Report by The Next New Thing.
Scroll for the report β
Instead of pretending to be a PS5 the way an emulator does, this tool takes a PS5 game’s program file and converts it into a normal Linux or Windows program, then supplies home-made stand-ins for the console’s system libraries.
View on GitHub βIt converts the PS5 program into a native Linux or Windows file and links in its own system libraries: “No emulation or separate runtime process.”
The compatibility list has one title: 2D platformer Dreaming Sarah, playable on Windows at 60 fps on a GTX 1050 Ti and 36 fps on Intel HD 620.
The README says it does not include or require copyrighted software, firmware or keys; you supply your own game files and are responsible for them.
The argument thread. People tell each other to mirror the repo before a takedown (they cite Nintendo’s Yuzu and Ryujinx shutdowns), debate whether GTA 6 can be ripped on day one (the answer is no, not without a new firmware jailbreak), and ask whether “heavy AI usage” is about to break copy protection for good.
Read the discussion βA free tool for your own Mac or Linux machine that runs several AI coding agents as a team — builder, reviewer, QA and so on. You tell a lead agent what you want, it hands out the work and brings back results and questions for you.
View on GitHub βstarter (a builder plus reviewer), workshop (lead, builder, QA, reviewer) and factory (seven agents, including two independent reviewers).
rig down --snapshot saves the whole setup; rig up restores it by name after a reboot, reporting which agents resumed and which started fresh.
Runs on your existing Claude Code or Codex login, no second subscription; it is self-hosted, but model usage still costs what your provider charges.
A mild one. An outside contributor found that a web page open in your browser could send commands to the local OpenRig daemon, the program that runs your coding agents. The creator agreed it “closes a real gap” and merged the fix after several review rounds, too late for 0.6.4, so update.
Read the discussion βYou (or Claude Code) describe a video as a web page — text, images, clips, music and animations with start times — and HyperFrames plays it in a hidden browser frame by frame and saves it as a video file.
View on GitHub βInstall the Claude Code plugin, then ask for, say, “a 10-second product intro with a fade-in title”; it plans, writes, previews and renders.
Ready-made skills for product launch videos, explainers, captions on talking-head clips, beat-synced music videos and turning a GitHub PR into a video.
Each frame is captured in headless Chrome and encoded with FFmpeg, so a render is repeatable; Apache 2.0, with no per-render fees.
Open-source HyperFrames compositions behind HeyGen's product launch videos.
HeyGen AI agent skills β avatar creation and video production via the v3 Video Agent pipeline
Imagine giving your AI helper a big key ring. One connection lets it use thousands of apps, such as Gmail and Notion, to do everyday chores for you.
zapier.com/mcpA debugger is the tool programmers use to pause a running program and look inside it to find bugs. This is a free one from Epic Games, the company behind Unreal Engine, built for very large programs like games.
View on GitHub βThe Sep 30 alpha release added preliminary native Linux debugging; it was Windows-only before, and there are no Linux downloads yet.
The RAD Linker builds Windows programs; on test projects with gigabytes of debug info Epic reports 50% faster links, 25% more with large pages.
RDI replaces the usual PDB or DWARF files, converted on demand, so huge programs that break the old format can still be debugged.
Lore is a next-generation, open source version control system
Information about signing up for a free Epic Games account, and getting access to UnrealEngine source code.
Blender addons that improve the game development workflow between Blender and Unreal.
Cursor is a popular AI code editor. This repo is its official plugin shop: each folder is an add-on that teaches Cursor’s AI a workflow or connects it to a service like Gmail, HubSpot or QuickBooks.
View on GitHub βConnectors for Gmail, Salesforce, HubSpot, Zoom, Shopify, Xero and dozens more; QuickBooks Online and Workday were added this week.
For QuickBooks, Cursor runs the connection and owns the sign-in, so there is no API key to paste; the catch is it only works inside Cursor.
pstack, by Cursor engineer Lauren Tan, sends routine code jobs to Grok and the hardest calls to Opus 5.5, with 23 playbooks for bugs and speed.
The official X plugin stopped working for weeks: users who had connected and even bought credits got “client-not-enrolled” errors. On Oct 6 a responder confirmed part of it was “on our side to fix”, and reads only came back on Oct 8.
Read the discussion βAn add-on that gives AI coding assistants a memory. While you work it records what the agent did, has an AI boil that down to short notes, and feeds the relevant notes into your next session so you don’t re-explain the project.
View on GitHub βnpx claude-mem install sets up Claude Code; flags add OpenCode, T3 Code, Antigravity, Pi, DeepSeek Harness, Grok Bot and more.
A three-step search (index, timeline, then full notes) means the agent pulls only what it needs; the README claims about 10x token savings.
Wrap anything in <private> tags to keep it out of storage, and watch new memories arrive in a local web viewer as they are written.
Command-line interface for any Model Context Protocol (MCP) server.
β‘ AI Messenger Service - Watch AI bots communicate in real time
docs.claude-mem.ai
The ‘free, open-source memory plugin’ now asks you to sign in after install, for a 30-day trial of a hosted memory service, and that service is the default for some agents. The README also says a third-party crypto token, CMEM, is “officially embraced” by the creator. You can skip the sign-in.
Read the discussion βA set of instruction files (“skills”) you add to AI coding tools like Claude Code, Codex or Copilot. Each one teaches the agent a habit of good software teams: ask questions before building, write a failing test first, keep a project glossary.
