Deep Learning

Project management skill that analyzes tasks, breaks them down into stages, coordinates other skills, and ensures proper workflow. Use when planning features, managing complex multi-step implementations, or need help organizing development tasks. Routes work to specialized skills (testing, security, components, etc.).

Comprehensive unit testing expertise covering Vitest, Jest, test-driven development (TDD), mocking strategies, test coverage, snapshot testing, test architecture, testing patterns, dependency injection, test doubles (mocks, stubs, spies, fakes), async testing, error handling tests, parametric testing, test organization, code coverage analysis, mutation testing, and production-grade unit testing best practices. Activates for unit testing, vitest, jest, test-driven development, TDD, red-green-refactor, mocking, stubbing, spying, test doubles, test coverage, snapshot testing, test architecture, dependency injection, async testing, test patterns, code coverage, mutation testing, test isolation, test fixtures, AAA pattern, given-when-then, test organization, testing best practices, vi.fn, vi.mock, vi.spyOn, describe, it, expect, beforeEach, afterEach.

Validates Azure DevOps projects and resources exist, creates missing resources automatically. Smart enough to prompt user to select existing or create new projects. Supports multiple projects for project-per-team strategy, area paths for area-path-based strategy, and teams for team-based strategy. NEW - Per-project configuration support - AZURE_DEVOPS_AREA_PATHS_{ProjectName} and AZURE_DEVOPS_TEAMS_{ProjectName} for hierarchical organization. Activates for ado setup, ado validation, ado configuration, missing ado project, azure devops .env setup, per-project area paths, per-project teams.

Validates Jira projects and boards exist, creates missing resources automatically. Smart enough to prompt user to select existing or create new projects. For boards, accepts either IDs (validates existence) or names (creates boards and updates .env with IDs). NEW - Per-project configuration support - JIRA_BOARDS_{ProjectKey} for hierarchical board organization across multiple projects. Activates for jira setup, jira validation, jira configuration, missing jira project, missing jira boards, jira .env setup, per-project boards.

Computer vision ML pipelines for image classification, object detection, semantic segmentation, and image generation. Activates for "computer vision", "image classification", "object detection", "CNN", "ResNet", "YOLO", "image segmentation", "image preprocessing", "data augmentation". Builds end-to-end CV pipelines with PyTorch/TensorFlow, integrated with SpecWeave increments.

Guidance for implementing tensor parallelism in PyTorch, including ColumnParallelLinear and RowParallelLinear layers. This skill should be used when implementing distributed tensor parallel operations, sharding linear layers across multiple GPUs, or simulating collective operations like all-gather and all-reduce for parallel computation.

This skill provides guidance for implementing PyTorch pipeline parallelism for distributed training of large language models. It should be used when implementing pipeline parallel training loops, partitioning transformer models across GPUs, or working with AFAB (All-Forward-All-Backward) scheduling patterns. The skill covers model partitioning, inter-rank communication, gradient flow management, and common pitfalls in distributed training implementations.

Guidance for implementing CLI tools that perform inference using PyTorch models in native languages (C/C++/Rust). This skill should be used when tasks involve extracting weights from PyTorch .pth files, implementing neural network forward passes in C/C++, or creating standalone inference tools without Python dependencies.

Expert guidance for creating syntactically correct Mermaid diagrams. Use when creating flowcharts, sequence diagrams, class diagrams, state diagrams, Gantt charts, ER diagrams, or data lineage visualizations.

This skill provides guidance for implementing tensor parallelism in PyTorch, specifically column-parallel and row-parallel linear layers. Use when implementing distributed neural network layers that split weights/activations across multiple ranks, working with torch.distributed for model parallelism, or implementing ColumnParallelLinear and RowParallelLinear classes.

Guidance for implementing neural network inference (like GPT-2) under extreme code size constraints. This skill should be used when tasks require implementing ML model inference in minimal code (code golf), parsing model checkpoints in constrained environments, or building transformer architectures in low-level languages like C with strict size limits.

Guidance for recovering PyTorch model architectures from state dictionaries, retraining specific layers, and saving models in TorchScript format. This skill should be used when tasks involve reconstructing model architectures from saved weights, fine-tuning specific layers while freezing others, or converting models to TorchScript format.

Generate hierarchical AGENTS.md structure for codebases to optimize AI agent token usage. Use when creating AGENTS.md files, documenting codebase structure, setting up agent guidance, organizing project documentation for AI tools, implementing JIT indexing, or working with monorepos that need lightweight root guidance with detailed sub-folder documentation. Covers repository analysis, root AGENTS.md generation, sub-folder AGENTS.md creation, and token-efficient documentation patterns.

Guidance for processing financial documents (invoices, receipts, statements) with OCR and text extraction. This skill should be used when tasks involve extracting data from financial PDFs or images, generating summaries (CSV/JSON), or moving/organizing processed documents. Emphasizes data safety practices to prevent catastrophic data loss.

Guidance for extracting weight matrices from black-box ReLU neural networks using only input-output queries. This skill applies when tasks involve model extraction attacks, recovering hidden layer weights from neural networks, or reverse-engineering ReLU network parameters from query access.

This skill should be used when reconstructing PyTorch models from weight files (state dictionaries), checkpoint files, or partial model artifacts. It applies when the agent needs to infer model architecture from saved weights, rebuild models without original source code, or recover models from corrupted/incomplete saves. Use this skill for tasks involving torch.load, state_dict reconstruction, architecture inference, or model recovery in CPU-constrained environments.

Monitor and refactor large files into smaller, AI-friendly modules. Use when user asks to check file sizes, split large files, or organize the codebase. Ensures tests pass before and after refactoring.

Guidance for implementing batching schedulers for LLM inference systems with compilation-based accelerators. This skill applies when optimizing request batching to minimize cost while meeting latency thresholds, particularly when dealing with shape compilation costs, padding overhead, and multi-bucket request distributions. Use this skill for tasks involving batch planning, shape selection, generation-length bucketing, and cost-model-driven optimization for neural network inference.

Guidance for extracting weight matrices from black-box ReLU neural networks using only input-output queries. This skill applies when tasked with recovering internal parameters (weights, biases) of a neural network that can only be queried for outputs, particularly two-layer ReLU networks. Use this skill for model extraction, model stealing, or neural network reverse engineering tasks.

Guidance for creating standalone CLI tools that perform neural network inference by extracting PyTorch model weights and reimplementing inference in C/C++. This skill applies when tasks involve converting PyTorch models to standalone executables, extracting model weights to portable formats (JSON), implementing neural network forward passes in C/C++, or creating CLI tools that load images and run inference without Python dependencies.