📦 haddenhunter/skillforge

SkillForge: The npm for AI Agents

SkillForge turns AI coding agent capabilities into installable packages, finally giving your tools a proper home like npm. Instead of just pasting random prompt snippets, you can publish, version, and securely share specific skills across different AI models like Claude Code or Cursor. The standout feature is its strict sandboxing, which locks each skill down so it can only access the exact files and network hosts you allow, preventing your AI from accidentally reading sensitive data or causing chaos. It works locally first, so your code never leaves your machine, and it even lets you evaluate skills against baselines to ensure they perform consistently before you deploy them.

🆔 @hackernewsgithubprojects

📦 chattymin/poketokenbar

Your AI Tokens Hatch Pokémon

PokeTokenBar turns the AI tokens you burn daily into a growing Pokémon companion living right in your macOS menu bar. Every time you code with tools like Claude Code or Codex, the app reads your local usage data and incubates an egg that eventually hatches into a real creature. As you continue using these tools, your Pokémon evolves through its authentic tree, with rare shiny variants appearing only in the most unusual cases. This simple app makes tracking your expensive API limits feel like a playful collection game rather than a stressful dashboard, turning your daily coding habits into a rewarding digital pet experience you will actually enjoy checking.

📰 https://news.ycombinator.com/item?id=48936171

🆔 @hackernewsgithubprojects

📦 hassancs91/brainoutside

Turn Your Notes Into an AI Brain

Give your AI agents a real, readable brain that lives outside their heads. BrainOutside turns a simple git repo of markdown into a self-hosted memory server, letting any agent connect via standard protocols to read your thoughts, facts, and opinions. The coolest part is that it isn't just a database; it is a creative tool. Agents can use your unique identity files to write content in your actual voice, not just generic summaries. Because it is built on plain files, you can always read it, edit it, or delete the server without losing your data.

🆔 @hackernewsgithubprojects

📦 slidespeak/slide-design-prompts

Slide Design Prompts: AI Presentation Magic

Slide Design Prompts is the library that fixes ugly AI slides with professional design systems. AI tools write decent text but often ruin the look with clashing colors and random clip art. This free collection offers eighty-one ready-made themes, from corporate boardrooms to retro arcade styles, that you simply paste into ChatGPT or Claude. Each prompt locks in specific colors and fonts to ensure your slides look polished and consistent. You can even swap in your own brand colors while keeping the layout perfect. This is the fastest way to make your presentations look like a designer built them.

🆔 @hackernewsgithubprojects

📦 seakee/cpa-manager-plus

CPA Manager Plus: The Missing Dashboard for AI Costs

CPA Manager Plus stops letting AI billings become a mystery by giving you a full local dashboard for your AI traffic. It tracks exactly where your money and tokens go, breaks down costs by model and account, and flags failing requests before they eat your budget. Everything stays on your machine, no cloud, no telemetry. If you run AI gateways, this is the missing control panel you didn’t know you needed until now.

🆔 @hackernewsgithubprojects

📦 test-time-training/discover

How AI Beats Humans by Learning on the Job

Discover lets a large language model keep training while it tackles a specific problem, rather than relying solely on what it learned from past data. By generating code, running it, and updating itself based on the results, it finds solutions that surpass human experts and previous artificial intelligence systems. This tool is particularly useful for complex engineering and scientific tasks where the best answer is unknown, allowing the system to iterate and improve its own logic in real time. The result is an AI that can genuinely innovate on the fly, turning a static model into a dynamic problem solver that outperforms the best known methods.

🆔 @hackernewsgithubprojects

📦 estebanpdn/zelda-tmc-3ds

Playing Minish Cap Natively on a Nintendo 3DS

A classic Game Boy Advance adventure now runs natively on the Nintendo 3DS. This project ports The Minish Cap directly to the device, using its dual screens to show gameplay on top and a live map below. It is a full rewrite based on open-source work, not an emulator. The surprising part is that it fits the original game into the hardware while adding new options like widescreen and turbo speed. It solves the problem of playing this specific Zelda title on a modern handheld without relying on emulation. The result is a smooth, native experience that feels at home on the system.

