AI agents are incredible right up until they confidently generate something completely wrong and move on like nothing happened.
What I Ran Into
Over the past year, I've spent a huge amount of time working with tools like Claude Code, Cursor, and Goose, and one pattern keeps showing up: the code looks correct until you actually run it.
That was one of the first things I ran into after joining Cloudinary.
I asked my agent to help me set up a simple Cloudinary integration in an application I was testing. The prompt itself was straightforward:
generate a transformation URL, optimize an image, resize it to 800px wide, and apply automatic quality and format optimization.
The output looked convincing enough that I almost trusted it. Which, honestly, was the problem. The transformation syntax wasn't following best practices, the SDK imports were outdated, and some delivery URL parameters would have behaved unpredictably in production. It didn't fail loudly. It failed quietly, which is worse.
That’s exactly why we built the Cloudinary Skills Pack.
What the Cloudinary Skills Pack Does
The Cloudinary Skills Pack turns your AI agent into a Cloudinary expert with up to date workflows, documentation context, and implementation patterns built in.
Instead of relying entirely on scattered examples pulled from docs, forums, blog posts, and old repositories, the agent can use skills that understand:
- Cloudinary transformation syntax
- current SDK usage patterns
- delivery URL workflows
- media optimization best practices
- documentation-aware implementation guidance
The current Skills Pack includes three skills:
| Skill | Purpose |
|---|---|
cloudinary-docs |
Pulls relevant documentation context directly from the latest Cloudinary docs |
cloudinary-transformations |
Converts natural language requests into valid Cloudinary image and video transformation URLs |
cloudinary-react |
Provides modern React SDK integration patterns and troubleshooting guidance |
💡 The
cloudinary-docsandcloudinary-transformationsskills are framework agnostic and work regardless of whether you're using React, Next.js, Vue, Svelte, or another stack entirely. The React skill exists specifically for teams using the Cloudinary React SDK.
What Changed When I Tested It
After installing the Skills Pack, I reran the exact same prompts.
The difference was immediate.
Documentation Answers Became More Reliable
One of the biggest improvements was documentation accuracy.
Prompt: "How do I sign a Cloudinary delivery URL to make it secure?"
Without the skill:
With the skill:
Without the skill, my agent explained the concept of signing but missed the actual implementation. With the cloudinary-docs skill installed, it pulled context directly from the latest Cloudinary documentation and returned working Node.js code and a CLI command I could run immediately.
That distinction matters more than it sounds. Incorrect answers around delivery URLs and signing workflows often don't fail immediately. They become production issues later when security constraints start to matter.
Transformation URLs Became Production Ready
The clearest improvement showed up in transformation generation.
Prompt: "Write me a Cloudinary transformation URL that resizes an image to 800px wide, uses face aware cropping, and applies automatic format and quality optimization."
Without the skill:
With the skill:
Without the skill, my agent used c_thumb instead of c_fill for the crop mode. Technically the URL worked. Realistically it was one "why is this image cropped like that?" debugging session away from ruining my afternoon. With the cloudinary-transformations skill installed, it returned the correct crop mode, face aware gravity, and proper optimization parameters on the first attempt.
This is also where the framework agnostic nature of the Skills Pack becomes important. The transformation skill works independently of React and improves Cloudinary workflows no matter what frontend framework or backend stack you are using.
The React SDK Imports Were Correct
One of the workflows I tested used the Cloudinary React SDK, which is where the cloudinary-react skill came in.
Prompt: "Show me how to set up Cloudinary in a React app and display an optimized image using the Cloudinary React SDK."
Without the skill:
With the skill:
Without the skill, my agent recommended the deprecated cloudinary-react package. The code looked believable enough that I probably would have copied it directly into my project if my terminal hadn't immediately humbled me. With the skill installed, it recommended the correct @cloudinary/react and @cloudinary/url-gen packages and set up the right environment variables for Vite.
Instead of debugging the generated integration, I could immediately continue building.
MCP Servers and Skills Solve Different Problems
One thing became very clear while testing this.
There's a growing conversation around whether skills are necessary if you already have MCP servers configured for your coding agent. After spending the last year heavily working with agents and skills, I don't think these are competing ideas at all.
They solve different problems.
| MCP Servers | Skills |
|---|---|
| Give agents tools and actions | Give agents workflows and implementation knowledge |
| Execute operations | Teach best practices and platform specific patterns |
| Help agents do something | Help agents do it correctly |
A Cloudinary MCP server can help an agent create a named transformation in your account.
The skill helps the agent understand:
- when named transformations are the better architectural choice
- how transformation syntax should be structured
- which optimization patterns are recommended
- how to avoid common implementation mistakes
The two approaches complement each other. MCP servers extend what an agent can do, while skills improve how accurately the agent understands a platform.
That distinction clicked for me very quickly because I spent the last few years working on Goose, where skills are a core part of how agents learn reusable workflows. Once I started testing Cloudinary integrations through that same lens, the value became obvious almost immediately.
AI Agents Need Platform Context
The biggest shift happening right now is that AI agents are quickly becoming the interface developers use to interact with platforms.
That changes the responsibility for developer platforms too.
It's no longer enough to expose APIs and publish documentation. Platforms now need ways to teach agents:
- platform specific syntax
- implementation workflows
- SDK patterns
- optimization best practices
- architectural recommendations
That's what skills enable.
Try It
# run in your terminal
npx skills add cloudinary-devs/skills
Full installation options, marketplace plugins, and setup instructions are available in the Cloudinary Skills Guide →
We're continuing to expand the Cloudinary Skills Pack with new skills and capabilities over time, so stay tuned for more. And if you have feedback or ideas you'd like to see supported, we'd love to hear from you in the GitHub repo.
| Cloudinary ❤️ developers |
|---|
| Ready to level up your media workflow? Start using Cloudinary for free and build better visual experiences today. |
| 👉 Create your free account |
Top comments (0)