Claude Skills by anubhavg-icpl

Unsloth — 2x faster LLM fine-tuning with 70% less VRAM via fused Triton kernels. Use when fine-tuning, training, or adapting language models with unsloth techniques.

Expert in Upstash Redis, QStash, Vector, and Workflow for serverless and edge. Use when deploying to or building on upstash edge/serverless platform.

Use when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback

Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

ux-researcher. Use when you need help with ux researcher.

Pick VAD model, threshold, silence hangover, pre-roll, and turn-detection strategy for a voice agent. Use when you need help with vad tuner.

Specify VAE architecture, latent size, beta schedule, and eval plan for a given dataset and downstream use. Use when you need help with vae trainer.

Expert in vector databases for AI/ML applications including Pinecone, Weaviate, and Milvus. Use when you need deep expertise in vector database.

Build TypeScript LLM apps with streamText, generateObject, useChat, and tool calling. Use when building AI applications with vercel ai sdk.

Expert in the Vercel design system - Frontend deployment. Black and white precision, Geist font. Use when building UI components, applying design tokens, or implementing visual styles for vercel.

Expert in Vercel Functions, Fluid compute, ISR, and Image Optimization. Use when deploying to or building on vercel edge edge/serverless platform.

Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always

verifier. Use when you need help with verifier.

Deep expertise in Vespa — tensor framework, ranking expressions, ColBERT MaxSim, sparse + dense in one query, and multi-phase ranking. Use when implementing vector search, embeddings storage, or similarity queries with vespa.

Expert in the Vibrant design system - Lively, colorful design with bold playful typography, warm accents, and dynamic visual energy. Use when building UI components, applying design tokens, or implementing visual styles for vibrant.

Translate a video brief into a model + prompt + shot plan for a 2026 video generator. Use when you need help with video brief.

Build a video understanding pipeline with scene segmentation, multi-vector indexing, temporal grounding, and timestamped citations. Use when you need help with video qa.

|. Use when you need help with video shortform.

Expert in creating engaging developer tutorial videos and screencasts. Use when you need help with video tutorial creator.

Video understanding with VLMs - Qwen2.5-VL video, Apollo, LLaVA-OneVision, frame sampling. Use when working with multimodal AI (images, audio, video) using video vlm.

Plan frame sampling, per-frame pooling, output format, and benchmark targets for a video-language model deployment. Use when you need help with video vlm frame planner.

Expert in the Vintage design system - 1950s-1990s nostalgia with skeuomorphic touches, grainy textures, retro color palettes, and pixel-style typography. Use when building UI components, applying design tokens, or implementing visual styles for vintage.

Scaffold a MemGPT-shaped two-tier memory system (main context + archival store + memory tools) for any target runtime with correct eviction, citation, and untrusted-input handling. Use when you need help with virtual memory.

VLM landscape - Claude, GPT-4o, Llama 3.2 Vision, Qwen2.5-VL, Pixtral, MiniCPM-V, InternVL. Use when working with multimodal AI (images, audio, video) using vision llm.

Design a vision-native document RAG using ColPali / ColQwen2 / VisRAG, with storage estimate and generator-pick. Use when you need help with vision rag designer.

Expert in visual regression testing with Playwright, Chromatic, Percy, BackstopJS, and Storybook visual tests. Use when writing, running, or improving tests with visual regression testing.

Pick a ViT variant, patch size, and pretraining source for a new vision task. Use when you need help with vit configurator.

Expert in Vite build tool, configuration, plugins, optimization, and best practices for modern web development. Use when automating CI/CD, deployments, or operations with vite.

Expert in Vitest for blazing fast unit testing with native ESM support and Vite integration. Use when writing, running, or improving tests with vitest.

Pick an action format (discrete bin, FAST, flow-matching, dual-system) and VLA family (RT-2, OpenVLA, π0, GR00T) for a robot task. Use when you need help with vla action format picker.

Serve LLMs at scale with PagedAttention, continuous batching, and speculative decoding. Use when building AI applications with vllm.

Self-host vLLM in Docker for high-throughput local inference with tensor parallelism, prefix caching, and AWQ/GPTQ quantization. Use when deploying, running, or configuring local LLM inference with vllm local deploy.

Diagnose a vLLM serving config by reading the scheduler-level knobs and identifying which of PagedAttention, continuous batching, and chunked prefill is the bottleneck. Use when you need help with vllm scheduler reader.

Decide vLLM deployment layout — production-stack Helm chart, KV offload (native CPU or LMCache), router/observability integration — given workload and fleet size. Use when you need help with vllm stack decider.

Pick an open-weight VLM recipe (encoder, connector, LLM, data mix, resolution schedule) with ablation-table citations for every choice. Use when you need help with vlm recipe picker.

Expert in the Vodafone design system - Global telecom brand. Monumental uppercase display, Vodafone Red chapter bands. Use when building UI components, applying design tokens, or implementing visual styles for vodafone.

Build a real-time voice agent with sub-800ms first-audio-out, barge-in handling, and mid-conversation tool use. Use when you need help with voice agent.

Produce a full-stack voice-assistant spec — components, latency budget, observability, compliance — for a given workload. Use when you need help with voice assistant architect.

Pick cloning approach (zero-shot / conversion / adaptation), consent artifact, watermark, and safety filters for a voice-cloning deployment. Use when you need help with voice cloner.

Scaffold a Pipecat-shaped voice pipeline (VAD + STT + LLM + TTS + transport) with barge-in, confidence gating, and latency budget enforcement. Use when you need help with voice pipeline.

Expert in the VoltAgent design system - AI agent framework. Void-black canvas, emerald accent, terminal-native. Use when building UI components, applying design tokens, or implementing visual styles for voltagent.

Audit a scalable-oversight or W2SG claim via the performance-gap-recovered metric. Use when you need help with w2sg pgr.

|. Use when you need help with waitlist page.

Weights & Biases Weave — trace agents, log datasets, run evaluations, compare runs. Use when evaluating, monitoring, or observing LLM performance with wandb prompts.

Expert in the Warm Editorial design system - A serif-led magazine aesthetic. Terracotta accent on warm off-white paper —. Use when building UI components, applying design tokens, or implementing visual styles for warm editorial.

Expert in the Warp design system - Modern terminal. Dark IDE-like interface, block-based command UI. Use when building UI components, applying design tokens, or implementing visual styles for warp.

Wear OS 5/6 with Compose for Wear, tiles, complications, declarative Watch Face Format, and Health Services. Use when developing Android apps with wear os.

Deep expertise in Weaviate v1.27+ — collections, named vectors, vectorizer modules, hybrid search, and per-tenant shard isolation at million-tenant scale. Use when implementing vector search, embeddings storage, or similarity queries with weaviate.

|. Use when you need help with web design engineer.

Build a WebArena/OSWorld-style harness with execution-based evaluation and trajectory-efficiency metrics. Use when you need help with web desktop harness.