Generate a reusable, platform-agnostic Python SDK for any task domain using the Task as Code (TaC) pattern — the generalized form of Perplexity's Search as Code architecture. Use this skill whenever someone wants to make a complex multi-step task reusable and programmable: building an SDK for a domain (design systems, data analysis, document processing, code review, market research, competitive intelligence), creating a workflow that benefits from parallelism and composable primitives, or packaging agentic task logic so any LLM can orchestrate it via generated code. Trigger on: build me an SDK for X, make X programmatic, create a reusable workflow for X, turn X into code, generate a design system SDK, I want to automate X with an LLM, package this as a reusable skill, or any request to structure a complex domain task as composable Python building blocks.
Detect and remediate AI-generated slop in copywriting/text, frontend UI/design, React components, and backend/architecture code. Use this skill whenever reviewing code before shipping, auditing AI-generated content, doing code reviews, evaluating copywriting, or when asked to improve text/UI/code that looks or feels AI-generated. Trigger on phrases like: does this look AI-generated, clean up this copy, this looks like AI slop, review this code for quality, make this less AI, improve this UI, audit for vibe coding anti-patterns, this copy sounds robotic, or any request to review or improve AI-assisted output across text, UI, or code.
Embodies the accumulated wisdom of a 20+ year career at high-performing software companies. Use this skill whenever the user asks for engineering guidance, code review feedback, architecture decisions, system design critique, or help avoiding common engineering mistakes. Also trigger when users ask "how should I approach X", "is this a good pattern", "what would a senior engineer do here", "how do I influence without authority", "how should I structure this system", or any question about engineering best practices, technical decision-making, code quality, or working effectively in engineering teams. This skill encodes hard-won lessons that take decades to learn — apply it liberally to elevate the quality of any engineering work. When orchestrating subagents, use Part 6 for delegation patterns and Part 8 for coordination protocols.
Expert LLM workflow and agent platform architect skill for designing scalable, production-grade agentic AI systems using API-based LLMs. Use this skill whenever you are designing, building, or reviewing any LLM agent system, workflow platform, or multi-agent architecture. Trigger on requests involving agent system design, multi-agent coordination, orchestrator/worker patterns, RAG pipeline architecture, tool/function calling design, agent memory and state management, LLM observability and evaluation pipelines, context window management, agentic workflow decomposition, or any system where LLMs call other LLMs. Also trigger when asked to build a copilot, assistant, agent, pipeline, or workflow of any kind. This skill encodes 2025-era best practices distilled from Anthropic, OpenAI, Microsoft, AWS, and Google engineering teams.
Expert Chrome developer skill covering advanced browser APIs and capabilities. Use this skill whenever working on Chrome extensions (Manifest V3), WebAssembly (Wasm 3.0), WebGPU (graphics and compute), WebNN (neural network inference), Chrome Built-in AI (Gemini Nano Prompt API, Summarizer, Translator), local ML model inference in the browser, IndexedDB (as SQL/NoSQL/vector/graph database), downloading and running HuggingFace models locally with WebGPU, or any advanced Chrome platform feature. Trigger this skill for requests about: building Chrome extensions, MV3 service workers, offscreen documents, declarativeNetRequest, running LLMs or ML models client-side (Qwen3, Llama, Phi, Gemma, Mistral, SmolLM), GPU compute shaders, WGSL shader language, WebAssembly SIMD/threads/GC, on-device AI without server infrastructure, browser performance optimization with WASM/GPU, WebGPU for machine learning, ONNX Runtime Web, Transformers.js v3, WebLLM, model quantization (q4/q4f16/fp16), IndexedDB as a database (key-value,
Expert ML engineering skill covering the full lifecycle — data ingestion, feature engineering, model training, evaluation, MLOps, and production deployment. Covers classical ML (scikit-learn, XGBoost) AND deep learning (PyTorch, CNNs, transformers, LoRA/QLoRA, LLMs). Trigger on building ML pipelines, training/evaluating models, experiment tracking, model versioning/deployment, drift detection, hyperparameter tuning, PyTorch training loops, fine-tuning BERT or LLMs, building CNNs, LoRA fine-tuning, MLflow setup, DVC, BentoML, FastAPI serving, or any MLOps question. Use for architecture decisions around PyTorch, scikit-learn, XGBoost, Hugging Face Transformers, PEFT, TRL, Kubeflow, or any ML framework.
Create Product Requirements Documents (PRDs) optimized for implementation by LLM Coding Agents like Claude Code, Cursor, Devin, and Aider. Use when building PRDs that will be consumed by autonomous AI coding agents, when converting existing PRDs for AI implementation, when creating technical specifications for AI-driven development, or when the user mentions creating specs for "AI agents," "LLM coding tools," "Claude Code," "Cursor," "autonomous coding," or similar AI-powered development workflows.
Expert LLM prompt engineering skill for creating new prompts and optimizing existing ones. Use this skill whenever a user wants to write, improve, refine, audit, compress, or optimize any prompt for any LLM (Claude, GPT, Gemini, Llama, etc.). Trigger on requests like "write me a prompt", "improve this prompt", "make my prompt shorter", "why isn't my prompt working", "optimize this system prompt", "help me get better results from AI", "reduce tokens in my prompt", or any time the user pastes a prompt and wants help with it. Also use when building system prompts for AI applications, agents, or workflows.