Today's GitHub trending is a snapshot of a standard solidifying in real time: the agent skills package, a directory of reusable prompts and workflows that give AI coding agents repeatable, engineer-defined behaviors. Three independent skills collections — from TypeScript educator Matt Pocock, Google Chrome's Addy Osmani, and Google's own engineering team — simultaneously broke into the top five trending repositories, suggesting the community has moved past debating whether to use AI coding agents and is now focused on how to make them reliable and shareable.
mattpocock/skills: Engineering Discipline in a Slash Command
Matt Pocock, best known for his TypeScript courses and tooling, published his personal .agents/ skill directory as an open-source package. The eight skills target what he calls the four AI failure modes — misalignment, verbosity, code quality decay, and architecture decay — with slash commands like /grill-with-docs for tightening requirements before any code is written, /tdd for enforcing test-driven cycles, and /code-review for reviewing output against both engineering standards and the original spec. The framing is explicitly anti-vibe coding: these skills are designed to impose professional discipline on agents that would otherwise drift toward plausible-but-wrong outputs.
addyosmani/agent-skills: The Full Development Lifecycle in 24 Skills
Addy Osmani of Google's Chrome team released a 24-skill collection that maps the entire software development lifecycle into eight slash commands: /spec, /plan, /build, /test, /review, /webperf, /code-simplify, and /ship. The package also ships four specialist personas — code reviewer, test engineer, security auditor, and performance auditor — that agents can embody for targeted work. Compatibility is declared across 70+ agents including Claude Code, Cursor, Codex, and GitHub Copilot, positioning this as a model-agnostic engineering standard rather than a tool-specific add-on.
github.com/addyosmani/agent-skills
google/skills: Official Agent Workflows for Google Cloud
Google published an official skills package covering its full cloud and AI product surface — GKE, BigQuery, AlloyDB, Bigtable, Spanner, the Gemini API, RAG engines, model tuning, Cloud Monitoring, and more, structured across the Google Well-Architected Framework's six pillars. The repo is actively maintained with 17 open pull requests and installs via the same npx skills add convention used by the community packages, making Google Cloud infrastructure guidance a first-class citizen in AI-assisted development workflows.
prime-agent holds at #1 for a second straight day
PrimeIntellect's self-improving TypeScript coding agent — which added 2,200 stars on its debut — picked up another 2,400 stars today, staying atop GitHub trending and suggesting the initial surge was more than novelty. A self-improving agent designed for long-running, multi-session workflows is an obvious complement to well-defined skill sets: the skills encode what good engineering looks like, and the agent's reinforcement loop trains toward it. The two trends arriving together in the same week points to a maturing picture of what production-grade AI-assisted development actually requires.