrate OpenClaw into your daily routine. Instead of staring at a blank prompt, you can use these recipes to automate everything from YouTube content pipelines and smart home maintenance to personal habit tracking and even hands-free video editing. It solves the biggest hurdle of modern tech by turning abstract capabilities into practical, life-improving workflows. Dive in to start building your own autonomous assistant and finally reclaim your schedule.

🆔 @hackernewsgithubprojects