{"slug":"yolo","title":"yolo","summary":"Use for ANY task involving Ultralytics Platform, the ultralytics Python package, yolo CLI, YOLO model weights (.pt), dataset annotation, training, validation, prediction, tracking, export, deployment, or the detect / segment / semantic / depth / classify / pose / OBB vision tasks","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-10-22T21:02:44.70075Z","repo":{"url":"https://github.com/ultralytics/skills","stars":38,"forks":0,"license":"AGPL-3.0","updatedAt":"2026-11-08T15:04:07Z"},"bodyHtml":"<hr>\n<h2>name: yolo\ndescription: &gt;\nUse for ANY task involving Ultralytics Platform, the ultralytics Python package, yolo CLI, YOLO model weights (.pt), dataset annotation, training, validation, prediction, tracking, export, deployment, or the detect / segment / semantic / depth / classify / pose / OBB vision tasks.</h2>\n<h1>Ultralytics YOLO</h1>\n<p>Run the same lifecycle on three complementary surfaces:</p>\n<ul>\n<li><strong><a href=\"https://platform.ultralytics.com\">Ultralytics Platform</a></strong> — the fastest start: upload or clone data, annotate in the browser, train on cloud GPUs, inspect metrics, test predictions, export, and deploy a dedicated endpoint without local setup.</li>\n<li><strong><code>ultralytics</code> package / <code>yolo</code> CLI</strong> — use local or remote compute, scripts, notebooks, custom pipelines, and exported artifacts directly.</li>\n<li><strong><code>ul</code> CLI</strong> (Python 3.11+, installed by <code>ultralytics</code> or <code>ultralytics-platform</code>) — script the Platform API itself: <code>ul cloud &lt;resource&gt; &lt;operation&gt; key=value</code> lists, creates, clones, trains, exports, and deploys Platform resources from a terminal (see <code>platform-cli</code>).</li>\n</ul>\n<p><code>ul cloud train|predict|export</code> upload local inputs as needed and run on Platform. Training returns after submission; add <code>watch</code> to follow and download results. <code>ul cloud download</code> uses the printed model URI and waits if needed; Ctrl-C stops waiting without canceling training. All four shortcuts need <code>ultralytics</code> installed; the <code>ul cloud &lt;resource&gt; &lt;operation&gt;</code> commands do not.</p>\n<pre><code>ul cloud train model=yolo26n.pt data=ul://username/datasets/helmets epochs=100 project=helmets name=exp1 # → run URI and download command\nul cloud download model=ul://username/helmets/exp1                                                       # → weights/best.pt and results\nul cloud predict model=ul://username/helmets/exp1 source=video.mp4                                       # → annotated output\nul cloud export model=ul://username/helmets/exp1 format=onnx                                             # → downloaded artifact\n</code></pre>\n<p>Mix them freely. Set <code>ULTRALYTICS_API_KEY</code>, use a Platform dataset as <code>data=ul://username/datasets/dataset-slug</code>, and set <code>project=username/project-slug name=experiment</code> during local training to stream its metrics back to Platform.</p>\n<p>The <code>yolo</code> CLI and Python share one API. The CLI grammar is <code>yolo TASK MODE arg=value ...</code>, and Python takes the same argument names:</p>\n<pre><code>yolo detect train data=data.yaml model=yolo26n.pt epochs=100 imgsz=640\n</code></pre>\n<pre><code>from ultralytics import YOLO\n\nmodel = YOLO(\"yolo26n.pt\")\nmodel.train(data=\"data.yaml\", epochs=100, imgsz=640)\n</code></pre>\n<ul>\n<li>TASK ∈ <code>detect</code> <code>segment</code> <code>semantic</code> <code>depth</code> <code>classify</code> <code>pose</code> <code>obb</code> — usually inferred from the weights, so it can be omitted.