Llm

regex-vs-llm-structured-text skill for choosing regex or LLM in structured text extraction. Start with deterministic parsing, add LLM validation for low-confidence edge cases, and use a cheaper, more reliable pipeline for documents, forms, invoices, and data analysis.

llm-trading-agent-security is a practical guide for securing autonomous trading agents with wallet authority. It covers prompt injection, spend limits, pre-send simulation, circuit breakers, MEV-aware execution, and key isolation to reduce financial-loss risk in a Security Audit.

foundation-models-on-device helps you build Apple FoundationModels features on iOS 26+ with on-device text generation, guided output with @Generable, tool calling, snapshot streaming, and availability checks for privacy-first apps.

cost-aware-llm-pipeline helps you build LLM workflows that control API spend with model routing, immutable cost tracking, retry handling, and prompt caching. Ideal for batch jobs, document pipelines, and Workflow Automation where output volume and quality tradeoffs need clear rules.

fact-checker is a prompt-driven skill for structured claim verification, source evaluation, and clear verdicts with confidence and context. Install it from Shubhamsaboo/awesome-llm-apps to fact check statements, rumors, statistics, and misleading claims with a repeatable workflow.

deep-research is a lightweight agent skill for structured web research. It helps clarify scope, gather multiple sources, evaluate credibility, and synthesize cited findings from a single SKILL.md workflow.

cso is a Chief Security Officer–style security audit skill for agents. It helps review codebases and workflows for secrets exposure, dependency and supply-chain risk, CI/CD security, and LLM/AI security using OWASP Top 10 and STRIDE. Use cso for structured Security Audit reviews with confidence gates, active verification, and trend tracking.

The evaluation-methodology skill explains PluginEval scoring for Model Evaluation, including layers, rubrics, composite scoring, badge thresholds, and practical guidance for interpreting results and improving weak dimensions.

prompt-engineering-patterns is a practical skill for production prompt design, covering install context, reusable templates, few-shot examples, structured outputs, and prompt optimization workflows for Context Engineering.

rag-implementation is a practical skill for planning RAG systems with vector databases, embeddings, retrieval patterns, and grounded-answer workflows. Use it to compare stack options, shape architecture decisions, and guide install and usage for document Q&A, knowledge assistants, and semantic search.

similarity-search-patterns helps you choose distance metrics, index types, and hybrid retrieval patterns for semantic search and RAG workflows. Use it to plan production vector search tradeoffs around recall, latency, and scale.

The hybrid-search-implementation skill shows how to combine vector and keyword retrieval with RRF, linear fusion, reranking, and cascade patterns for RAG and search systems.

langchain-architecture is a design guide for building LangChain 1.x and LangGraph applications. Use it to choose between chains, agents, retrieval, memory, and stateful orchestration patterns before implementation.

Use the llm-evaluation skill to design repeatable evaluation plans for LLM apps, prompts, RAG systems, and model changes with metrics, human review, benchmarking, and regression checks.

embedding-strategies helps you choose and optimize embedding models for semantic search and RAG workflows, with practical guidance on chunking, model tradeoffs, multilingual content, and retrieval evaluation.

ai-prompt-engineering-safety-review is a prompt audit skill for reviewing LLM prompts for safety, bias, security weaknesses, and output quality before production, evaluation, or customer-facing use.

agentic-eval is a GitHub Copilot skill that shows how to build evaluation loops for AI outputs using reflection, rubric-based critique, and evaluator-optimizer patterns.

develop-ai-functions-example helps you create or modify runnable AI SDK examples in vercel/ai under examples/ai-functions/src/. Use it to choose the right category, match repo conventions, and build minimal examples for provider validation, demos, or fixtures.

huggingface-papers helps you read Hugging Face paper pages in markdown and extract structured metadata from the papers API, including authors, linked models, datasets, Spaces, GitHub repos, and project pages. Use it for Hugging Face paper URLs, arXiv URLs or IDs, and Academic Research workflows that need paper-page evidence.

huggingface-llm-trainer helps you train or fine-tune language and vision models on Hugging Face Jobs with TRL or Unsloth. Use this huggingface-llm-trainer skill for SFT, DPO, GRPO, reward modeling, dataset checks, GPU selection, Hub saving, Trackio monitoring, and GGUF export for backend development workflows.

huggingface-local-models helps you find Hugging Face models that run locally with llama.cpp and GGUF, choose a practical quant, and launch on CPU, Apple Metal, CUDA, or ROCm. It covers model discovery, exact GGUF file lookup, server vs CLI setup, and a fast path for backend development and private local inference.

huggingface-community-evals helps you run Hugging Face Hub model evaluations locally with inspect-ai or lighteval. Use it for backend selection, smoke tests, and a practical guide to vLLM, Transformers, or accelerate. Not for HF Jobs orchestration, model-card PRs, .eval_results publishing, or community-evals automation.

llm-patterns helps you design AI-first application logic where LLMs handle reasoning, extraction, and generation while code handles validation, routing, and error handling. Use the llm-patterns skill for clearer prompt structure, testable LLM workflows, and practical guidance for Skill Authoring.

detecting-ai-model-prompt-injection-attacks is a cybersecurity skill for screening untrusted text before it reaches an LLM. It uses layered regex, heuristic scoring, and DeBERTa-based classification to flag direct and indirect prompt injection attacks. Useful for chatbot input validation, document ingestion, and Threat Modeling.