[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"extension-skill-Orchestra-Research-modal-de":3,"guides-for-Orchestra-Research-modal":1841,"similar-k175v5fe4bt509v7dma1w27wbd86mwg7-de":1842},{"_creationTime":4,"_id":5,"children":6,"community":7,"display":9,"evaluation":15,"identity":229,"isFallback":210,"parentExtension":235,"providers":293,"relations":297,"repo":298,"tags":1839,"workflow":1840},1778695116697.1846,"k175v5fe4bt509v7dma1w27wbd86mwg7",[],{"reviewCount":8},0,{"description":10,"installMethods":11,"name":13,"sourceUrl":14},"Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as APIs, or running batch jobs with automatic scaling.",{"claudeCode":12},"Orchestra-Research/AI-Research-SKILLs","modal-serverless-gpu","https://github.com/Orchestra-Research/AI-Research-SKILLs",{"_creationTime":16,"_id":17,"extensionId":5,"locale":18,"result":19,"trustSignals":208,"workflow":227},1778696387488.196,"kn7f87qr0mzq66h06f6xapf28986nwjt","en",{"checks":20,"evaluatedAt":175,"extensionSummary":176,"features":177,"nonGoals":183,"promptVersionExtension":188,"promptVersionScoring":189,"purpose":190,"rationale":191,"score":192,"summary":193,"tags":194,"targetMarket":201,"tier":202,"useCases":203},[21,26,29,32,36,39,43,46,50,54,58,61,64,68,72,75,79,83,87,90,94,98,101,104,107,110,113,116,120,124,128,132,135,138,141,144,148,151,154,157,160,163,166,168,170,172],{"category":22,"check":23,"severity":24,"summary":25},"Practical Utility","Problem relevance","pass","The description clearly states the problem of managing ML infrastructure and names specific use cases like deploying models and batch jobs.",{"category":22,"check":27,"severity":24,"summary":28},"Unique selling proposition","Modal offers a distinct value proposition by providing on-demand GPU access with serverless architecture and Python-native definitions, differentiating it from direct infrastructure management or simpler APIs.",{"category":22,"check":30,"severity":24,"summary":31},"Production readiness","The skill provides comprehensive documentation, examples, and covers the full lifecycle for deploying ML workloads, including infrastructure, model loading, and deployment patterns.",{"category":33,"check":34,"severity":24,"summary":35},"Scope","Single responsibility principle","The skill focuses exclusively on providing serverless GPU infrastructure and deployment capabilities for ML workloads, without venturing into unrelated domains.",{"category":33,"check":37,"severity":24,"summary":38},"Description quality","The displayed description accurately reflects the capabilities described in the SKILL.md, focusing on serverless GPU platforms for ML.",{"category":40,"check":41,"severity":24,"summary":42},"Invocation","Precise Purpose","The description clearly defines the artifact (ML workloads on Modal's platform) and the user intent (on-demand GPU access, deploying models, batch jobs) with clear boundaries provided in the 'Use alternatives instead' section.",{"category":40,"check":44,"severity":24,"summary":45},"Concise Frontmatter","The frontmatter is concise and self-contained, clearly stating the core capability and listing relevant trigger phrases within a reasonable character limit.",{"category":47,"check":48,"severity":24,"summary":49},"Documentation","Concise Body","The SKILL.md body is concise, focusing on key concepts and examples, with detailed information deferred to reference files.",{"category":51,"check":52,"severity":24,"summary":53},"Context","Progressive Disclosure","The SKILL.md outlines the main flow and links to separate reference files for advanced topics like multi-GPU training and container configurations.",{"category":51,"check":55,"severity":56,"summary":57},"Forked exploration","not_applicable","This skill does not involve deep exploration or research requiring a forked context.",{"category":22,"check":59,"severity":24,"summary":60},"Usage examples","The SKILL.md includes multiple end-to-end, runnable examples demonstrating GPU usage, inference endpoints, and core concepts.",{"category":22,"check":62,"severity":24,"summary":63},"Edge cases","The troubleshooting guide addresses common issues like installation failures, GPU problems, cold starts, and web endpoint errors with suggested solutions.",{"category":65,"check":66,"severity":56,"summary":67},"Code Execution","Tool Fallback","The skill does not appear to rely on external MCP servers or custom tools that would require fallbacks.",{"category":69,"check":70,"severity":56,"summary":71},"Safety","Halt on unexpected state","The skill's focus on infrastructure and deployment does not inherently involve states that would require halting the workflow due to unexpected conditions.",{"category":40,"check":73,"severity":24,"summary":74},"Non-Goals","The description explicitly lists alternatives, clearly delineating when to use Modal and when other tools might be more appropriate.",{"category":76,"check":77,"severity":24,"summary":78},"Install","Installation instruction","The SKILL.md provides clear installation instructions, including CLI commands for setup and usage examples.",{"category":80,"check":81,"severity":24,"summary":82},"Errors","Actionable error messages","The troubleshooting guide provides specific error messages and actionable solutions for common problems.",{"category":84,"check":85,"severity":24,"summary":86},"Execution","Pinned