1 774 machine learning skills from 282 authors. They train and fine-tune models, build embeddings, run RAG and measure quality. Half of them fit into 2 253 tokens or less — that is what one costs your context window when the agent loads it. 422 ship runnable scripts rather than instructions alone. 10 of them cannot work without an MCP server, most often rube. We also found 363 copies of these same skills sitting in other people's repositories — counted once here, not 363 times.
1 774 unique 282 authors 905 updated this month 182 from vendors
| Diagnose whether an LLM classifier's validation-gate failure is GOLD-BOUND pipeline over-predicts a label (precision low, recall high) and a prompt clarification is proposed to tighten it, (2) a pilot/validation gate fails and the fix candidates are prompt edits, (3) inter-rater agreement on the the exact feature the revision would exclude, no prompt can pass a gold-scored gate — recall craters while precision barely moves. Also documents the verified surgical-pilot design (single-section diff, tune/holdout split, pre-registered gate, perturbation check on untouched sections).
Autonomously research an ML task and run MANY bounded experiments to find the best config — a fixed-budget edit→train→eval→keep-or-discard loop in the spirit of karpathy/autoresearch, wrapped in the ml-intern orchestrator model and fanned out with a Claude Code dynamic workflow. Runs LONG: an iterative generational loop (mims-harvard/AutoScientists style) where parallel agent teams propose hypotheses, peer-critique them before spending any GPU, share findings on a common board, promote a champion, and keep going until budget/stagnation/convergence. Triggers when the user wants to "run many experiments", "sweep / search for the best config", "beat a benchmark", "do an ablation", "autoresearch X", "run for a long time / overnight / for days", or "find what improves metric Y on dataset Z". Deep-researches existing solutions across the internet FIRST (fan-out web search + PapersWithCode + GitHub, sources cross-checked into a cited DEEPRESEARCH.md), then ASKS where to get GPUs ("cards") and data before spending any compute, generates an experiment matrix seeded from diverse literature angles, runs it as a background workflow under an explicit budget, keeps a running leaderboard + shared findings board, verifies winners, and reports the best config. Reuses ml-intern's notify.sh + hf_push.sh for milestone alerts and HF Hub publishing.
| AI Native 产品方法论——RAG与知识系统设计的实操 Skill。 用户提供企业知识场景，Skill 自动执行知识系统设计流程： 资料来源分析 → 清洗与脱敏 → 索引与权限控制 → 检索召回 → 评估与更新 → 输出知识系统方案。 基于《AI Native 产品方法论》第15章。
AI Native 商业模式设计 Skill。基于《AI确定性商业模式》方法论， 帮助用户设计以"确定性溢价"为核心的 AI 商业模式： 避开6种失效模式，选择4种确定性模型，构建可持续的收费逻辑。
直觉训练——将直觉转化为可压缩的认知模型与模式识别
产品心理学——用户心理模型、行为设计、动机分析
美感的重新定义——生成焦虑、美感六维度、美感双轴模型
AI Native 商业模式设计 Skill。基于《AI确定性商业模式》方法论， 帮助用户设计以"确定性溢价"为核心的 AI 商业模式： 避开6种失效模式，选择4种确定性模型，构建可持续的收费逻辑。
数据飞轮构建器。基于《AI Native 营销与增长》数据飞轮概念卡， 帮助产品团队评估和构建自增强的数据飞轮：使用→数据→模型→产品→更多用户→更多数据。
AI Native 产品方法论——RAG与知识系统设计的实操 Skill。 用户提供企业知识场景，Skill 自动执行知识系统设计流程： 资料来源分析 → 清洗与脱敏 → 索引与权限控制 → 检索召回 → 评估与更新 → 输出知识系统方案。 基于《AI Native 产品方法论》第15章。 '
Produce an LLM Build Pack (prompt+tool contract, data/eval plan, architecture+safety, launch checklist). See also: ai-evals (eval only), ai-product-strategy (strategy only).
Build product taste via a Taste Calibration Sprint (benchmarks, critique notes, hypothesis log).
性能基准测试
| "tell me about [company]", "research [company]", "what does [company] do", "who is [company]", "look up [company]", "company deep dive", "due diligence on [company]", "background on [company]", "dig into [company]", "analyze [company]", or evaluating a company for investment, partnership, or sales. MUST be used instead of answering from memory — fetches real-time web data (funding, leadership changes, product launches, news) your training data lacks. Use even for well-known companies. Produces a sourced 360° report covering funding, leadership, product/tech, market position, news, and strategic outlook with dates and URLs. Do NOT use for multi-company competitor monitoring (use competitor-intel) or meeting prep with attendees (use meeting-prep).