</li>\n<li>MODE ∈ <code>train</code> <code>val</code> <code>predict</code> <code>track</code> <code>export</code> <code>benchmark</code>.</li>\n<li>Install/upgrade: <code>pip install -U ultralytics</code>. Environment check: <code>yolo checks</code>.</li>\n</ul>\n<h2>Whole lifecycle in five commands</h2>\n<pre><code>yolo detect train data=data.yaml model=yolo26n.pt epochs=100 # → runs/detect/train/weights/best.pt\nyolo val model=best.pt data=data.yaml                        # mAP, per-class metrics\nyolo predict model=best.pt source=video.mp4 save=True        # any source: image/dir/URL/RTSP/webcam\nyolo track model=best.pt source=video.mp4                    # + persistent object IDs\nyolo export model=best.pt format=onnx                        # exported model loads back into YOLO()\n</code></pre>\n<h2>Whole lifecycle in Platform</h2>\n<ol>\n<li>Open <a href=\"https://platform.ultralytics.com\">Platform</a> and choose the data region during onboarding.</li>\n<li>Clone a public dataset from <strong>Explore</strong>, or create one under <strong>Annotate</strong> and upload images, videos, an archive, or NDJSON.</li>\n<li>Label in the fullscreen editor; use SAM or a compatible YOLO model in <strong>Smart</strong> mode where available.</li>\n<li>Create a project, click <strong>New Model</strong>, select the dataset, pretrained model, GPU, and epochs, then monitor the run.</li>\n<li>Use the completed model's <strong>Predict</strong>, <strong>Export</strong>, or <strong>Deploy</strong> tab.</li>\n</ol>\n<p>Start with the <a href=\"https://docs.ultralytics.com/platform/quickstart\">Platform quickstart</a>. Use the stage skill below for both Platform and package details.</p>\n<h2>Route before coding</h2>\n<p>Read the skill for the stage you're working on BEFORE writing code — each is a sibling folder of this one (<code>../&lt;skill&gt;/SKILL.md</code>) and contains exact formats, argument tables with defaults, recipes, and symptom→fix tables. A request spanning stages (\"train and deploy\") → read each relevant skill.</p>\n<table>\n<thead>\n<tr>\n<th>Working on</th>\n<th>Skill</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>choosing a model family/size/task, YOLO26 vs YOLO11, YOLO-World/YOLOE, SAM, RT-DETR</td>\n<td><code>yolo-models</code></td>\n</tr>\n<tr>\n<td>data.yaml, labels, annotation conversion, auto-labeling, dataset analysis/errors, splits</td>\n<td><code>yolo-datasets</code></td>\n</tr>\n<tr>\n<td>training, fine-tuning, hyperparameters, augmentation, OOM / NaN / low mAP, reading runs</td>\n<td><code>yolo-training</code></td>\n</tr>\n<tr>\n<td>hyperparameter tuning, Ray Tune, systematic model improvement, \"autotraining\"</td>\n<td><code>yolo-tuning</code></td>\n</tr>\n<tr>\n<td>predict on images/video/streams, Results API, tracking IDs, counting/heatmaps/Solutions</td>\n<td><code>yolo-inference</code></td>\n</tr>\n<tr>\n<td>ONNX / TensorRT / CoreML / Core AI / OpenVINO / LiteRT / NCNN / NPUs, quantization, benchmarking</td>\n<td><code>yolo-export</code></td>\n</tr>\n<tr>\n<td>Platform API from a terminal: <code>ul cloud &lt;resource&gt; &lt;operation&gt;</code> commands, <code>ultralytics-platform</code>, scripted changes to resources, trash, deployments, and cloud runs</td>\n<td><code>platform-cli</code></td>\n</tr>\n</tbody>\n</table>\n<h2>CLI specifics</h2>\n<p>Special commands (no TASK/MODE):</p>\n<pre><code>yolo help   # full syntax reference\nyolo checks # env report: version, torch, CUDA, disk — run when anything is weird\nyolo version\nyolo settings # view; `yolo settings key=value` to set; `yolo settings reset`\n# keys incl. datasets_dir, runs_dir, wandb, mlflow, tensorboard, ...