Skill UI

llama.cpp is a pure C/C++ library designed for efficient LLM inference, optimized to run on resource-constrained environments. It excels in scenarios where high-end NVIDIA GPUs (CUDA) are unavailable, making it ideal for Apple Silicon Macs, AMD/Intel GPUs, or edge deployment devices like Raspberry Pi. It supports GGUF quantization formats, enabling significant memory reduction and speed improvements for local, cross-platform AI applications.

Enables Claude to monitor and analyze application CPU usage, detect hotspots, inefficient loops and regexes, and offer optimization strategies such as asynchronous processing or algorithmic improvements to boost performance.

This skill simulates an expert performance engineer capable of diagnosing complex bottlenecks across the entire technology stack. It covers modern observability techniques (OpenTelemetry, APM), advanced profiling (CPU, Memory, I/O), and sophisticated load testing strategies (k6, Chaos Engineering). Use this when optimizing scalability, reducing latency, improving throughput, or architecting highly reliable, performant distributed systems in cloud environments.

Automates load, stress, spike, and endurance tests through the performance-test-suite plugin, collecting response time, throughput, and error-rate metrics to spot CPU/memory/database/network bottlenecks and deliver graphs plus optimization guidance.

Profiles application performance by analyzing CPU usage, memory allocation, and execution time with the application-profiler plugin, then identifies bottlenecks and suggests code-level optimizations for faster execution.

A comprehensive guide for profiling, analyzing, and optimizing Python code. It addresses identifying performance bottlenecks related to CPU usage, excessive memory consumption, I/O operations, and slow data processing. Use this skill to improve application speed and efficiency in production environments.

Guides adding PyTorch FSDP2’s fully_shard integration to training loops, covering correct distributed init, bottom-up sharding, mixed precision/offload policies, DTensor-aware optimizers, and distributed checkpointing so oversized models or DeviceMesh sharding goals work reliably.

Ray Data offers scalable distributed data processing for ML workloads, streaming across CPU/GPU clusters and integrating with PyTorch, TensorFlow, and Ray Train. Use it for batch inference, ETL, and multi-modal preprocessing from a laptop up to hundreds of nodes.

This skill provides comprehensive guardrails for securing Replit-hosted applications. It covers critical security policies including preventing hardcoded secrets, implementing resource usage limits (CPU/Memory), enforcing endpoint authentication, and ensuring proper deployment configuration. Use this when you need to harden a Replit app for production or audit its security posture.

RWKV is an RNN+Transformer hybrid architecture with linear-time inference, infinite context, and GPT-style training. It delivers constant-memory, streaming generation on CUDA and CPU, ideal for long-context, memory-constrained deployments.

A guide to the Swift 6.2 concurrency model, which mandates default single-threaded execution to eliminate accidental data races. Learn how to use MainActor for isolated consistency and utilize the @concurrent attribute for explicitly offloading CPU-intensive tasks to background threads, ensuring safe and predictable application architecture.

Tracks CPU, memory, disk/network I/O, DB connections, and threads via the resource-usage-tracker plugin to reveal performance bottlenecks, recommend right-sizing, and guide cost-saving actions.