代理技能革新了與AI的互動與工作流程

代理技能革新了與AI的互動與工作流程

Hacker News·

Eleanor Berger 探討了「代理技能」如何讓使用者將其領域專業知識注入AI代理,實現個人化且更有效率的工作流程。這些可重複使用的指令和資源,適用於各種AI平台,讓使用者無需程式設計知識即可客製化AI的能力。

Image

Eleanor on Everything

Agent Skills Changed How I Work with AI

A new way of working with AI that I'm genuinely excited about

Ready to make Agent Skills your AI superpower? 💪

👉 Join me for Mastering Agent Skills, a live workshop for AI power users.

I’ve been working with Agent Skills for a few months now, and I’m genuinely thrilled with what they’ve made possible.

Skills are reusable instructions and resources you give to an AI agent to make it better at specific tasks. Think of them as training materials: you show the agent how you want something done, what resources to use, what your preferences are, and it remembers. Every time you need that capability again, the agent just knows.

And you don’t need to be a programmer to create them.

What Are Agent Skills, Really?

Agent Skills started in Claude but have since spread to other AI agents — Claude Code, GitHub Copilot, OpenAI’s Codex, OpenCode, and more. There’s now an open standard for skills that works across many agents.

At their simplest, a skill is just a markdown file with instructions. At their most powerful, skills include templates, reference documents, and scripts that let the agent do complex work on your behalf.

The point is: skills let me inject my domain expertise into the agent. I’m not limited to what the AI was trained on. I can teach it how I do things, with my templates, following my standards.

Skills I’ve Built

To give you a sense of what’s possible, here are some skills I’ve created over the past few months:

Flashcard Generator — Converts any text into Anki flashcard decks. I wrote about this one in more detail at Are Skills the New Apps? because it demonstrates something interesting: skills can replace standalone applications entirely.

Promptify — Transforms casual, dictated requests into well-structured prompts. Useful when I’m speaking to the agent rather than typing.

Markdown Converter — Uses markitdown to convert PDFs, Word documents, PowerPoint, Excel, images, and audio files to Markdown. Replaces a whole MCP server with a simple skill.

GitHub Copilot CLI — Lets me access Gemini, GPT-5, and Codex from within Claude. Useful for getting a second opinion from other models without switching contexts.

Nano Banana Pro — Generates and edits images using Gemini 3 Pro Image at up to 4K resolution.

GPT Image 1.5 — Similar image generation using OpenAI’s model. Having both means I can pick the one that suits the task.

YouTube Transcript — Fetches transcripts from YouTube videos. (Note: this works in Claude Code but not the Claude web app due to network restrictions.)

Invoice Generator — Generates PDF invoices from a Word template. I built this as a teaching example, but it demonstrates the pattern of document generation that’s useful across many domains.

Upstash Redis KV — Reads and writes to a Redis-compatible key-value store. This one’s more technical, but shows that skills can connect to external services.

Lorem Ipsum Generator — Generates placeholder text in various formats. Simple, but useful.

Some of these are practical tools I use daily. Others are examples for teaching. All of them were built through conversation with the AI — showing it what I wanted, iterating until it worked, then saving the result as a skill.

Why This Matters for AI Power Users

The common assumption is that skills are a developer tool. They’re not.

Yes, some skills include scripts. But you don’t necessarilly need to write those scripts. The AI can do that. Your job is to describe what you want, test whether it works, and refine until it does. If you’re comfortable with Excel formulas or automation tools like Zapier, you have enough technical fluency to create skills.

What you bring to the table is domain expertise. The AI knows how to code; it doesn’t know how your team formats reports, what your invoice template looks like, or which workflow you prefer. That’s your contribution.

How to Get Started with Agent Skills

If you’re curious about skills and want to get started, I’ve put together a free video series:

Image

Introduction to Agent Skills is a mini series of YouTube videos that guide you through some of the most common patters. No programming background required. You’ll learn what skills are, create and use your first skills, and progress to more sophisticated workflows.

And if you want to go deeper — to really master skills and build a personal library that compounds over time — I’m running a live workshop:

Image

Mastering Agent Skills is a 4-hour hands-on workshop. We’ll cover the architecture, work through progressively complex examples, and you’ll leave with skills you built yourself plus all the templates and materials you need to keep going.

It’s small-group and limited places, so have a look if you’re interested.

No posts

Ready for more?

Hacker News

相關文章

  1. 代理技能

    3 個月前

  2. 2026年建構AI代理程式的最簡單方法

    3 個月前

  3. 理解與部署AI代理:一份綜合指南

    3 個月前

  4. 代理技能:用於 React 和 Next.js 工作流程的 AI 代理

    3 個月前

  5. 透過單一CLI指令輕鬆管理AI代理技能

    3 個月前