📦 ethanfel/comfyui-krea2textencoder

Better Krea2 Prompts with Vision-Aware Encoding

ComfyUI-Krea2TextEncoder is the custom node that finally provides proper vision-aware conditioning for the Krea2 model. It solves issues with the default encoder by forcing the correct Krea2 descriptor template and removing unnecessary VAE inputs that the model ignores. This tool allows you to feed multiple reference images and optional masks into the Qwen3-VL-4B vision path to influence your generations effectively. It supports auto-growing input slots and configurable padding for precise image cropping, ensuring your prompts interact correctly with reference visuals. Upgrade your ComfyUI workflow to achieve more consistent and visually responsive output for your creative projects.

🆔 @hackernewsgithubprojects

📦 jordancoin/codemap

Codemap

Give your AI agents instant project context without wasting valuable API tokens. Written in Go, codemap runs locally to provide structured architectural overviews, dependency flow mapping, and quick code diffs. It features a skills framework that delivers context-aware guidelines to Claude or other LLMs based on the languages and files you are editing. You can use its built-in model context protocol server, serve context over a local HTTP API, or capture multi-agent handoffs when switching from Claude to Cursor. Download codemap today to make your AI coding sessions faster, cheaper, and vastly more accurate.

🆔 @hackernewsgithubprojects

📦 browser-use/browser-agent-template

Build Your Own Web-Browsing AI Agent

This open-source template lets you deploy a functional web-browsing agent that navigates the real internet using a cloud-based browser. By integrating the eve framework with a live browser environment, it enables you to create agents that scrape data, fill out forms, and perform tasks on actual websites. The architecture includes a side panel that displays the agent's work in real time, giving you full transparency into its interactions. Because the agent manages its browser session server-side, you can build reliable tools for automation without handling complex scraping APIs. It is a powerful foundation for building your own intelligent browsing assistants.

🆔 @hackernewsgithubprojects

📦 ynulihao/agentskillos

AgentSkillOS

Orchestrate complex AI agent workflows automatically by pulling from a massive library of over two hundred thousand specialized skills. AgentSkillOS works like an operating system for agents, letting you search a structured capability tree to discover non-obvious, functionally relevant skills and compose them into directed acyclic graph pipelines. It solves the massive scalability bottleneck of finding and combining the right tools for multi-step tasks, featuring a clean web UI for human-in-the-loop intervention and a batch CLI with rich progress terminal logs. Give AgentSkillOS a spin to build highly capable, auditable agent pipelines that execute flawlessly.

🆔 @hackernewsgithubprojects

📦 fareedkhan-dev/claude-code-from-scratch

claude-code-from-scratch

claude-code-from-scratch is the terminal agent harness that replicates the exact multi-agent architecture powering Anthropic's production tools from the ground up [README.md]. Instead of relying on fragile prompts, it introduces developers to harness engineering by progressively building a twenty-three-session agent framework starting with a minimal perception-action loop [README.md]. You will learn how to implement essential runtime features like three-layer context compression, YAML rule-based permissions, parallel tool execution, and official Model Context Protocol integrations [README.md]. It is a masterclass in building reliable agentic software that remains highly stable over long sessions [README.md]. Grab the code, boot up the scripts, and start building production-grade AI agents today [README.md].

🆔 @hackernewsgithubprojects

📦 noblerworks-hq/ironsight

Ironsight

Monitor real-time open-source intelligence across the Middle East by spinning up a localized command center dashboard on your machine. This Next.js and TypeScript project aggregates critical security data from over fifty free sources without requiring a single API key. You can display live military aircraft tracking via adsb.lol, naval vessel telemetry in the Persian Gulf, real-time translated Telegram channels, and NASA FIRMS satellite thermal markers for fire detection. It even loops missile trajectory arcs on an interactive Leaflet map during active alerts. It is a brilliant example of client-side data aggregation that puts a complete tactical operating picture right in your browser.

🆔 @hackernewsgithubprojects

📦 snu-vgilab/extend3d

Extend3D

Generate massive town-scale 3D scenes from a single input image using your local NVIDIA graphics card. The extend3d project solves the massive scale bottleneck of traditional 3D generation by expanding neural reconstruction pipelines to model entire virtual towns. Built on top of the Trellis framework, this codebase processes a flat source image to output both highly detailed 3D Gaussians and optimized FlexiCubes meshes ready for virtual worlds. You can run the entire pipeline in PyTorch to export complex environment assets as standard GLB files with high-fidelity textures. Check out this project to start building expansive, render-ready environments from simple images.

