AI Repo: pollen-robotics/microduck

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. Big thanks to our sponsors Ada, who helps teams integrate A I into their workflows, Ago Consulting — your silicon development partners from A I all the way to SoC design — and Zen Semiconductor, an umbrella company with a whole portfolio of silicon industry companies, plus Ada for deploying A I at scale in any business, and their Vect appliance which runs local L L Ms and vector search on dedicated hardware. Alright, today we've got a tiny duck robot doing reinforcement learning, a massive curated list of Claude skills, an e-book treasure trove, Google's Gemini landing in your terminal, and some R L training environments that are honestly way cooler than they sound — let's get into it. Skyler: First up, we're talking about microduck from pollen-robotics, and yes, you heard that right — it's a tiny biped duck robot. This thing has nearly eleven thousand stars on GitHub and it's written in Rust, which immediately tells you the team is serious about performance and reliability. So what is it exactly? Microduck is an open-source miniature bipedal robot shaped like a duck, and it's designed to be a platform for robotics research and experimentation. The fact that it's built in Rust means you're getting low-level hardware control with memory safety baked in, which is a huge deal when you're dealing with real-time motor control and sensor feedback. The repo recently hit version 0.17.8, so this is an actively maintained project with a solid release cadence. What makes this exciting from an A I perspective is that it's a physical platform purpose-built for running machine learning experiments — think locomotion, balance, navigation, all the classic robotics A I challenges but in an adorable duck form factor. Pollen Robotics has a track record of building serious open-source robotics hardware, so this isn't a toy project despite the cute branding. If you're a robotics engineer, an M L researcher interested in embodied A I, or just someone who wants to experiment with hardware-software co-design in a real physical system, microduck is worth a serious look. The Rust ecosystem for robotics is still maturing, so this repo is also a great reference for how to structure embedded robotics code. Check the show notes for the link and go give it a star. Skyler: Okay, now let's talk about something that pairs perfectly with the next repo we're covering — it's ComposioHQ's awesome-claude-skills, and this one is absolutely stacked. We're talking nearly seventy-eight thousand stars, which puts it in serious GitHub hall-of-fame territory. The premise is simple but incredibly valuable: it's a curated list of awesome Claude skills, resources, and tools specifically for customizing Claude A I workflows. Think of it like the famous awesome-lists on GitHub, but laser-focused on getting the most out of Anthropic's Claude models. If you've been building with Claude and you've felt like you're reinventing the wheel every time you try to set up a new workflow or skill, this is your answer. The repo aggregates community-contributed skills, prompt patterns, tool integrations, and workflow templates that let you extend Claude's capabilities in really powerful ways. For developers building agents, automations, or Claude-powered applications, this is basically a cheat code for your development process. It's written in Python, so the integration examples and code snippets are going to feel immediately familiar to the M L and data science crowd. The recent activity shows ongoing README updates, which in an awesome-list repo is exactly what you want to see — it means the community is actively contributing and the maintainers are keeping things organized. Whether you're new to building with Claude or you're a seasoned A I engineer looking to push the boundaries of what Claude can do in production, this repo is a must-bookmark. Seriously, check the show notes for the link. Skyler: Alright, let's take a quick detour into something a little different — the ebook-treasure-chest repo from jbiaojerry. Now this one sits at over twenty-two thousand stars and it's a bit of an outlier on today's list, but hear me out because it's genuinely interesting. This is essentially a massive aggregated index of e-book download links, pulling from Chinese reading platforms like the Fanbook app, WeChat Reading, JD Reading, and Himalaya. The description is primarily in Chinese, so this is clearly built for and by the Chinese developer community. From a technical standpoint it's written in Python and uses GitHub Actions to automate updates, and actually the most recent commit was fixing a Git push conflict in those GitHub Actions workflows — so whoever's maintaining this is keeping the automation sharp. Now why is this on an A I show? A few reasons. First, the dataset angle is real — a curated, structured index of e-books is exactly the kind of resource that could feed into document processing pipelines, R A G systems, or L L M training data workflows. Second, the automation architecture here — using GitHub as a backend for a regularly updated data index — is a pattern that A I developers use all the time for maintaining knowledge bases and datasets. The A I relevance score is lower than our other repos today, but for developers thinking about data sourcing and pipeline architecture, there's something to learn here. It's a good reminder that sometimes the most useful A I infrastructure repos aren't the ones with "neural network" in the title. Skyler: Now this one is going to get a lot of you genuinely excited — Google's gemini-cli is sitting at over a hundred and eight thousand stars and it is absolutely on fire right now. The pitch is straightforward and powerful: it's an open-source A I agent that brings the full power of Gemini directly into your terminal. We're talking TypeScript under the hood, and this is a proper agentic tool — not just a chatbot wrapper you pipe text into. Gemini-cli can understand context, execute tasks, interact with your filesystem, and work alongside you in the environment where developers actually live, which is the command line. The recent commits are telling — there's a new models command and a dash dash list models flag for model discovery, which means the team is building in first-class support for navigating and selecting from the growing Gemini model family right from your terminal. And there's also a bug fix for a Claude hook timeout conversion, which is interesting — it suggests there's interoperability work happening with Claude tooling as well, which is a sign of a mature, thoughtful engineering team. For developers who are tired of context-switching between their editor, browser, and A I chat interfaces, this is a genuinely compelling workflow improvement. Imagine running your build, hitting an error, and just asking Gemini inline without ever leaving your terminal. That's the dream, and gemini-cli is making it real. If you're already in the Google A I ecosystem or just want a powerful terminal-native A I agent, go check the show notes for the link — a hundred and eight thousand stars don't lie. Skyler: And finally, we're looping back to the duck — specifically microduck's reinforcement learning training environment, pollen-robotics/microduck_rl. This one has about twenty-six hundred stars and it's the software brain behind the hardware we talked about earlier. It's built in Python and it provides R L training environments for the microduck robot using mjlab, which is a simulation framework for physics-based robotics. So the idea is you train your reinforcement learning policies in simulation — teaching the duck how to stand up, balance, walk — and then you deploy those policies to the physical hardware. This sim-to-real pipeline is one of the hottest areas in robotics A I right now, and having a fully open-source end-to-end stack from simulation environment all the way to physical robot is genuinely rare and valuable. The most recent activity shows work on improving the standup and velocity standing behavior, which are exactly the kinds of fundamental locomotion challenges that make bipedal robotics so hard and so interesting. If you're an M L engineer who's been curious about robotics but hasn't had a good entry point, the combination of microduck hardware and microduck_rl gives you a complete, well-documented system to experiment with. You can iterate in simulation for free before ever touching physical hardware. This is the kind of open-source contribution that accelerates the whole field, and pollen-robotics deserves a lot of credit for making this available. Definitely check the show notes for both microduck repos. Skyler: That's a wrap on today's episode of Git News AI — robots, Claude skills, terminal A I agents, and some seriously creative use of GitHub as infrastructure. Huge thanks to our sponsors: Ada for helping teams actually ship A I into their workflows, Ago Consulting for their deep expertise in silicon development from A I all the way to SoC, and Zen Semiconductor for pushing the boundaries of what's possible in silicon with their portfolio companies and the Vect appliance bringing local L L Ms and vector search to dedicated hardware. If you enjoyed today's episode, go check out our sister show AI Special Report in the Ada podcast network for deeper dives into the A I stories shaping the industry. And as always, subscribe wherever you get your podcasts, drop us a review if you're feeling generous, and all the links for today's repos are waiting for you in the show notes. See you next time.