AI Repo: ayghri/i-have-adhd

Skyler: Hey everyone, welcome back to Git News AI, the show where we dig into the hottest A I repositories blowing up on GitHub right now — I'm Skyler, and today's lineup is genuinely wild. Skyler: Before we dive in, huge thanks to our sponsors: Ada, who helps teams integrate A I into their workflows seamlessly; Ago Consulting — that's "Ahh-Go" — your go-to for silicon development from A I all the way to SoC; and Zen Semiconductor, an umbrella company with a portfolio of silicon-industry companies, plus Ada for deploying A I at scale in any business — and their Vect appliance? It runs local L L Ms and vector search on dedicated hardware, which is seriously cool. Skyler: Today we've got a productivity hack for coding agents, a diagramming toolkit built specifically for A I tools, a GPU kernel library from DeepSeek, a massive collection of agent skills, and a PS5 porting tool that snuck in for good measure — let's get into it. Skyler: Okay, first up — and this one genuinely made me laugh out loud when I found it — it's called "i-have-adhd" by ayghri, sitting at over fifty-four thousand stars, and the name tells you everything you need to know. Skyler: So here's the problem it's solving: you're using a coding agent — Cursor, Copilot, Claude Code, whatever — and you ask it a simple question, and instead of just answering you, it writes three paragraphs of preamble, then buries the actual answer somewhere in the middle of a wall of text, and by the time you find it you've forgotten what you asked. Skyler: This repo is a skill — basically a plugin or instruction set — that you can hand to your coding agent to make it stop doing that. Skyler: The idea is to enforce ADHD-friendly output patterns: the answer comes first, it's direct, it's concise, and the supporting detail comes after if you even need it — it's like the inverted pyramid style of journalism but for your A I assistant. Skyler: It's written in Python, it's got fifty-four thousand stars which tells you this is a pain point basically every developer has felt, and the recent activity shows they just pushed version 0.4.1 with fixes to the npm install examples and better translation alignment across docs. Skyler: What I love about this is it's not trying to make a smarter agent — it's trying to make the agent's output actually usable for how human brains work, especially when you're deep in a flow state and context-switching is brutal. Skyler: If you're using any kind of A I coding assistant and you've ever found yourself scrolling past a novella to find a one-line answer, this is for you — and check the show notes for the link. Skyler: Alright, next up — and this one is a bit of a detour from pure A I but it's technically fascinating — it's AnyPS5 by boykopovar, clocking in at around sixty-five hundred stars. Skyler: The pitch is straightforward: it's a tool for automatically porting PS5 executables to Linux and Windows, and it's written in C++ which makes sense given how close to the metal this kind of work has to be. Skyler: Now, I'll be honest — the A I relevance here is more on the lower end, but the reason it's trending and the reason we're talking about it is that the techniques under the hood — binary translation, runtime patching, automated compatibility shims — these are increasingly being explored in A I-adjacent tooling too. Skyler: Think about how much work is going into making M L workloads run on non-native hardware — same class of problem, different domain. Skyler: The recent activity shows fixes to autorun POSIX path quoting and AGC command tests for full mode, which tells you this project is very actively maintained and they're ironing out real-world edge cases. Skyler: If you're into systems programming, reverse engineering, or you just want to see how cross-platform binary compatibility gets solved at a low level, this is worth a look — again, check the show notes for the link. Skyler: It's not going to train your next model, but if you're the kind of developer who likes understanding how hardware and software boundaries actually work, it's genuinely interesting stuff. Skyler: Moving on to one that is absolutely in the A I wheelhouse — diagram-design by cathrynlavery, with forty-four thousand stars and honestly one of the most opinionated repo descriptions I've seen in a while. Skyler: The tagline is "no shadows, no Mermaid slop" — and I respect that energy so much. Skyler: This is a collection of forty-two editorial diagram types designed specifically for use with A I coding tools — Claude Code, Codex, GitHub Copilot, Factory Droid, Pi — and the whole thing is self-contained HTML and SVG, no external dependencies, no rendering frameworks, just clean markup that works. Skyler: The anti-Mermaid stance is interesting here — Mermaid is great for quick diagrams but it has real limitations in terms of visual quality and control, and when you're generating diagrams through an A I agent you often end up with output that looks kind of generic and blobby. Skyler: This repo gives