🆔 @hackernewsgithubprojects

📦 cathrynlavery/diagram-design

Diagram Design

Diagram-design is the Claude Code skill that finally generates gorgeous, editorial-quality diagrams instead of generic, rounded-box artificial intelligence slop. When you ask Claude to map out your architecture or database, it skips the ugly shadows and instead crafts beautiful, self-contained HTML and SVG schematics that match your brand in sixty seconds. Just point the skill at your website, and it automatically extracts your exact colors and typography, checking contrast levels to map out pristine timelines, flowcharts, or state machines. It relies on a meticulous, grid-aligned system with gorgeous typography and hand-drawn sketchy variants that look like a professional designer built them. You will get stunning, production-ready visuals tailored specifically to...

🆔 @hackernewsgithubprojects

📦 himself65/finance-skills

finance-skills

Supercharge your AI agents with the ability to perform institutional-grade investment research. The finance-skills repository provides a structured open standard framework to load advanced financial skills directly into Claude Code. Your AI can instantly pull live market data through yfinance, map complex options payoff charts, and run discounted cash flow valuations. It even reads TradingView desktop app states and tracks Strait of Hormuz shipping risks. It is the ultimate toolkit for developers who want their AI assistants to analyze the markets like seasoned hedge fund analysts. Grab the repository and install the plugins to start running automated, multi-perspective financial analyses today.

🆔 @hackernewsgithubprojects

📦 jackywxsz/jacky-motion

Jacky Motion

Your AI-generated video scripts are probably dead on arrival because they lack visual structure. jacky-motion is a specialized agent skill designed to audit your raw Chinese voiceover scripts and instantly transform them into perfectly timed, presentation-ready sixteen-by-nine HTML motion graphics. It operates on a rigorous six-stage quality check that makes sure every single beat of your spoken script maps directly to a physical, on-screen visual action. Instead of generic templates, it serves up highly tailored CSS styles like polished glass-panel Apple Keynote mockups and illuminated financial studio cards. It is a brilliant way to build professional, video-ready information screens without touching a video editor.

🆔 @hackernewsgithubprojects

📦 resciencelab/opc-skills

OPC Skills

opc-skills is the automation repository that equips your AI agents with specialized marketing, research, and branding capabilities. Instead of relying on generic prompts, it delivers self-contained modular folders complete with instructions and scripts that integrate directly into tools like Claude Code, Cursor, and Windsurf. You get immediate access to tactical utilities, such as checking domain availability and pricing across registrars, analyzing user demand via the Rust-based requesthunt CLI, generating vector logos, and applying Generative Engine Optimization to boost visibility in AI search results. Install these utilities instantly to supercharge your terminal workflow and let your AI agents run your solopreneur business for you.

🆔 @hackernewsgithubprojects

📦 refloow/refloow-photo-studio

Refloow Photo Studio

You can now edit photos on a local desktop app with absolutely zero ads, watermarks, accounts, or background data collection. Refloow-photo-studio is a free, privacy-first image editor built with Electron that processes everything directly on your computer. It features built-in background removal using a local machine learning model, multi-photo layering, custom text overlays, and over forty professional filters. The application supports simple drag-and-drop actions, crop adjustments, and an active undo history of up to fifteen steps. It is a highly practical, secure, and lightweight tool for quick, everyday desktop photo editing.

📰 https://news.ycombinator.com/item?id=48619100

🆔 @hackernewsgithubprojects

📦 traderalice/auto-quant

Auto-Quant

The LLM-native autonomous quant research loop that applies Karpathy's autoresearch pattern to FreqTrade crypto strategies. Auto-Quant lets an AI agent independently own, modify, and evolve files inside a strategies directory to optimize trading logic across a five-pair crypto portfolio. By executing an iterative loop guided by instructions in a markdown file, the agent runs in-process batch backtests over multi-year regimes, evaluates multi-timeframe indicators, and applies dynamic position sizing. It logs every action to a gitignored event list to preserve historical lessons through hard resets, while multi-objective gates prevent the agent from gaming metrics. See how autonomous agents can systematically refine quantitative strategies without human intervention.

