If yesterday's trending page was about structuring AI coding agents with composable skills, today's is about expanding where agents actually operate. New repos are pushing agent-native tooling into scientific research, GUI software automation, and voice interfaces — a sign that the "coding agent" framing is giving way to something broader. The common thread: developers want agents that can reach into previously inaccessible domains without writing custom wrapper code from scratch.
135 Pre-Built Skills Bring Agent-Native Workflows to Science
K-Dense Inc. published a curated library of 135 agent skills targeting 17+ scientific domains including genomics, drug discovery, molecular docking, and single-cell RNA sequencing. Each skill is production-ready and tuned to work with libraries like RDKit, Scanpy, and PyTorch Lightning, with unified access to 100+ databases including PubChem, ChEMBL, and ClinicalTrials.gov. The repo works with Claude Code, Cursor, and Codex, and gained 673 stars today — a strong signal that demand for domain-specific agent tooling extends well beyond software development.
github.com/K-Dense-AI/scientific-agent-skills
CLI-Anything Automates the Path from GUI Software to Agent-Ready Interfaces
The HKUDS research group released CLI-Anything, a framework that analyzes the source code of existing applications and automatically generates structured command-line interfaces that AI agents can control directly. Rather than relying on fragile UI automation or incomplete APIs, the system runs a seven-phase pipeline producing Click-based CLIs with comprehensive test suites, JSON output modes, and full documentation — for applications like GIMP, Blender, and LibreOffice. With 2,280+ tests and integrations for Claude Code and OpenCode, the project puts a concrete engineering path behind the idea that any software with accessible source should eventually be agent-native.
Dograh Brings Open-Source Competition to Voice Agent Platforms
Dograh, built by YC alumni, is a self-hostable voice agent platform aimed at teams that want production voice assistants without the pricing opacity or vendor lock-in of managed services like Vapi and Retell. It supports Twilio, Vonage, and Telnyx for telephony, allows custom TTS and STT model integration, and ships with Docker and a drag-and-drop workflow builder that can get a working bot running in about two minutes. The repo gained 287 stars today on a pitch centered on full source transparency, flexible model choices, and complete data residency control.
Bun v1.3.14 Ships as the Runtime Holds Steady Near 92k Stars
The Bun JavaScript runtime released version 1.3.14 on May 13, with contributions from 11 developers, continuing its cadence of frequent, focused updates. Bun's all-in-one positioning — runtime, bundler, test runner, and package manager in a single binary — keeps it trending consistently on GitHub even between major feature releases. The project sits at just under 92,000 total stars and added 397 more today, reflecting sustained developer curiosity rather than a one-time spike.