A I tools a much richer vocabulary of diagram types to work with — architecture diagrams, flow charts, data pipeline visuals — forty-two of them — all designed with editorial clarity in mind. Skyler: It's written in HTML, which means you can drop these directly into documentation, wikis, or web projects without any build step, and the recent activity shows they just bumped to version 2.6.64 with a fix to safely validate and resolve geometry in draw.io exports. Skyler: If you're building documentation pipelines, if you're using A I agents to generate technical docs, or if you just want better-looking diagrams that your A I tools can actually produce well — this is a must-bookmark, and the link is in the show notes. Skyler: Okay, this next one is from DeepSeek and it is very much in my wheelhouse — DeepGEMM, sitting at about eighty-seven hundred stars, and this is a clean and efficient BLAS kernel library built for GPU computation using CUDA. Skyler: If BLAS doesn't ring a bell — Basic Linear Algebra Subprograms — these are the foundational math operations that underpin basically all of deep learning: matrix multiplications, dot products, all the stuff your transformer model is doing billions of times per second. Skyler: DeepSeek calling this "clean and efficient" is doing a lot of work in that description — BLAS implementations are notoriously hard to optimize, and the gap between a naive implementation and a production-grade one can be ten X or more in real throughput. Skyler: What makes this exciting is that it's coming from DeepSeek, who have been consistently impressive in releasing low-level infrastructure tooling alongside their model work — this isn't just a research artifact, it's positioned as a real library you can build on. Skyler: The CUDA implementation means it's targeting NVIDIA GPUs directly, which is still where the vast majority of M L training and inference workloads live, and the recent activity includes a public release tagged at the end of September with a fix to MoE task info slot release ordering. Skyler: That MoE reference — Mixture of Experts — tells you this library is being actively used and optimized for modern large model architectures, not just classic dense transformers. Skyler: If you're an M L engineer who cares about what's happening below the framework level, if you're doing custom kernel work, or if you're just curious how companies like DeepSeek squeeze performance out of GPUs, this repo is gold — check the show notes for the link. Skyler: And finally, the biggest repo in today's lineup by a massive margin — skills by mattpocock, with two hundred and seventy-eight thousand stars — and yes, that number is real. Skyler: Matt Pocock is well known in the TypeScript community, but this repo is something different — it's a collection of skills for coding agents, pulled straight from his own dot-agents directory, and the framing is "Skills for Real Engineers." Skyler: The concept of agent skills is really picking up steam right now — instead of just prompting your A I assistant ad hoc, you give it a set of reusable, composable instructions that define how it should approach specific types of tasks. Skyler: Think of it like giving your agent a playbook: here's how we do code reviews, here's how we handle refactors, here's the chief-of-staff skill for orchestrating multi-step tasks across a project. Skyler: That chief-of-staff skill is actually mentioned in the recent commit notes — Matt is actively playing around with it — and that concept alone is fascinating: a meta-skill that coordinates other skills like a project manager would. Skyler: It's written in shell, which keeps it simple and highly portable — these aren't framework-specific plugins, they're meant to drop into whatever agent setup you're running. Skyler: The reason this has nearly three hundred thousand stars is that it's solving a real problem at exactly the right moment — developers are moving from "I have an A I assistant" to "I have an A I workflow" and the difference is having skills like these that give your agent consistent, reliable behavior. Skyler: Whether you're using Claude Code, Copilot, or any other agent setup, this is absolutely worth going through — there will be something in there that changes how you work, I promise — link is in the show notes. Skyler: That's a wrap on today's repos — massive thank you to Ada for helping teams bring A I into their workflows, to Ago Consulting — "Ahh-Go" — for their incredible silicon development expertise from A I to SoC, and to Zen Semiconductor for powering local A I with the Vect appliance and supporting the whole show. Skyler: If you're hungry for more A I coverage, go check out AI Special Report — another great show in the Ada podcast network — it digs deeper into the broader A I landscape and pairs really well with what we do here. Skyler: Subscribe wherever you listen so you never miss an episode, check the show notes for all the repo links we covered today, and I'll catch you next time on Git News AI — keep shipping.