// Skill profile
addyosmani/agent-skills is a 77K-star bundle of production-grade engineering skills for AI coding agents, MIT-licensed and the highest-starred agent-skills directory on GitHub as of July 2026. Authored by Addy Osmani (Chrome team, formerly Google Web Platform DevRel), the project ships 24 SKILL.md-formatted instruction bundles organized across the full development lifecycle — defining, planning, building, verifying, reviewing, and shipping — with 8 slash commands and support for 70+ AI agents. The bundle loads into Claude Code, Cursor, Anthropic's Antigravity IDE, or any harness that respects the runtime-skills convention. v0.6.3 (July 2026) is the latest release with active development (338 commits, 83 open PRs). Pairs naturally with mattpocock/skills (Claude Code-first) and ComposioHQ/awesome-codex-skills (Codex-first) — the three together effectively cover the agent IDE/CLI landscape.
addyosmani/agent-skills is a curated bundle of production-grade engineering skills for AI coding agents — MIT-licensed, distributed via a shell-script installer, and authored by Addy Osmani, one of the most visible figures in mainstream web developer experience. As of July 2026 the repo has reached 77,000 stars with 8,300 forks, making it the highest-starred agent-skills directory on GitHub by a wide margin. The project ships 24 skills organized across the full development lifecycle with 8 slash commands and support for 70+ AI agents. The repo’s GitHub topics — claude-code, cursor, antigravity, antigravity-ide, skills, agent-skills — cover the four leading agent harnesses simultaneously.
The bundle ships SKILL.md-formatted instruction files that load at runtime into any agent harness following the skills convention: Claude Code, Cursor, Anthropic’s Antigravity IDE, and SKILL.md-aware tools like Codex CLI. Each skill encapsulates a repeatable engineering workflow — patterns refined for production codebases rather than toy examples. Installation is a single shell command that drops the skills into the harness’s expected location. There is no proprietary runtime — every skill is auditable markdown.
When Addy Osmani — a mainstream web-platform DevRel voice — publishes a curated skills bundle and tags it explicitly for antigravity-ide, that’s a legitimacy signal for the SKILL.md format itself. The skills convergence has been visible since late 2025, but as of May 2026 every credible runtime-skills directory ships in the same format and targets the same four harnesses. Vendor-specific skills bundles are no longer competitive against the cross-harness pattern. addyosmani/agent-skills is the highest-profile example of that convergence in mainstream web DX, alongside ComposioHQ/awesome-codex-skills (Codex-side) and mattpocock/skills (Claude Code-side).
Useful for engineering teams adopting any SKILL.md-aware agent harness who want a curated baseline of production-quality skills without authoring them from scratch. Particularly valuable for teams running Antigravity IDE, where the bundled skills have been validated against the newest entrant in the agent-IDE space. Pairs naturally with mattpocock/skills (TypeScript/Claude Code emphasis) and ComposioHQ/awesome-codex-skills (Codex-CLI emphasis) — installing all three together gives an agent harness very strong default capability across the modern coding workflow.
The skills directory is opinionated — Addy’s preferences for engineering style and tooling are embedded in the prompts. Teams with strongly differing conventions should fork and adapt rather than ship the bundle as-is. As an open-source skills directory there is no SLA, hosted runtime, or vendor support; if you need that, the relevant agent harness vendor (Anthropic, Cursor, Google Antigravity) is the support contract — addyosmani/agent-skills is the content, not the infrastructure.
agent-skills fits engineering teams adopting Claude Code, Cursor, Antigravity, or Codex who want a credible, mainstream-DX-validated baseline of skills before authoring their own. It’s also useful for SKILL.md authors studying high-quality production examples — the repo is small enough to read fully and large enough to demonstrate the breadth the format supports. If you’re not yet using a SKILL.md-aware harness, this directory is also one of the cleanest arguments for starting.
AI agents that work well with Agent Skills.
Single 70-line CLAUDE.md derived from Karpathy's LLM coding-pitfall observations — #1 GitHub trending for 28 consecutive days.
100+ ready-to-use Claude Code agents, commands, hooks, and integrations — installable via CLI or interactive dashboard at aitmpl.com.
Injects up-to-date, version-specific documentation into LLM context — eliminates hallucination from stale training data.