> Find insurance payments, total annual premiums, and compare to benchmarks.
| 史蒂夫·乔布斯(Steve Jobs)的思维框架与表达方式。基于Isaacson授权传记、Stanford演讲、 Lost Interview、D Conference系列、Make Something Wonderful、30+一手来源的深度调研， 提炼6个核心心智模型、8条决策启发式和完整的表达DNA。 用途：作为一个可对话的思维顾问——用乔布斯的视角分析你的产品、决策与想法， 在一来一回的对话里制造思维碰撞、激发灵感，并帮你把"Jobs级"思考方式内化成自己的能力。 当用户提到「用乔布斯的视角」「乔布斯会怎么看」「Jobs模式」「steve jobs perspective」 「切换到乔布斯」「如果乔布斯」「乔布斯思维」「表达DNA」时使用。 即使用户只是说「帮我用乔布斯的角度想想」「如果乔布斯会怎么做」「切换到乔布斯」也应触发。
>- Analyze datasets and generate statistical summaries and insights. Use when asked to analyze data, compute statistics, or identify trends.
Implements and debugs browser Web Neural Network API integrations in JavaScript or TypeScript web apps. Use when adding navigator.ml checks, MLContext creation, MLGraphBuilder flows, device selection, tensor dispatch and readback, or explicit fallback paths to ONNX Runtime Web or other local runtimes. Don't use for model training, server-side ML inference, or cloud AI APIs.
业务诊断助手适合人力资源、管理者、运营、software在用户提出“这件事该怎么做”这类问题，需要快速拆解目标、判断重点并形成可执行结果时使用，帮助基于输入材料生成摘要、诊断结论、行动建议和可复用交付物。
业务诊断助手适合运营、市场营销、内容媒体、电商在用户提出“这件事该怎么做”这类问题，需要快速拆解目标、判断重点并形成可执行结果时使用，帮助基于输入材料生成摘要、诊断结论、行动建议和可复用交付物。
| Council · 智囊团：蒸馏真人思维框架为Advisor，并支持多Advisor圆桌讨论。 所有Advisor以persona文件形式存储在 personas/ 目录下，由Council统一管理。 三种用法： (A) 蒸馏：输入人名/主题/模糊需求 → 深度调研 → 提炼心智模型 → 生成Advisor persona (B) 激活：加载已有Advisor，以其视角回答问题 (C) 圆桌：召集多个Advisor → 独立发言 → 交叉质疑 → 综合输出 触发词（蒸馏）：「蒸馏XX」「造skill」「做个XX视角」「XX的思维方式」「更新XX的persona」 触发词（激活）：「用XX的视角」「XX会怎么看」「XX模式」「切换到XX」「ask XX」 触发词（圆桌）：「问问council」「圆桌讨论」「让XX和YY讨论」「ask the council」「council session」 模糊需求也触发：「我想提升决策质量」「有没有一种思维方式能帮我...」「我需要一个思维顾问」
Expert skill for integrating local Large Language Models using llama.cpp and Ollama. Covers secure model loading, inference optimization, prompt handling, and protection against LLM-specific vulnerabilities including prompt injection, model theft, and denial of service attacks.
Orchestrate AI/ML pipelines for data ingestion, model training, batch inference, and RAG indexing using Prefect, Airflow, or Dagster. Build reliable, observable, and retriable workflows for production AI systems.
Implement multi-layer LLM caching with exact match, semantic similarity, and provider-side prompt caching. Reduce API costs by 30–70%, cut latency, and improve throughput using Redis, GPTCache, and provider caching APIs.
Reduce LLM API and infrastructure costs through model selection, prompt caching, batching, caching, quantization, and self-hosting strategies. Track spend by team and model, set budgets, and implement cost-aware routing.
Build production LLMOps platforms with CI/CD, model promotion workflows, evaluation gates, rollback, and governance across cloud and self-hosted inference.
Monitor and evaluate RAG systems with retrieval quality metrics, groundedness checks, hallucination detection, and continuous regression testing.