🆔 @hackernewsgithubprojects

📦 tradecatlabs/human_infra

human_infra

The human_infra repository operates as a docs-as-code knowledge base designed to map out human runtime infrastructure, longevity evidence, and neuroethical boundaries. Instead of treating longevity or cognitive engineering as vague science fiction, this project treats the human body and mind as systems architecture, organizing research on everything from disembodied central nervous system interfaces and memory-editing limits to time-dilation thought experiments like the black-hole waiting room. Built using an Astro and D3 research frontend, human_infra introduces a quantitative life-path prediction model contract to calculate survival curves and relative time. It is a brilliant, highly structured approach to human optimization that reads like a system administrator's manual for humanity.

🆔 @hackernewsgithubprojects

📦 px4/hawkeye

Hawkeye Flight Visualizer

Hawkeye is a real-time 3D flight visualizer for PX4 that lets you watch live simulations, replay flight logs, and analyze multi-drone swarms without installing any heavy dependencies. It renders up to sixteen vehicles simultaneously using Raylib and MAVLink, displaying pitch, roll, and altitude with zero lag. You can compare different flights by overlaying them as translucent ghost models at thirty-five percent opacity, or use the automatic takeoff alignment to synchronize separate runs to the exact moment of liftoff. It is an incredibly fast, lightweight utility for developers to visually inspect their flights frame-by-frame and pinpoint issues instantly.

🆔 @hackernewsgithubprojects

📦 ai4s-research/awesome-ai-for-science

Awesome AI for Science

Skip the endless search for research tools and grab a massive, curated cheat sheet for scientific discovery. The awesome-ai-for-science repository is a crowd-sourced goldmine of open-source AI models, datasets, and frameworks designed to crack complex problems in physics, chemistry, and biology. Instead of hunting through obscure academic forums, you get a clean, vetted directory of game-changers like AlphaFold alongside specialized datasets for materials science. It is basically a shortcut to the best machine learning tools currently reshaping how we understand the physical world. Check it out, find your next research companion, and start building something brilliant.

🆔 @hackernewsgithubprojects

📦 valuecell-ai/valuecell

valuecell

valuecell is the multi-agent financial platform that puts a team of virtual investment analysts right on your desktop. It connects smart AI agents directly to your stock and crypto accounts to help you research, track, and automate your trading. Instead of sending your sensitive API keys to a random cloud, valuecell runs locally on your device so your data stays yours. You get specialized agents for deep research, news analysis, and strategy execution, all talking to top language models like DeepSeek. It is like having a private hedge fund team running on your machine.

🆔 @hackernewsgithubprojects

📦 star23/day1global-skills

Day1Global Skills

Supercharge your AI agents with professional-grade investing capabilities using day1global-skills. This repository lets you instantly equip your AI with specialized analytical frameworks for evaluating tech earnings, value investing, market sentiment, global liquidity, and Bitcoin cycles. Instead of asking a generic chatbot for basic summaries, you can install these skills directly into tools like Claude Code to run automated, institutional-grade deep dives. For example, the Bitcoin cycle model tracks thirteen weighted on-chain and sentiment indicators to score market heat from zero to one hundred, giving clear accumulation or distribution signals. It turns your daily AI assistant into a highly specialized financial analyst with just a single command.

🆔 @hackernewsgithubprojects

📦 opengeos/geoagent

geoagent

geoagent is the unified AI assistant layer that lets you control your geospatial Python packages and map widgets using plain English. Instead of writing custom AI code for every different mapping library, this clever project centralizes everything into one smart, shared interface. It connects directly to live map widgets like leafmap, connects search tools to massive space agency catalogs, and even docks right into desktop QGIS to let you manipulate map layers, run complex algorithms, or write fallback scripts on the fly. It is the ultimate bridge between natural language and complex geographic data. Check it out and bring your maps to life with simple prompts.

🆔 @hackernewsgithubprojects

📦 sagar-tamang/friday-tony-stark-demo

friday-tony-stark-demo

You can now run a voice-controlled digital assistant on your own machine that acts just like Iron Man's AI, Friday. The friday-tony-stark-demo project splits the assistant into two clever parts. First, a backend server exposes real-time tools like system diagnostics and web searches. Second, a live voice pipeline listens to your microphone, reasons using Gemini, and talks back to you instantly. Instead of lagging, it uses a protocol to fetch live information mid-conversation. Setting it up is surprisingly simple with the right API keys. If you want a hands-free assistant that actually feels responsive and alive, this is a fantastic weekend project.

🆔 @hackernewsgithubprojects