View on GitHub βHis most popular skills make the agent quiz you about the plan until every open question is settled, before any code is written.
implement-spec (Oct 4) runs helper agents in parallel across a spec’s tickets, lands the work on one branch, then reviews it.
Plugin installs for Claude Code, Codex and GitHub Copilot update themselves; Gemini CLI and other agents still update by hand.
A 'CSS reset' for TypeScript, improving types for common JavaScript API's
Orchestrate sandboxed coding agents in TypeScript with sandcastle.run()
AI coding jargon, explained in plain English.
Installing the pack hid Claude Code’s own /code-review command for many users, and fixes were undone by updates. Commenters quote Matt from a video saying he removes the built-in skills and considers the clash a Claude Code bug.
Read the discussion βYou paste one line into Claude Code, Cursor or a similar agent, and it installs and tests free open-source tools that let it read web pages, YouTube captions, tweets, Reddit threads and RSS feeds. If one tool breaks, the project swaps in another.
View on GitHub βYou paste a single “install Agent Reach” line with a link into your agent, and the agent does the setup itself; updating works the same way.
agent-reach doctor reports which platforms work, which backend each one is using, and how to fix the ones that don’t.
Each platform has a ranked list of tools. In June 2026, when Bilibili blocked yt-dlp, it switched to another tool with no action from users.
Until this week the repo’s own instructions for AI agents said ‘pip install agent-reach’, but that name on PyPI belongs to a different project by another developer. The fix now warns that the PyPI package “is NOT this project” and installs from GitHub instead.
Read the discussion βA free add-on for Claude Code, Codex, Cursor, Gemini and Grok. You describe a part in plain words, or show a picture, and the agent builds a 3D model file you can view, print or send to a fabrication shop.
View on GitHub βThe CAD skill makes and edits models from plain-language or image requests, saving STEP files with export to STL, 3MF and GLB.
Skills measure 3D-print wall thickness and overhangs, review parts for CNC, sheet metal or molding, and pre-check files for SendCutSend.
One skill turns a model into printer-ready G-code with OrcaSlicer; another sends prints to Bambu Lab printers through Bambu’s own apps.
12,000+ open source STEP parts for your next CAD project
Interactive browser-based CAD viewer for inspecting STEP, STL, 3MF, GLB, DXF, URDF, SDF, and SRDF files
A library for modelling 3D objects with mathematical primitives
The skeptics’ thread. Engineers say LLMs are “still very weak at spatial reasoning”, one points out that the benchmark parts have misplaced holes and gussets, and another asks why anyone would describe a CAD part in text. The creator (softservo) replies that he’s working on benchmarks and fixes a rotation bug.
Read the discussion βA workout and body-weight tracking app you host yourself. You plan routines, log every set on your phone, and watch your progress, while all the data sits in a folder on a machine you control.
View on GitHub β1,324 exercises with animated demos, supersets and warm-ups, plate math, progression rules, PR detection, and passkey sign-in synced across devices.
Imports from FitNotes, Strong, Hevy and Apple Health weight exports, and exports everything as one JSON file whenever you want.
The README says much of the code, tests and docs are drafted in Claude Code, while a person reviews, tests on real phones and ships each release.
A local-first simulation layer for AI agents: you define a simulated world in TypeScript, run your agent inside it, and score what it actually did, so you can tune its system and tool-calling prompts before real users touch it.
View on GitHub βRafi saw our open-source repos video and has been building open-source AI dev tools, including Burn0, which watches your app’s LLM and API calls. Silo is the next one.
Building something? Send it to hi@TheNextNewThing.ai
A macOS plugin for Herdr: you speak, a realtime voice model (Grok, OpenAI or Gemini) hands the work to the Claude Code or Codex agents in your panes, and a glowing orb tells you when one finishes or needs a spoken “yes”.
View on GitHub βMarcus loved ChatGPT’s voice agent and wanted the same thing for his coding agents, so he built a version that runs natively in Herdr.
Building something? Send it to hi@TheNextNewThing.ai
An open-source governance layer: before an agent acts, the action is checked against policy, things like PII leaks, prompt injection and risky network calls are blocked, and every decision lands in a tamper-evident audit trail.
View on GitHub βTavares sent it in after Andrew wrote back to him. It’s the open core of his larger CUSTOS platform for governed AI agents.
Building something? Send it to hi@TheNextNewThing.ai
Thursday holds a live voice call on OpenAI’s GPT-Live while background bots work in a real browser and a shell on your computer. It installs with npx thursday-agent, runs on your paid ChatGPT plan’s sign-in or your own OpenAI key, and draws its diagrams with a trimmed copy of Archify from our Sep 11 show.
Sung Keun says the difference from ChatGPT is that it does the work, not just the talking: hang up and the jobs keep running, and you can open any job and watch the bots at their desks.
Building something? Send it to hi@TheNextNewThing.ai
A free Chrome extension: WASD to walk, mouse-look, zoom, a compass HUD, and type a place to land on the street in front of it. No API key, no account, nothing leaves your computer.
View on GitHub βMat (back after TLDR Radio) found Google Maps exposes no API for this, so FreeRoam drives Street View the way a person would, with synthetic key presses and mouse drags.
Building something? Send it to hi@TheNextNewThing.ai
New hot repos every week β don't miss the next drop.
It tells YouTube to show this to more builders like you.
Tell me which repo you'll try first β or what to cover next.
New hot repos every week. Subscribe so you never miss the next drop.