dependencies","The installation section shows `pip install modal` and the SKILL.md shows example `pip_install` commands, implying dependency management is handled.",{"category":33,"check":88,"severity":56,"summary":89},"Dry-run preview","The skill focuses on infrastructure setup and deployment, which typically do not have a 'dry-run' mode in the same sense as state-changing commands.",{"category":91,"check":92,"severity":24,"summary":93},"Protocol","Idempotent retry & timeouts","The documentation mentions increasing `timeout` and handling retries, suggesting consideration for these protocol aspects.",{"category":95,"check":96,"severity":24,"summary":97},"Security","Secret Management","The SKILL.md details how to use secrets for authentication (e.g., Hugging Face token) and implies secure handling through environment variables.",{"category":95,"check":99,"severity":56,"summary":100},"Injection","The skill itself is a Python script defining infrastructure, not processing untrusted external data as instructions.",{"category":95,"check":102,"severity":56,"summary":103},"Transitive Supply-Chain Grenades","The skill defines infrastructure configurations and does not fetch or execute arbitrary remote code or content.",{"category":95,"check":105,"severity":24,"summary":106},"Sandbox Isolation","The skill defines infrastructure and container configurations, operating within the defined Modal environment, not modifying external files.",{"category":95,"check":108,"severity":24,"summary":109},"Sandbox escape primitives","The skill's code does not contain primitives for escaping sandbox environments.",{"category":95,"check":111,"severity":24,"summary":112},"Data Exfiltration","The skill focuses on infrastructure and deployment configurations, with no indication of reading or submitting confidential data.",{"category":95,"check":114,"severity":24,"summary":115},"Hidden Text Tricks","The bundled content and documentation appear free of hidden text tricks or obfuscation.",{"category":117,"check":118,"severity":24,"summary":119},"Hooks","Opaque code execution","The code in the SKILL.md and references is plain Python and does not involve obfuscated or dynamically fetched code execution.",{"category":121,"check":122,"severity":24,"summary":123},"Portability","Structural Assumption","The skill primarily defines infrastructure configuration and code within its own environment, not making assumptions about user project structure outside of necessary file paths for image builds.",{"category":125,"check":126,"severity":24,"summary":127},"Trust","Issues Attention","With 4 open and 8 closed issues in the last 90 days, maintainers show good engagement.",{"category":129,"check":130,"severity":24,"summary":131},"Versioning","Release Management","The SKILL.md frontmatter includes a `version: 1.0.0`, indicating clear release management.",{"category":65,"check":133,"severity":24,"summary":134},"Validation","The Python code and examples demonstrate proper input handling and configuration, implying validation of parameters.",{"category":95,"check":136,"severity":24,"summary":137},"Unguarded Destructive Operations","While deployment can be destructive, the skill itself defines configurations rather than executing direct destructive commands without gates. Modal's platform handles deployment safety.",{"category":65,"check":139,"severity":24,"summary":140},"Error Handling","The troubleshooting guide implicitly covers error handling by providing solutions to common errors, indicating that errors are surfaced and addressed.",{"category":65,"check":142,"severity":24,"summary":143},"Logging","The 'Monitoring and Observability' section and debugging tips mention structured logging and viewing container logs, suggesting logging capabilities.",{"category":145,"check":146,"severity":56,"summary":147},"Compliance","GDPR","The skill focuses on infrastructure configuration and does not appear to operate on personal data.",{"category":145,"check":149,"severity":24,"summary":150},"Target market","The extension's focus on cloud infrastructure and ML workloads is globally applicable, with no regional restrictions detected.",{"category":121,"check":152,"severity":24,"summary":153},"Runtime stability","Modal's serverless nature and Python-native definition promote portability across different environments where Modal is supported.",{"category":47,"check":155,"severity":24,"summary":156},"README","The main README.md for the repository provides a comprehensive overview of the AI Research Skills library, including the Modal skill.",{"category":33,"check":158,"severity":56,"summary":159},"Tool surface size","This is a skill defining infrastructure and deployment patterns, not a set of distinct tools.",{"category":40,"check":161,"severity":56,"summary":162},"Overlapping near-synonym tools","The skill does not expose multiple tools with overlapping or synonymous functionality.",{"category":47,"check":164,"severity":24,"summary":165},"Phantom features","All features described in the documentation, including GPU options, container configurations, and web endpoints, have corresponding implementations and examples.",{"category":47,"check":52,"severity":24,"summary":167},"The SKILL.md is concise, and detailed information is available in separate reference files for advanced