Sam: NVIDIA is promising 'deterministic query generation' for AI.
Kai: I'm Kai!
Sam: And I'm Sam.
Kai: And this is Trending Repos for Friday, November 7th, 2026. We've got three incredible open-source projects for you.
Sam: Let's see if they hold up to scrutiny.
Kai: So, NVIDIA is trying to make AI more explainable with a new graph-based framework... and it has some very big ambitions.
Sam: Then there’s a new repo claiming to have the essential skills for 'Real Engineers'. I think we'll be the judge of that.
Kai: And get this: one of Silicon Valley's biggest names has open-sourced his entire C-suite... as a stack of Claude prompts.
Kai: Alright, let's dive into Auto Ontology. It's a new Python framework from NVIDIA, and Sam, this thing looks like a game-changer.
Sam: 'Game-changer' is what they always say, Kai. The repo talks about building 'semantic, graph-based reasoning layers'... which is a mouthful of buzzwords.
Kai: But it's what those buzzwords do! You can create knowledge graphs over your existing data, and it makes context retrieval for your AI way, way better.
Sam: And it enables 'deterministic query generation'—your cold open tease. I'm skeptical. 'Deterministic' and 'AI' in the same sentence usually means one of them isn't true.
Kai: It means when you ask a question, it generates a predictable SQL or Cypher query that you can actually read and verify. It’s not a black box.
Sam: So it’s a fancy, graph-aware text-to-SQL generator. Fine. But the other promise is 'more explainable AI'. Is it actually explainable, or does it just produce a graph that looks like an explanation?
Kai: It's a huge step! You can trace the reasoning path. So instead of getting a magic answer from an LLM, you get the query. Something you can actually run, debug, and optimize.
Sam: A traceable path is better than no path, I'll grant you that. But let's not pretend this solves the AI alignment problem.
Kai: No, but it makes building RAG systems that don't hallucinate a lot more achievable. That's a win.
Sam: A win I'll believe when I see it running in production without a team of PhDs to babysit it.
Sam: Alright, next up: a repo called 'skills' from Matt Pocock.
Kai: Oh, the TypeScript wizard! I love his stuff. The repo description is just: 'Skills for Real Engineers. Straight from my .agents directory.' That's the whole thing!
Sam: The confidence. So what are these 'skills'? Are we talking nunchuck skills, bow hunting skills...
Kai: It’s a collection of his personal scripts and tools for AI agents. Think of it as a toolbelt you can give to your coding assistant. There are skills for refactoring, analyzing repos, generating components...
Sam: So... dotfiles for AI. And what makes them skills for 'Real Engineers'? Is there an entrance exam?
Kai: It's a bit tongue-in-cheek, sure. But the idea is what's important. Instead of a generic chatbot, you give your agent specialized, high-quality tools. It makes them more capable and reliable.
Sam: Or, it's a collection of scripts that work perfectly for his workflow and will break spectacularly in anyone else's. 'It works on my machine' is not a skill.
Kai: Look, you're not meant to just clone it and expect magic. It's a design pattern. It's a fantastic reference for how to structure your own agent's tools.
Sam: Fair enough. It's an interesting look behind the curtain. But I'm waiting for the first pull request that tries to add 'arguing on Hacker News' as a skill.
Kai: It would be the most-used skill in the whole repo. Okay, so if a toolbelt isn't enough, how about an entire C-suite in a repo?
Sam: Oh boy. You're talking about 'gstack'.
Kai: From the one and only Garry Tan! It's his exact Claude setup—23 opinionated tools to act as your CEO, Designer, Eng Manager, Release Manager...
Sam: ...Doc Engineer, and QA. I read the README. It promises the whole org chart.
Kai: You can basically run a one-person company with Garry Tan's brain as your co-pilot! How cool is that?
Sam: Let's be real. It's a set of 23 very opinionated prompts. And naming a stack after yourself, 'gstack'... it feels a little... much.
Kai: Come on, it's branding! And the prompts are solid. It's a whole system for using an LLM to manage everything from product strategy to a press release. It’s a framework for thinking.
Sam: Here's the trap, though. You're not just adopting a tool; you're adopting an entire, rigid philosophy of how to build a company. What happens when Garry Tan's 'CEO' prompt gives you advice you know is dead wrong for your market?
Kai: Or, you get a massive head start by standing on the shoulders of one of the most successful investors and founders out there. For a solo dev trying to ship, these guardrails could be invaluable.
Sam: I prefer building my own guardrails, thank you. Not someone else's 'g-rails'.
Kai: Okay, before this gets any worse... So, NVIDIA is making AI queries more predictable with Auto Ontology.
Sam: Matt Pocock shared his agent's 'skills'—for 'real engineers'.
Kai: And Garry Tan released 'gstack' to let you clone his entire executive team using Claude.
Sam: Before we go, a quick one: did you see the latest update to `oh-my-zsh`?
Kai: Don't tell me, a new theme that finally makes my terminal sentient?
Sam: Almost. It's a new plugin that plays the 'Law & Order' dun-dun sound effect whenever a command fails. It's completely useless and I've already installed it.
Kai: And that's our show! We'll have links to all these projects—and maybe that 'dun-dun' plugin—in the show notes. Thanks for tuning into Trending Repos.
Sam: We'll be back tomorrow. Try not to install any 'g-stacks' before then.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.