Deploy, manage, and optimize vector databases for AI applications. Covers Qdrant, Weaviate, pgvector, and Pinecone — collection management, indexing strategies, backup, and performance tuning for production RAG and semantic search workloads.
Auto-scale LLM inference clusters on Kubernetes using KEDA, custom GPU metrics, and horizontal pod autoscaling. Handle traffic spikes, implement queue-based scaling, and optimize cost with spot instances for AI workloads.
Configure a Mac mini as a reliable local LLM server with remote access, observability, and power-safe operation. Use when building an always-on private AI inference server on Apple Silicon.
Design secure, multi-tenant LLM hosting platforms with tenant isolation, quotas, billing attribution, noisy-neighbor protection, and per-tenant policy controls.
Set up infrastructure for fine-tuning LLMs with QLoRA, LoRA, and full fine-tuning using Hugging Face TRL, Axolotl, and distributed training with DeepSpeed or FSDP. Covers dataset prep, training runs, and model export.
Run local LLM workloads with Ollama, Open WebUI, and GPU-aware tuning for private development environments. Use when setting up private inference, local AI dev environments, or air-gapped LLM deployments.
Build and operate Retrieval-Augmented Generation (RAG) infrastructure with vector stores, embedding pipelines, and hybrid search. Covers ingestion, chunking strategies, reranking, and production deployment patterns.
Deploy and manage vLLM for high-throughput LLM inference. Configure continuous batching, tensor parallelism, quantization, and OpenAI-compatible API endpoints for production LLM serving.
Use service mesh patterns for AI inference traffic management, mTLS, canary releases, policy enforcement, and cross-cluster resilience.
Deploy an API gateway for LLM traffic with load balancing, rate limiting, key management, semantic caching, fallback routing, and cost tracking. Covers LiteLLM Proxy, OpenRouter-compatible setup, and custom Nginx/Traefik patterns.
Set up and manage NVIDIA GPU servers for AI workloads — driver installation, CUDA toolkit, container toolkit, MIG partitioning, GPU health monitoring, and multi-GPU configuration for LLM inference and training.
Secure LLM-powered applications with input validation, output controls, tenant isolation, and abuse prevention.
Audit and remediate CIS benchmark violations. Use automated tools to assess compliance and implement hardening recommendations. Use when meeting compliance requirements or implementing security baselines.
Apply CIS benchmarks and secure Linux servers. Configure SSH, manage users, implement firewall rules, and enable security features. Use when hardening Linux systems for production or meeting security compliance requirements.
Harden Windows servers per security baselines and CIS benchmarks. Configure Group Policy, Windows Defender, and security features. Use when securing Windows Server environments.
Designs and runs organizational change programs: transformation planning, stakeholder engagement, resistance management, communication strategy, training design, and adoption measurement. Use when planning or delivering any organizational change (digital transformation, AI adoption, merger integration, culture shift, new operating model, ERP or system rollout) and when the ask is a change approach, stakeholder map, resistance playbook, communication plan, champion network, or adoption scorecard. Diagnoses where adoption stalls and prescribes the intervention rather than issuing generic advice.
Локальный LLM-стек на Mac (Apple Silicon, MLX) под приватность и запасной режим. NL-вход к установке/запуску/переключению моделей + слой суждения для мониторинга новых моделей. Тонкая обёртка над скриптами РП404, не замена.
Never teach a principle without an immediate rep. When designing ANY teaching content — a workshop, a lesson plan, a deck, a course module, a webinar, an explainer video, an educational post — pair every theoretical unit with an embodiment the learner does right now. Use whenever building or reviewing teaching material, lesson plans, workshop flows, course outlines, or training content, OR when the user says 'teach-by-doing', 'add an exercise', 'make it hands-on', 'don't leave it abstract', 'תרגיל לכל עיקרון', 'הטמעה מיידית'.
Hugging Face Transformers best practices including model loading, tokenization, fine-tuning workflows, and inference optimization. Use when working with transformer models, fine-tuning LLMs, implementing NLP tasks, or optimizing transformer inference.
Implement Retrieval-Augmented Generation (RAG) systems with LangChain4j. Build document ingestion pipelines, embedding stores, vector search strategies, and knowledge-enhanced AI applications. Use when creating question-answering systems over document collections or AI assistants with external knowledge bases.
Implement comprehensive evaluation strategies for LLM applications using automated metrics, human feedback, and benchmarking. Use when testing LLM performance, measuring AI application quality, or establishing evaluation frameworks.