fiftyone-skills

Use when driving the FiftyOne App via the Playwright MCP — plugin/operator verification, demo or screencast recording, or any end-to-end UI automation against `fo.launch_app(...)`. Covers the critical "do not navigate after reload_dataset" crash, launcher patterns (remote=True, trigger-file IPC), MUI-specific input/dropdown gotchas, sidebar tag filtering, `data-cy` selectors, dialog scrolling, session refresh strategies, and cleanup. Trigger on mentions of "Playwright + FiftyOne", "automate FiftyOne App", "operator demo", "browser_navigate crash", "reload_dataset", or any FiftyOne session that silently dies after a UI action.

Writes Python code following FiftyOne's official conventions. Use when contributing to FiftyOne, developing plugins, or writing code that integrates with FiftyOne's codebase.

Creates Jupyter notebooks for FiftyOne workflows including getting-started guides, tutorials, recipes, and full ML pipelines. Use when creating notebooks, writing tutorials, building demos, or generating FiftyOne walkthroughs covering data loading, exploration, inference, evaluation, and export.

End-to-end dataset curation for FiftyOne: inspect schema and quality, audit annotations, analyze class distributions, explore embeddings, find duplicates, create curated subsets, and build train/val/test splits. Works with any computer vision dataset type.

Exports FiftyOne datasets to standard formats (COCO, YOLO, VOC, CVAT, CSV, etc.) and Hugging Face Hub. Use when converting datasets, exporting for training, creating archives, sharing data in specific formats, or publishing datasets to Hugging Face.

Imports datasets into FiftyOne with automatic format detection. Supports all media types (images, videos, point clouds, MCAP multimodal recordings), label formats (COCO, YOLO, VOC, KITTI), multimodal grouped datasets, LeRobot v3 robot-learning episode datasets, and Hugging Face Hub datasets. Use when importing datasets from local files or Hugging Face, loading autonomous driving data, importing robotics/AV sensor logs (MCAP, ROS bag), importing LeRobot robot-learning episodes (teleop recordings, observation.state/action data), or creating grouped datasets.

Run ML model inference on FiftyOne datasets. Use when running models for detection, classification, segmentation, or embeddings. Discovers available models dynamically from the Zoo, plugin operators, or custom sources — never assumes a fixed model list.

Develops custom FiftyOne plugins (operators and panels) from scratch. Use when creating plugins, extending FiftyOne with custom operators, building interactive panels, or integrating external APIs.

Visualizes datasets in 2D using embeddings with UMAP or t-SNE dimensionality reduction. Use when exploring dataset structure, finding clusters, identifying outliers, or understanding data distribution.

Evaluates FiftyOne plugins for quality, security, and agent-readiness. Use when reviewing a plugin before installation, auditing an existing plugin, validating a plugin you just built, or assessing community plugins for safety. Produces a structured report with scores and actionable recommendations.

Finds duplicate or near-duplicate images in FiftyOne datasets using brain similarity computation. Use when deduplicating datasets, finding similar images, or removing redundant samples.

Generate a Data Lens connector from an external database schema. Use when users want to connect an external data source (PostgreSQL, BigQuery, Databricks, MySQL, etc.) to FiftyOne Data Lens, or when they have a database schema and want to browse/import that data through the FiftyOne App.

Triages FiftyOne GitHub issues by categorizing as fixed, won't fix, not reproducible, or still valid. Use when reviewing GitHub issues, triaging bugs, or closing stale issues in the voxel51/fiftyone repository.

Evaluate model predictions against ground truth using COCO, Open Images, or custom protocols. Use when computing mAP, precision, recall, confusion matrices, or analyzing TP/FP/FN examples for detection, classification, segmentation, or regression tasks.

Answers FiftyOne Python SDK questions with accurate, runnable code. Use when the user asks how to do something in the FiftyOne Python SDK, asks a docs question, no skill or operator covers their goal, or the agent is unsure of the correct FiftyOne method, argument, or field path while writing code — e.g. "how do I do X in Python", "show me the SDK for filtering", "there's no operator for this, can I do it in code", "what's the API for embeddings", "write me a Python script to do Y".

Diagnose and fix common FiftyOne issues automatically. Use when a dataset disappeared, the App won't open, changes aren't saving, MongoDB errors occur, video codecs fail, notebook connectivity breaks, operators are missing, or any recurring FiftyOne pain point needs solving.

Build FiftyOne UIs using VOODO (@voxel51/voodo), the official React component library. Use when building plugin panels, creating interactive UIs, styling FiftyOne applications, or choosing components, text sizes, colors, and spacing. Looks up the installed version's components and token values with `npx @voxel51/voodo` before writing code.

Use when integrating a model into FiftyOne's remote model zoo — detection, classification, segmentation, embedding, keypoint, or vision-language (VLM) models loaded via `register_zoo_model_source` and `load_zoo_model`, then applied with `dataset.apply_model`. Also for debugging zoo registration, `manifest.json` issues, custom `fom.Model` / `TorchModelMixin` subclasses, DataLoader pickle errors, or worker `ModuleNotFoundError` from spawned DataLoader workers.