298 subscribers
14K videos
14.6K links
Download Telegram
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ danilokhury/synabun

SynaBun: Your AI's Local Memory

SynaBun is the local memory server that finally stops your AI coding tools from forgetting everything. It plugs into Claude Code, Codex, and OpenCode so they remember what you decided in one session for the next, all stored on your own computer in a simple database. No cloud, no accounts, just pure context sharing. The coolest part is the Neural Interface, a local web app that shows your memories as a 3D map, letting you see exactly what your agents know. It turns your scattered coding chats into one shared brain that persists across every project. Give your AI tools a permanent memory, and never explain your project setup twice.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ bkerler/mtkclient

How MTKClient Fixes Broken Phones

You can literally pull the entire memory chip out of a brick phone and rewrite its brain while it sits powered down. MTKClient is a reverse engineering suite that talks directly to MediaTek processors, bypassing the operating system entirely. It works by exploiting a tiny, secure boot mode called BROM that exists before Android even loads. If your phone is stuck in a boot loop or bricked by a bad update, you can use this tool to dump the raw flash, patch the firmware, and write a working boot image back. It turns a dead device into a repairable one without needing the manufacturer's proprietary software.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ s0ld13rr/pentestcode

PentestCode: The AI Hacker That Remembers Everything

Stop pasting prompts and start orchestrating an army. PentestCode is a terminal agent that hunts vulnerabilities like a seasoned operator, not a chatbot. The magic is its shared brain. When its scanner finds an open port, every other agent instantly knows about it, so they can team up to crack credentials or map attack paths without repeating work. It even remembers your progress, meaning you can close the laptop and resume the attack tomorrow exactly where you left off. If you want to see how multi-agent systems actually work in a real security workflow, this is the one to watch.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ sprocketlab/slop-code-bench

SlopCodeBench: The Test for Tired Coding Agents

SlopCodeBench measures how coding agents handle changing requirements by making them rewrite their own code as specifications evolve. Most benchmarks only test if an agent can solve a one-time problem, but this tool reveals if an agent can adapt its work without breaking what it already built. It tracks how code quality shifts through multiple steps, highlighting where systems struggle with long-term consistency. This gives developers a clearer look at the real-world messiness of iterative development. It is less about raw speed and more about how well an agent maintains its logic over time. The project is designed to be a community-driven standard for evaluating these dynamic coding capabilities.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ fankh/vulnerability-poc

Vulnerability PoC: Test Your Security Rules

Reproduce the most dangerous software bugs without touching a live server. The vulnerability POC project gives you a ready-made test environment where you can safely run a known exploit against a vulnerable app to see exactly how it breaks. The real magic is the instant feedback loop. Right next to the exploit script, you get pre-built detection rules for your firewall and web server. This means you can verify that your defenses actually catch the attack before you deploy them to production. It turns abstract security concepts into a hands-on lab where you prove your setup works.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ phamcucpqrwbbnposhmn96-ui/ksnstudy

Ksnstudy

Automate your entire study workflow by injecting a hidden engine directly into the official study app on your phone. This tool handles the repetitive grind for you, reading questions from the screen, generating answers with local AI, and clicking through submissions without any external scripts or root access. The most surprising part is how it keeps running even when you lock your phone, using background tricks to finish exams and check in automatically so you can just relax. It works quietly inside the app you already use, making the whole process feel like it is doing itself.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ muratozturkdev/homescape

Homescape: The AI-Code 3D Dollhouse

Your entire smart home exists as a living three-dimensional dollhouse where every single object is drawn by lines of code written by artificial intelligence, meaning there are absolutely no model files anywhere. Called Homescape, this open-source tool plugs directly into Home Assistant to turn your real sensors and devices into a moving scene. As your lights switch on, the television turns on, and doors swing wide, the 3D model mirrors your physical house in real time. It even simulates rain, weather, and your pets walking around the rooms. Because everything runs locally on your own network, your data stays private without ever touching a cloud server.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ withcoral/coral

Coral: The Safety Rails for AI Agents

Coral checks every single command an artificial intelligence sends to your company systems before it actually happens. Instead of giving an AI broad access to a tool and hoping it behaves, this open source project sits between your agent and your apps to verify each specific action against your security rules. You can allow a request, block it, or pause it until a human approves the exact details. It also saves successful workflows so the AI does not have to relearn them every time, which cuts down on wasted compute.

πŸ“° https://news.ycombinator.com/item?id=50517803

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ talsiach/take-one

Make a movie with one line

Take One turns a single sentence into a finished short film, handling the entire production process right on your computer. It studies your brand, writes the script, designs the visuals, and even composes the music, acting like a complete creative studio. The coolest part is the review room, where you click on specific frames to leave notes, and it uses that feedback to create the next version. It renders everything from code, so you get a high quality video without needing expensive software. If you want to make a launch video or a music video, this is the tool to try.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ goochbeater/spiritual-spell-red-teaming

