Hey everyone! Maker here. Thanks for checking out askill. Here's what triggered this project: AI coding agents like Claude Code, Codex, and Openclaw are transforming how we write software. A growing ecosystem of "skills" — instruction files that teach agents specific workflows — is emerging across GitHub. But we noticed a critical gap: there's no trust layer. Think about it. A SKILL.md file literally tells your AI agent what to do and how to do it. It shapes your agent's behavior in your codebase. Would you install one from a stranger's repo without checking it first? Most people do — because there's no practical alternative. That's the problem askill solves. We built an AI-powered quality and safety layer for agent skills. Every skill on askill.sh is automatically reviewed by AI across 5 dimensions: • Safety — flags hardcoded secrets, dangerous commands, destructive operations • Clarity — is the skill well-documented and structured? • Reusability — does it work across projects, or is it repo-specific? • Completeness — does it actually cover what it claims? • Actionability — are the instructions concrete and executable? Scores, detailed breakdowns, and AI-generated review summaries are all visible before you install. Skills get re-evaluated automatically whenever their content changes on GitHub. Think of it as npm audit meets code review — but for agent skills, powered by AI. On top of that, askill is a proper package manager: • Dependency management — skills can depend on other skills • Semantic versioning — lock versions, manage updates safely • One command (askill add) installs across 40+ AI agents • Open ecosystem — we auto-index SKILL.md files from public GitHub repos We believe the agent skills ecosystem needs what every healthy open-source ecosystem has: a way to surface quality and filter out risk, at scale. Humans can't manually review thousands of skills. AI can — and that's exactly what askill does. Open source, free to use. We'd love to hear: what would make you trust (or distrust) an AI agent skill? And what skills would you want for your workflow?

askill.sh

Every Agent Skill, One Registry

Every Agent Skill, One Registry