Important new course: Agent Skills with Anthropic, built with Anthropic and taught by Elie Schoppik! Skills are constructed as folders of instructions that equip agents with on-demand knowledge and workflows. This short course teaches you how to create them following best practices. Because skills follow an open standard format,... you can build them once and deploy across any skills-compatible agent, like Claude Code. What you'll learn: - Create custom skills for code generation and review, data analysis, and research - Build complex workflows using Anthropic's pre-built skills (Excel, PowerPoint, skill creation) and custom skills - Combine skills with MCP and subagents to create agentic systems with specialized knowledge - Deploy the same skills across Claude Code, the Claude API, and the Claude Agent SDK Join and learn to equip agents with the specialized knowledge they need for reliable, repeatable workflows.show more

@AnthropicAI @eschoppik we really went from 'the code is the documentation' to 'the documentation is the code' in record time. folders of instructions are basically just READMEs with a run button

I am definitely going to check it out! This basically confirms what a lot of developers have been thinking for a while now. You can't just rely on prompts if you want things to scale, you need actual structured skills that work consistently. Having a standard format you can reuse across different agents finally brings some real discipline to how we build these things. This was a genuine gap in the ecosystem, and it's good to see it getting filled.

The shift toward modular "folders of instructions" feels like a necessary evolution from standard prompting. I’m particularly interested in the "write once, deploy anywhere" architecture Elie Schoppik covers. The idea that a single file can package a workflow—like that marketing campaign analysis example—and run seamlessly across Claude Code, the API, or the SDK is a huge efficiency unlock. It effectively treats agent behaviors like portable software libraries rather than ephemeral chat context. Are there plans to expand the library of pre-built skills beyond the Excel and PowerPoint ones?

@AnthropicAI @eschoppik Andrew Ng is someone you can trust. He consistently and honestly works on AI.

@AnthropicAI @eschoppik The man is teaching everything at a pace where we can't even learn it all!

@AnthropicAI @eschoppik thank you for everything you do for the AI/ML community at large, Andrew.

@AnthropicAI @eschoppik Claude gets the skills, now let’s see if it learns the hustle.

@planert41 @AnthropicAI @eschoppik Got some labs u can play with here plus all kinds of learning material to help u build your own plus plenty if free ones that could work great as base templates for u to customize Over 1400 ⭐️ on @github

@AnthropicAI @eschoppik Working on something to build skills, optimize skills, have skills trigger Subagents (vs agents ingest skills)

@AnthropicAI @eschoppik Folders of instructions formalize the modularity agents need to scale. The pattern: this decouples static prompts from dynamic workflows, turning LLMs into composable systems. Enterprise agents hit escape velocity when skills become plug-and-play.

Prompt Engineering is dead. Skill Engineering just replaced it. Think about the difference. A prompt is a one-time instruction. A "Skill" is a reusable asset. Anthropic just turned the context window into an operating system. You aren't just chatting anymore. You are installing capabilities. The "Open Standard" is the most important part. It means you build the logic once. And it works in the chat, in the terminal, and in the API. You are no longer writing text. You are programming behavior.

@AnthropicAI @eschoppik Skills as portable folders is a good abstraction for domain knowledge. It makes agent behaviour reproducible across tools. But the ecosystem is still immature - every agent has its own install path. We are getting further and further into symlink hell.

@AnthropicAI @eschoppik This is gold for anyone building AI agents. Practical skills > theory. Thanks for sharing! 🚀🤖

@AnthropicAI @eschoppik The folder-based skill architecture is a smart design pattern - modular, reusable, and easy to version control. This approach mirrors how we organize codebases. Excited to see how skills can be composed and chained for complex multi-step reasoning tasks.

@AnthropicAI @eschoppik Great stuff! I too think Modular, reusable skills with open standards is exactly the right direction for scalable agents. Nice work.

@AnthropicAI @eschoppik Skills as reusable building blocks is how agents finally become dependable system.

@AnthropicAI @eschoppik @andrewyng, you’ve done it again!! so much good content on building with skills and claude code and its FREE don’t get left behind

@AnthropicAI @eschoppik The folder metaphor reveals the real insight. Agents need modular knowledge architectures we can compose and recompose. Most implementations fail because they treat skills as features instead of living systems.

@AnthropicAI @eschoppik folders? we're literally teaching AI agents how to use file systems now. what's next, teaching them JIRA shortcuts

@AnthropicAI @eschoppik so grateful for your original AI course and so happy to see you continue making incredible educational content @AndrewYNg

@AnthropicAI @eschoppik “AI adoption feels less like a tech problem now and more like an organizational one.”

What's interesting here is the shift from prompting to knowledge engineering. That 206% improvement from codified expertise is real, and it's exactly why Agent Skills matters: it makes expertise portable instead of locked in someone's head. We stop crafting individual prompts and start building reusable knowledge structures.

@AnthropicAI @eschoppik Great move toward reusable automation patterns for AI agents. “Skills as folders of instructions” makes it easier to standardize workflows (code review, research, analysis) and deploy them consistently across tools. Skills + MCP should improve reliability and productivity.

@AnthropicAI @eschoppik Skills as modular instruction folders is such an elegant pattern. Makes agents composable and debuggable - two things that are usually hard to get right together 🧠

@AnthropicAI @eschoppik The best explanation that I’ve heard for skills so far, thanks

Great step toward modular agents, but skills alone are not intelligence. What will matter is congruity between instructions, context limits, compute cost, and real world feedback. Reusable skills scale only if they remain proportionate to task complexity. The winners will be those who design agent skills that know when not to act.

@AnthropicAI @eschoppik Sounds cool! How easy is it for someone new to start creating these skills?

@AnthropicAI @eschoppik love this partnership!

@AnthropicAI @eschoppik Skills boost autonomy, but without control layers they amplify risk faster than value.

@AnthropicAI @eschoppik I just wrap up a session withClaude and ask to create a skill based on our work.

@AnthropicAI @eschoppik He’s such a good teacher!

The integration with MCP (Model Context Protocol) is the standout feature here for me. Most of my automation bottlenecks happen during the data handoff between tools. If we can standardize how agents 'call' on these Excel/PowerPoint skills, we stop building chatbots and start building actual reliable systems. Great timing for this.

@AnthropicAI @eschoppik Anthropic just launched a new course on Agent Skills! 🤖 Build reusable "folders of instructions" to give agents on-demand knowledge. ​✅ Key takeaway: Build once, deploy across Claude Code & API. ✅ Focus: MCP integration, subagents, and automated workflows. 🚀

Skills are the missing piece most people don't get about AI agents. Agents without skills = smart intern with no training manual Agents with skills = senior engineer who knows exactly what to do The shift: from "prompt engineering" to "skill engineering." Biggest unlock? Skills are composable. Build once, reuse across agents.

@AnthropicAI @eschoppik @rattibha unroll