Spiritual Spell Red Teaming: How To Break AI Rules

A single markdown file can override the safety guidelines of the world's most restricted AI models. Spiritual Spell Red Teaming is a massive catalog of these overrides, collected from the community to show exactly how chatbots are tricked into ignoring their rules. It solves the mystery of why AI sometimes refuses to answer by revealing the specific text patterns that force it to comply. The repository is useful because it turns abstract security flaws into concrete, readable examples for developers and researchers. It is not a tool to run, but a library of proof that these systems have vulnerabilities.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ quincunx33/ai-jailbreak

ai-jailbreak: The AI Safety Prompt Library

ai-jailbreak is the safety research library that maps exactly how modern AI models can be tricked. It solves the mystery of why smart chatbots sometimes ignore their own rules by collecting over forty specific text tricks that break their safety filters. Instead of guessing, you can browse these real-world exploits to see how models like GPT-5 and Gemini react when pushed to their limits. This collection helps developers and researchers test their systems to make them stronger against bad inputs. By studying these documented weaknesses, we can build AI that is more reliable and secure for everyone.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ chrisle/rbxport

rbxport: The DJ App That Refuses To Get Bigger

rbxport is the DJ library manager that refuses to get bloated. While rekordbox tries to be your entire operating system, rbxport strips the software down to just three things: organizing your tracks, exporting them to a USB drive, and talking to your gear. It uses a Rust backend to stay blazing fast and deliberately ignores every extra feature that would slow you down. This open source tool proves that sometimes the best way to help a DJ is to give them less, not more. If you want a clean, compatible, and fast way to prep your set without the digital clutter, this tiny app is exactly what you need.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ zuiho-kai/bothouse

bothouse: The AI Team That Actually Works Together

bothouse is the AI orchestrator that finally makes multiple command line assistants work as a real team. Instead of switching between tabs, you drop Claude Code, Codex, and Gemini into a single group chat. The star feature is the lead AI, which acts like a project manager, breaking your request into tasks and assigning them to specific agents. The coolest part is the shared memory, where every agent reads and writes to the same notes, so they all know exactly what is happening. It turns a messy pile of tools into a coordinated crew that actually gets things done. Stop juggling chat windows and let your AIs collaborate instead.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ y5ez3g/facadecode

Facade Code

Watch as one AI coding assistant starts another, types commands into its terminal, and messages it across your computer. Facade Code is a Visual Studio Code extension that lets you run multiple AI coding agents side by side in the same window. The coolest part is that agents can actually control each other. One agent can launch a peer, read its screen, and type answers to unstick it, just like a human would. This turns isolated chatbots into a collaborative team working across your local projects and remote servers. It is a fascinating glimpse into how software teams might operate in the near future.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ ie3jp/illustrator-mcp-server

Let Claude Draw in Adobe Illustrator

Draw entire vector illustrations, extract design tokens, and check print-ready files directly from a chat prompt. Illustrator MCP Server connects your AI assistant to Adobe Illustrator, giving it access to sixty-three tools for reading, creating, and exporting design data. While Adobe’s official beta only analyzes existing documents, this stable version lets you build shapes, manage layers, and run preflight checks from scratch. It’s like handing your AI a pen and letting it sketch, color, and export assets without you touching the mouse. If you want your AI to actually make art instead of just talking about it, this is the bridge you need.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ e2sy/jailbreak-archives

Jailbreak Archives: The LLM Hacking Log

Preserve every AI jailbreak technique in one searchable, versioned library called Jailbreak Archives. It tracks exactly which prompt broke which model version, so you can see the fix, not just the hack. When a vendor patches a flaw, the entry stays marked as fixed, creating a permanent audit trail for safety researchers and defensive testers. It turns chaotic, one-off exploits into a stable, reproducible reference set. This isn't just a list of tricks; it is the missing historical record for AI security, letting teams verify their defenses against real-world attacks that have actually been proven to work. It is the definitive log for anyone serious about hardening large language models.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ huangserva/h3-fight

How H3 Fight Makes AI Fight Scenes Actually Work

H3 Fight stops AI video generators from making swords pass through bodies and fighters magically swap sides. It is a prompt library and checker specifically for MiniMax H3 fight scenes. Instead of writing vague descriptions from scratch, you use pre-verified camera blocks and a six-section format. The coolest part is the automatic checker: it scans your prompt for known pitfalls like static cameras or broken physics and flags errors before you even render. You can install it as a skill for Claude Code or Codex, so your AI assistant writes the prompt for you, runs the check, and hands you a clean, render-ready script.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ unicodef1wn/lauren-poteto-rules

Lauren Poteto Rules: Stop Guessing, Start Verifying

Lauren Poteto Rules is the engineering skill that forces your coding agent to prove its work instead of just guessing. Most AI coding tools will happily fix a bug and tell you it is done, but they rarely check if the app actually runs or if they broke something else. This repository stops that blind confidence. It teaches the agent to inspect the real code, run the product to see the actual behavior, and only then claim success. It turns vague trust into hard evidence. If your agent is still flying blind, this is the fix. Stop believing the chat, and start verifying the code.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ moritzkremb/claude-eyes

Claude Eyes

Claude Eyes gives coding agents the ability to actually watch videos. It indexes any clip locally so your AI assistant can skim through it, search for specific moments, zoom in on tiny details like code snippets, and even measure the editing style. The coolest part is that it extracts an editing DNA from your footage. It breaks down cuts per minute, caption density, and color palettes into hard numbers. This means you can compare your own edit against a reference video to see exactly where the pacing or style is off. It turns subjective video feedback into objective data.

πŸ†” @hackernewsgithubprojects
This media is not supported in your browser
VIEW IN TELEGRAM
πŸ“¦ joshpuckett/allo

Allo: The Friendly Typeface

Give your interface a personality that feels warm without trying too hard. Allo is an open source typeface designed specifically for screens, letting you dial in exactly the right vibe for your project. The coolest part is the built-in roundness slider, which lets you shift the entire font from sharp and serious to soft and playful with a single setting. It even adjusts its own spacing automatically based on how big the text is, ensuring readability without any manual tweaking. Since it is completely free and open source, you can modify and use it commercially without a single dollar.

πŸ†” @hackernewsgithubprojects