addyosmani/agent-skills Hits 30k Stars: The Skills Spec That Just Won
Addy Osmani's agent-skills repo crossed 30,300 stars today and added 629 in the last 24 hours. Started a few weeks ago as a viral blog post about "production-grade engineering skills for AI coding agents," it's now a full Markdown-defined skills package. Twenty skills across six lifecycle phases — Define, Plan, Build, Verify, Review, Ship — plus three specialist personas (code reviewer, test engineer, security auditor) and reference checklists for testing, security, performance, accessibility.
The reason this matters editorially is the standardization moment. Three weeks ago, Anthropic shipped Skills as a first-class primitive. mattpocock and browserbase then shipped their own packs. Then ARIS today defined skills as research-tool primitives. addyosmani/agent-skills is the practitioner-facing canonical version — it's plain Markdown, integrates with Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro IDE. The version number says v0.6.0 but the install footprint says de facto standard.
Substance check: the skills aren't generic. They borrow heavily from "Software Engineering at Google" and Google's internal engineering culture, then encode the workflows step-by-step so an agent following the skill reproduces the senior-engineer behavior. The implicit thesis is that AI coding agents default to the shortest path — which means skipping specs, tests, security reviews — and skills are the lever that forces them to do the work properly.
If you're shipping agent-driven engineering at any scale, this is now table stakes. The new game is whether enterprise teams write their own internal skill packs that wrap their codebase conventions, or whether the open-source canon (Osmani + Anthropic + browserbase) gets adopted whole. Either way, hand-rolled prompts as the agent-quality control mechanism are over.
Repo: https://github.com/addyosmani/agent-skills
← Back to all articles
The reason this matters editorially is the standardization moment. Three weeks ago, Anthropic shipped Skills as a first-class primitive. mattpocock and browserbase then shipped their own packs. Then ARIS today defined skills as research-tool primitives. addyosmani/agent-skills is the practitioner-facing canonical version — it's plain Markdown, integrates with Claude Code, Cursor, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro IDE. The version number says v0.6.0 but the install footprint says de facto standard.
Substance check: the skills aren't generic. They borrow heavily from "Software Engineering at Google" and Google's internal engineering culture, then encode the workflows step-by-step so an agent following the skill reproduces the senior-engineer behavior. The implicit thesis is that AI coding agents default to the shortest path — which means skipping specs, tests, security reviews — and skills are the lever that forces them to do the work properly.
If you're shipping agent-driven engineering at any scale, this is now table stakes. The new game is whether enterprise teams write their own internal skill packs that wrap their codebase conventions, or whether the open-source canon (Osmani + Anthropic + browserbase) gets adopted whole. Either way, hand-rolled prompts as the agent-quality control mechanism are over.
Repo: https://github.com/addyosmani/agent-skills
Comments