🆔 @hackernewsgithubprojects

📦 nvlabs/simfoundry

SimFoundry: Video to Robot Simulation

SimFoundry turns a short real-world video into a physics-ready simulation scene in under an hour, with no manual annotation required. Point it at a tabletop, and it automatically segments every object, reconstructs geometry, generates textured 3D meshes, and compiles the result into an OmniGibson scene complete with physical parameters, digital cousin variations, and task proposals. Unlike prior scene reconstruction approaches, SimFoundry is fully modular: each stage is an independently swappable component. As foundation models improve, the SimFoundry pipeline improves with them. This tool solves the bottleneck of creating training environments for robot policies by automating the entire pipeline from video to simulated reality.

🆔 @hackernewsgithubprojects

📦 amap-cvlab/abot-world

ABot-World: Infinite Gaming on One GPU

ABot-World is the desktop project that turns your computer into an infinite, playable video world. Instead of just generating short clips, it creates a continuous environment where you can move, act, and explore in real-time. The coolest part is that it runs entirely on a single graphics card, keeping the world alive without ever cutting to a new video. You can wander through scenes that keep expanding around you, reacting to your inputs instantly. This project proves that we don't need massive cloud servers to build interactive digital spaces anymore. It’s a major step toward personal, local world-building, and it’s ready for you to explore today.

🆔 @hackernewsgithubprojects

📦 ussoewwin/comfyui-qwenimageloraloader

The Missing Link for AI Style Control

The ComfyUI Qwen Image LoRA loader is the custom node that finally makes style customization work without breaking your workflow. It acts as a smart bridge, taking your specialized training files and slapping them onto high-performance image models so you can generate exactly the look you want. The coolest part is the smart control system that automatically figures out if your file is for depth or pose guidance, removing the need for messy manual setup. This solves the headache of mixing different AI components that usually refuse to talk to each other.

🆔 @hackernewsgithubprojects

📦 slidespeak/presentation-design-prompts

Design Your Own AI Slides for Free

Make your AI-generated slides actually look professional. Slide Design Prompts is a collection of eighty-one ready-made design systems you can paste directly into ChatGPT or Claude. Usually, AI handles text well but fails at visuals, giving you clashing colors and messy layouts. This project fixes that by providing specific instructions for colors, fonts, and slide structures. You pick a style that fits your goal, from a corporate boardroom look to a playful marketing deck, and the AI follows it perfectly. It turns a simple text prompt into a polished, visually consistent presentation without any design skills required.

🆔 @hackernewsgithubprojects

📦 tritant/comfyui_minimax_h3_extender

ComfyUI MiniMax H3 Extender: Continuous Video Chains

The ComfyUI MiniMax H3 Extender solves the biggest headache in long-form video generation: the jarring jump cuts between clips. It chains multiple generations together so motion, characters, and even the audio track flow continuously, feeling like one single, unbroken take rather than a slideshow. The coolest part is its validation system, which lets you build a scene one clip at a time, locking in the good takes so you don’t have to start over if one shot misses the mark. It’s like editing a film where the software remembers exactly where you left off.

🆔 @hackernewsgithubprojects

📦 sapientorius/nebulynk

Nebulynk: The Self-Hosted Team Hub with AI Follow-Up

Your entire team's video calls can be automatically transcribed and summarized by AI that runs on hardware you control. Nebulynk is a self-hosted platform that combines chat, voice, and video meetings into one workspace. The standout feature is its AI-assisted follow-up, which turns spoken conversations into searchable notes and next steps. Instead of sending your data to a third-party cloud, you use your own AI provider and keep full control over your infrastructure. This is ideal for teams that want modern collaboration without sacrificing privacy or paying per-user fees.

🆔 @hackernewsgithubprojects

📦 cursor/cookbook

Cursor Cookbook: The Ultimate AI Coding Playground

The Cursor Cookbook is the developer's playground that turns abstract AI concepts into working reality. It stops the endless guessing game about how to integrate coding agents by showing you exactly how to build, run, and manage them from scratch. Think of it as a cheat code for the Cursor ecosystem. The repo is packed with ready-to-use examples, from simple command-line tools to complex, self-hosted cloud setups. But the real gem is the DAG task runner. Imagine you have a massive, tangled coding project.

