What are Ultralytics Agent Skills?

They are reusable instructions and reference material that help AI coding agents work with Ultralytics Platform, the ultralytics Python package, the yolo CLI, and the ul Platform CLI.

Which agents can use these skills?

The repository provides plugins for Claude Code and Codex. Cursor, Gemini CLI, and other agents that read the open Agent Skills format can install the skill folders directly or through the skills CLI.

Which skill should I install?

Install the complete plugin for automatic routing across the computer vision lifecycle and Platform CLI operations. To install one skill with the skills CLI, pass its name with --skill, such as --skill yolo-training or --skill platform-cli.

Do the skills run code or send data?

The plugin itself contains only instructions, so nothing runs when it loads. Commands your agent runs from the skills can download packages, weights, and datasets, and ul cloud commands upload the files you pass them to Ultralytics Platform with your API key. The ultralytics package also sends anonymous usage analytics, which yolo settings sync=False turns off.

Do the skills replace the Ultralytics documentation?

No. The skills guide an agent toward the relevant workflow, but the installed package and current Ultralytics documentation remain authoritative when behavior or supported options differ.

How do I report incorrect guidance?

Open a bug report in the ultralytics/skills repository with the prompt you used, the guidance the agent gave, and what should have happened.

Ultralytics Agent Skills#

Ultralytics Agent Skills give compatible AI coding agents instructions and reference material for working with the ultralytics Python package, the yolo CLI, and Ultralytics Platform, including automation with ul cloud. They follow the open Agent Skills format and load when a relevant task is requested.

How the Skills Work#

The yolo router skill maps a request to the lifecycle stage it involves, and each stage skill covers both routes: the fastest path in Ultralytics Platform, then the local Python and CLI workflow. Version-sensitive facts such as weight names, training arguments, and export formats live in catalogs grounded against a pinned ultralytics release, and every skill tells the agent to trust the installed package (yolo checks, yolo cfg, and error messages) when they differ. The platform-cli skill instead defers to the installed ul CLI help and the live Platform API. Platform's own Ask AI uses the same platform-cli skill.

Available Skills#

The repository contains one router, six lifecycle skills, and one Platform CLI skill:

Skill	Use for
`yolo`	Core Python and CLI usage, Platform workflows, and routing to the other skills
`yolo-models`	Selecting model families, sizes, tasks, and weights
`yolo-datasets`	Preparing, converting, validating, and troubleshooting datasets
`yolo-training`	Training, validation, resumes, multi-GPU use, and troubleshooting
`yolo-tuning`	Hyperparameter tuning and experiment improvement
`yolo-inference`	Prediction, results, tracking, and Solutions
`yolo-export`	Export, quantization, deployment formats, and benchmarking
`platform-cli`	Automating Ultralytics Platform with `ul cloud`: resources, uploads, cloud training, exports, and deployments

Installation#

claude plugin marketplace add ultralytics/skills
claude plugin install yolo@ultralytics

To update, run claude plugin update yolo@ultralytics and restart Claude Code.

Each version of the plugin is published as a GitHub release with release notes.

Connect the Platform CLI#

The platform-cli skill uses the ul CLI (Python 3.11+), included with pip install ultralytics or the standalone SDK (pip install ultralytics-platform). Create a Platform API key, make it available to your agent, and check the connection:

export ULTRALYTICS_API_KEY="YOUR_API_KEY" # or: ul login YOUR_API_KEY
ul cloud account summary

Example Prompts#

Once installed, ask the agent naturally. The yolo router selects the relevant guidance:

Goal	Example prompt
Try YOLO	"Run YOLO on a sample image and show me what it detects."
Train	"Turn my images into a custom YOLO model, labeled and trained in Ultralytics Platform."
Improve	"My model's mAP plateaued. Diagnose the run and propose the next experiment."
Deploy	"Export my trained YOLO model to TensorRT with INT8 quantization and benchmark it."
Analyze video	"Count the cars crossing a line in traffic.mp4 and save the annotated video."
Automate Platform	"List my Ultralytics Platform datasets and start cloud training on the PPE one."

See the ultralytics/skills repository for current installation commands, source files, updates, and issue reporting.

FAQ#

They are reusable instructions and reference material that help AI coding agents work with Ultralytics Platform, the ultralytics Python package, the yolo CLI, and the ul Platform CLI.
The repository provides plugins for Claude Code and Codex. Cursor, Gemini CLI, and other agents that read the open Agent Skills format can install the skill folders directly or through the skills CLI.
Install the complete plugin for automatic routing across the computer vision lifecycle and Platform CLI operations. To install one skill with the skills CLI, pass its name with --skill, such as --skill yolo-training or --skill platform-cli.
The plugin itself contains only instructions, so nothing runs when it loads. Commands your agent runs from the skills can download packages, weights, and datasets, and ul cloud commands upload the files you pass them to Ultralytics Platform with your API key. The ultralytics package also sends anonymous usage analytics, which yolo settings sync=False turns off.
No. The skills guide an agent toward the relevant workflow, but the installed package and current Ultralytics documentation remain authoritative when behavior or supported options differ.
Open a bug report in the ultralytics/skills repository with the prompt you used, the guidance the agent gave, and what should have happened.

Contributors

JAJaviChulvi² GLglenn-jocher¹ PDpderrenger¹

Created 1 month agoUpdated 2 hours ago