\nyolo cfg            # print every default argument (the ground truth for arg names)\nyolo copy-cfg       # copy default.yaml → default_copy.yaml to customize, use with cfg=\nyolo solutions help # prebuilt apps: count, heatmap, speed, ... (see yolo-inference)\n</code></pre>\n<p>Parsing rules that matter:</p>\n<ul>\n<li>Args are <code>key=value</code>, no <code>--</code> flags. A leading <code>--</code> and trailing commas are stripped with a warning; spaces around <code>=</code> are merged.</li>\n<li>A bare boolean arg sets it True: <code>yolo predict ... show</code> ≡ <code>show=True</code>.</li>\n<li><code>cfg=custom.yaml</code> loads the file's values: CLI arguments before or after it win, and missing keys still use built-in defaults (start with <code>yolo copy-cfg</code>).</li>\n<li>Missing args are auto-filled with warnings (sample source, task-default data/model, <code>format=torchscript</code>).</li>\n<li>Model stem selects the architecture: <code>rtdetr-*</code> → RT-DETR, <code>sam_*</code>/<code>sam2*</code>/<code>sam3*</code>/<code>mobile_sam</code> → SAM, <code>FastSAM-*</code> → FastSAM, <code>yolo_nas_*</code> → NAS, <code>yoloe-*</code>/<code>*-world*</code> → promptable YOLO (accepts <code>classes=\"person, bus\"</code>), everything else → YOLO.</li>\n</ul>\n<h2>Global directives</h2>\n<ol>\n<li><strong>Validate the dataset before training</strong> — run the task-appropriate checks in <code>yolo-datasets</code>, then a 1-epoch smoke test and inspect <code>runs/&lt;task&gt;/train/train_batch0.jpg</code>: annotations or targets must match each image.</li>\n<li><strong>Always fine-tune from pretrained <code>.pt</code></strong> — never <code>pretrained=False</code>, never a YAML architecture from scratch, unless the user is explicitly doing research.</li>\n<li><strong><code>stream=True</code> for videos/streams</strong> in Python predict/track — the default list mode OOMs on long videos.</li>\n<li><strong>Use <code>best.pt</code></strong> (not <code>last.pt</code>) from <code>runs/&lt;task&gt;/&lt;name&gt;/weights/</code> after training.</li>\n<li><strong>After export, verify parity</strong>: <code>yolo val</code> the exported artifact against the <code>.pt</code> baseline.</li>\n<li><strong>Prefer built-ins over custom code</strong>: dataset converters and checkers (<code>ultralytics.data</code>), trackers, and Solutions modules replace whole categories of hand-written glue.</li>\n<li><strong>Trust the installed version over memory</strong> — if an argument is rejected, the API has moved: <code>yolo cfg</code> and the error text list valid arguments, and <code>yolo checks</code> shows the version. Prefer them over any table in these skills.</li>\n</ol>\n","files":[{"path":"agents/openai.yaml","sizeBytes":249,"isText":true},{"path":"SKILL.md","sizeBytes":9123,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-11-12T21:44:09.498669Z","sha256":"0ADBBA887253ECEFD8E5305AA1A8F0B3708F1A343263F3C857112AE456DEE5A8","sizeBytes":4170},"review":null,"source":{"repositoryUrl":"https://github.com/ultralytics/skills","path":"skills/yolo","license":"AGPL-3.0","commit":"11e45578024f8bcfdcf1ff3faf2d9e8a80e3c919","subtreeSha":"70EDBD9B7999C2E3729D9EEF5DB04A7954946CCEC1FE0E4FB11F9C308122AE9D","lastSyncedAt":"2026-11-12T21:43:57.682266Z"},"reviewedAt":"2026-11-12T21:44:13.829498Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/ultralytics/skills/tree/main/skills/yolo"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install ultralytics-skills@llmmart"},{"target":"git","command":"git clone https://github.com/ultralytics/skills.git"}]}