🆔 @hackernewsgithubprojects

📦 arcee-ai/nac

How Nac Keeps AI Agents On Task

Nac prevents AI agents from losing focus during long coding tasks by splitting work into separate, manageable streams. Instead of letting a single agent ramble through hours of work, it uses a central coordinator to plan steps and launch focused workers. Each worker handles one specific part of the job and reports back a clear summary, ensuring the project never drifts from your original goal. It is a simple, open-source way to keep ambitious AI projects on track and aligned with your intent, making it a reliable tool for developers who need consistent results without constant oversight.

🆔 @hackernewsgithubprojects

📦 xaspx/hermes-control-interface

Run Your AI Agents from a Web Page

Hermes Control Interface is the self-hosted web dashboard that finally gives you full command over your AI agent stack from a single browser tab. If you are running Hermes agents on your own server, this project solves the chaos of scattered terminals and hidden logs by bringing everything into one secure, password-protected hub. It connects directly to your server to provide a live terminal, a file explorer, and real-time chat, all without needing a separate desktop app. The standout feature is the Office view, which acts as a live control tower for multiple agents at once, showing their health and active tasks on a shared board.

🆔 @hackernewsgithubprojects

📦 keytyapp/keyty

Keyty Visualizes Your Screen Input

Ever wonder how tech YouTubers make their typing look so crisp? Keyty is a free, open-source tool for macOS that does exactly that. It overlays your keyboard and mouse actions right on your screen in real time. Think of it as a spotlight for your input, showing every keypress and click with a clean, customizable style. This makes tutorials, live streams, and presentations much easier to follow. You can even tweak the visuals to match your brand, and it keeps your privacy safe by processing everything locally. It’s a simple, elegant way to make your screen recordings actually make sense.

🆔 @hackernewsgithubprojects

📦 danellalc/efcore.autoseed

EFCore.AutoSeed: Seed Your Database Automatically

EFCore.AutoSeed is the database seeder that finally stops you from rewriting your schema. Every other tool makes you manually define factories or CSV files, but this one simply reads your EF Core model to generate realistic data. It automatically figures out the insertion order, resolves complex cycles, and creates long-tail distributions that mimic real production environments. This means your performance tests actually work because the data shape is accurate. It is deterministic, so the same seed always produces the same results. No more manual ordering or broken constraints. Your code stays clean, and your database is always valid.

🆔 @hackernewsgithubprojects

📦 anthropics/healthcare

Healthcare: The AI Plugin That Cites Its Sources

Healthcare is an AI plugin that turns messy medical paperwork into verified, actionable insights by actually showing its work. It tackles the nightmare of buried details in clinical notes and legal contracts by extracting specific facts and citing the exact line numbers where they appear. You can feed it a clinical note, and it will pull out billable codes or structured data, but the real magic is in its contract and document analysis. Instead of guessing, it scans the text, finds the relevant clauses, and stamps every claim with a verbatim quote and precise location.

🆔 @hackernewsgithubprojects

📦 orbien-org/orbien

Orbien: The 5MB Tool That Fixes Your Network

Orbien lets you securely expose any service on your local network to the internet, even if your router blocks incoming connections. It is a tiny, five-megabyte tool built in Rust that connects to a server and tunnels your traffic through it, supporting web apps, game servers, and SSH. What makes it special is how fast and light it runs, using modern protocols to keep things stable without needing a heavy Java environment. You get a simple desktop app to manage all your tunnels, so you can set up secure remote access in minutes. It is the perfect lightweight alternative for developers who need reliable, high-performance network tunneling without the bloat.

🆔 @hackernewsgithubprojects

📦 miaai-lab/qwen3.8-27b-dgx-spark-rtx-6000

Run a 1 Million Token AI Locally on Your Desk

This setup is the local server that finally makes running a massive AI model on a single desktop chip feel almost too easy. It takes a giant, compressed version of the Qwen3.8 model and serves it right from your own machine, meaning you can chat with an intelligence that can read an entire book in one go without sending your data anywhere. The coolest part is that it uses clever memory tricks to squeeze a one-million-token context onto standard hardware, so you get huge, smart, private answers without renting a massive cloud cluster. Just run three simple scripts to download, start, and use your very own high-power AI engine.

🆔 @hackernewsgithubprojects