Skills

Trains Visual ChangeNet models using NVIDIA TAO Toolkit for detecting changes between image pairs, applicable to satellite imagery analysis, construction monitoring, and industrial inspection workflows.

Quickly retrieves and summarizes technical documentation across multiple sources to provide precise, context-aware programming references and solutions.

from google/skills

Explores any codebase first (domain, architecture, specs, failure history, optional AI chat-history mining) and then generates six grounded artefacts: `quality/QUALITY.md` constitution, language-appropriate spec-traced functional tests, `RUN_CODE_REVIEW.md` with anti-hallucination guardrails, `RUN_INTEGRATION_TESTS.md`, a Council-of-Three multi-model `RUN_SPEC_AUDIT.md`, and an `AGENTS.md` bootstrap file. Designed to go beyond test-stub generators and prevent generic "coverage theatre" by tying every output to specific functions, schemas, or defensive patterns in the actual code.

Deploys and runs NVIDIA TAO Toolkit workflows on the Brev cloud platform, providing a streamlined way to fine-tune and optimize pretrained vision AI models on cloud GPU instances.

Trains action recognition models with NVIDIA TAO Toolkit to classify human activities in video, fine-tuning pretrained models with custom data for applications like surveillance, sports analytics, and gesture control.

Trains NVPanoptix3D panoptic segmentation models through NVIDIA TAO Toolkit, producing unified 3D scene understanding with both semantic and instance-level segmentation for autonomous systems.

Evaluation framework for AI agents and LLM applications that covers test suites, dataset generation, tracing, online evaluations, and prompt iteration for RAG pipelines, multi-turn chatbots, and tool-using workflows.

from recoupable/skills

Sets up VectorBT environment and dependencies for trading strategy backtesting.

Generates referring expression annotations with NVIDIA TAO Toolkit, producing natural language descriptions that identify specific objects in images for training vision-language understanding models.

Trains CenterPose object pose estimation models using NVIDIA TAO Toolkit, leveraging pretrained vision AI models that can be fine-tuned with custom data and exported for production deployment.

Trains DINO self-supervised vision transformer models via NVIDIA TAO Toolkit, learning powerful visual features without labeled data that can be transferred to downstream detection and segmentation tasks.

Trains Masked Auto Encoder (MAE) self-supervised vision models through NVIDIA TAO Toolkit, learning rich visual representations from unlabeled image data for downstream fine-tuning tasks.

Trains metric learning recognition models with NVIDIA TAO Toolkit, learning embedding spaces for similarity-based visual recognition tasks such as re-identification, face verification, and product matching.

Trains NVDINOv2 self-supervised vision models using NVIDIA TAO Toolkit, leveraging NVIDIA's optimized DINOv2 variant for feature extraction and transfer learning across vision tasks.

Provides a streamlined single-step training workflow in NVIDIA TAO Toolkit, simplifying the process of training vision AI models by consolidating configuration, training, and export into one unified command.

from botpress/skills

A responsiveness check skill from a Claude Code skills collection that guides Claude through verifying and ensuring responsive design across different screen sizes.

Analyzes gaps in VLM (Vision Language Model) BCQ evaluation results within NVIDIA TAO Toolkit, identifying performance weaknesses and quality benchmarks to guide targeted model improvements.

Mines AOI (Automated Optical Inspection) images within NVIDIA TAO Toolkit's data pipeline, selecting and curating the most informative samples from inspection datasets for efficient model training.

Runs NVIDIA TAO Toolkit workflows on Slurm-managed HPC clusters, enabling large-scale distributed training of vision AI models across multi-node GPU environments in research and enterprise settings.

Trains Depth Anything V2 monocular depth estimation models using NVIDIA TAO Toolkit, fine-tuning this state-of-the-art depth model with custom data for robust depth prediction across diverse scenes.

Trains Mask Auto Label models through NVIDIA TAO Toolkit, automating segmentation mask generation for vision AI datasets using pretrained models and low-code workflows.

Trains Mask2Former universal segmentation models using NVIDIA TAO Toolkit, supporting panoptic, instance, and semantic segmentation with a unified architecture fine-tuned on custom datasets.

Trains PointPillars 3D object detection models through NVIDIA TAO Toolkit, processing LiDAR point cloud data for autonomous driving and robotics with optimized models for edge deployment.

`/em:hard-call` framework for irreversible founder decisions (firing a co-founder, layoffs, pivots, killing a product) using a six-step process: reversibility test, 10/10/10 view, Andy Grove fresh-CEO test, stakeholder impact map, pre-announcement test, and communication plan. Includes decision-specific playbooks for each scenario.

Part of a comprehensive library of 235 production-ready Claude Code skills and agent plugins for 12 AI coding tools. Includes skills covering engineering, DevOps, marketing, compliance, and C-level advisory, with 305 Python CLI tools and compatibility across Claude Code, OpenAI Codex, Gemini CLI, Cursor, and 8 more platforms.

Trains Mask Grounding DINO models via NVIDIA TAO Toolkit, combining open-set object detection with instance segmentation to detect and segment objects from text prompts using custom training data.

Validates dataset formats within NVIDIA TAO Toolkit before training, checking annotation consistency, file structure, and data integrity to prevent errors during model training and optimization.