Skills surge: nature-skills leads, Colibri runs 744B on 25GB
We don't tell you what's popular — popularity lags and is gameable. We tell you what's gaining speed right now, and whether it's worth your attention. 6 accelerating AI repos earned today's slot.
Top mover
This repo provides a curated set of skills (prompts/templates) for generating Nature-level academic text and scientific figures using Codex or similar LLMs. It's gaining traction because researchers want to streamline manuscript preparation and achieve publication-quality visuals without manual formatting. However, it's essentially a prompt library, not a tool—results depend heavily on the underlying model and user input.
🤖 Agents & automation
SkillOpt optimizes natural-language skills for frozen LLM agents by editing trajectories and validating updates, enabling self-evolving agents without retraining. It's gaining traction because it addresses the key challenge of improving agent performance efficiently. A caveat: the skill optimization may overfit to specific trajectories, limiting generalization.
This repo provides a structured way to define and share personal AI agent skills, enabling modular, reusable capabilities for agents. It's gaining traction because it simplifies building custom agent behaviors without heavy frameworks. However, it's early-stage and may lack production robustness.
⚙️ Inference & serving
Colibri enables inference of the massive GLM-5.2 (744B MoE) model on consumer hardware by streaming experts from disk, using pure C with zero dependencies. This matters because it dramatically lowers the hardware barrier for running state-of-the-art MoE models, but the trade-off is slower inference due to disk I/O and limited to single-batch processing.
Tooling & infra
This tool lets you run Apple iOS simulators headlessly and serve them via HTTP, enabling CI/CD testing and agent-based automation without a GUI. It's gaining traction because it fills a gap for developers needing to automate iOS testing in cloud or headless environments, similar to how npx serve simplifies local web serving. A caveat: it relies on Apple's simulator infrastructure, so it only works on macOS and may have stability issues with complex app interactions.
This tool converts technical book PDFs into structured skill files for Claude Code, enabling developers to query and reference book content directly in their coding workflow. It's gaining traction because it bridges the gap between static PDF knowledge and interactive AI-assisted development, but it depends on the quality of the PDF's text extraction and may not handle complex layouts or scanned pages well.