topics.",{"category":65,"check":133,"severity":24,"summary":169},"The provided Python code demonstrates structured configuration and parameter usage, suggesting validation.",{"category":91,"check":92,"severity":24,"summary":171},"The troubleshooting guide and documentation mention increasing timeouts and handling errors, which aligns with ensuring retryability and stable operations.",{"category":145,"check":173,"severity":56,"summary":174},"Telemetry opt-in","The skill focuses on infrastructure configuration and does not appear to emit telemetry.",1778696386856,"This skill provides guidance on using Modal, a serverless GPU cloud platform, for running ML workloads. It details how to define infrastructure in Python, deploy models as APIs, run batch jobs, and optimize performance with various GPU configurations and container images. It also covers advanced topics like web endpoints, persistent storage, and scheduling.",[178,179,180,181,182],"Serverless GPUs on-demand (T4, A10G, A100, H100, etc.)","Python-native infrastructure definition","Auto-scaling for ML workloads","Deploying ML models as REST APIs","Running batch processing jobs with automatic scaling",[184,185,186,187],"Using alternatives like RunPod for longer-running pods with persistent state","Using Lambda Labs for reserved GPU instances","Using SkyPilot for multi-cloud orchestration and cost optimization","Using Kubernetes for complex multi-service architectures","3.0.0","4.4.0","To enable users to run ML workloads on-demand with GPU access without managing infrastructure, by leveraging Modal's serverless platform for deployment and batch processing.","The skill is well-documented, production-ready, and provides clear value. Minor improvements could be made in explicitly detailing error handling for edge cases beyond the troubleshooting guide, but overall quality is very high.",95,"Excellent skill for serverless GPU ML workloads, offering robust documentation and clear utility.",[195,196,197,198,199,200],"infrastructure","serverless","gpu","cloud","deployment","mlops","global","verified",[204,205,206,207],"Running GPU-intensive ML workloads without managing infrastructure","Deploying ML models as auto-scaling APIs","Running batch processing jobs (training, inference, data processing)","Prototyping ML applications quickly",{"codeQuality":209,"collectedAt":211,"documentation":212,"maintenance":215,"popularity":222,"security":223,"testCoverage":226},{"hasLockfile":210},true,1778696368169,{"descriptionLength":213,"readmeSize":214},204,45313,{"closedIssues90d":216,"forks":217,"hasChangelog":210,"manifestVersion":218,"openIssues90d":219,"pushedAt":220,"stars":221},8,640,"1.0.0",4,1777352967000,8343,{"npmDownloads":8},{"hasNpmPackage":210,"license":224,"smitheryVerified":225},"MIT",false,{"hasCi":210,"hasTests":225},{"updatedAt":228},1778696387488,{"basePath":230,"githubOwner":231,"githubRepo":232,"locale":18,"slug":233,"type":234},"09-infrastructure/modal","Orchestra-Research","AI-Research-SKILLs","modal","skill",{"_creationTime":236,"_id":237,"community":238,"display":239,"identity":243,"parentExtension":246,"providers":278,"relations":289,"tags":290,"workflow":291},1778695116697.1702,"k17155ws9qc0hw7a568bg79sfd86max8",{"reviewCount":8},{"description":240,"installMethods":241,"name":242,"sourceUrl":14},"LLM architectures and implementations including LitGPT, Mamba, NanoGPT, RWKV, and TorchTitan. Use when implementing, training, or understanding transformer and alternative architectures.",{"claudeCode":232},"Agent-Native Research Artifact (ARA) Tooling",{"basePath":244,"githubOwner":231,"githubRepo":232,"locale":18,"slug":232,"type":245},"","plugin",{"_creationTime":247,"_id":248,"community":249,"display":250,"identity":254,"providers":256,"relations":272,"tags":274,"workflow":275},1778695116697.17,"k17755pkhk2ktxts0edcsj00s586nmvk",{"reviewCount":8},{"description":251,"installMethods":252,"name":253,"sourceUrl":14},"Comprehensive library of 98 AI research engineering skills enabling autonomous AI research from hypothesis to experimental verification",{"claudeCode":12},"AI Research Skills Library",{"basePath":244,"githubOwner":231,"githubRepo":232,"locale":18,"slug":232,"type":255},"marketplace",{"evaluate":257,"extract":265},{"promptVersionExtension":258,"promptVersionScoring":189,"score":259,"tags":260,"targetMarket":201,"tier":202},"3.1.0",99,[261,200,262,263,264],"ai-research","llm-skills","autonomous-agents","research-orchestration",{"commitSha":266,"license":224,"marketplace":267,"plugin":270},"HEAD",{"name":268,"pluginCount":269},"ai-research-skills",1,{"mcpCount":8,"provider":271,"skillCount":8},"classify",{"repoId":273},"kd70hj1y80mhra5xm5g188j5n586mg18",[261,263,262,200,264],{"evaluatedAt":276,"extractAt":277,"updatedAt":276},1778695131103,1778695116697,{"evaluate":279,"extract":286},{"promptVersionExtension":188,"promptVersionScoring":189,"score":280,"tags":281,"targetMarket":201,"tier":202},98,[282,283,284,285,261],"research","artifact","provenance","review",{"commitSha":266,"license":224,"plugin":287},{"mcpCount":8,"provider":271,"skillCount":288},96,{"parentExtensionId":248,"repoId":273},[261,283,284,282,285],{"evaluatedAt":292,"extractAt":277,"updatedAt":292},1778695555085,{"evaluate":294,"extract":296},{"promptVersionExtension":188,"promptVersionScoring":189,"score":192,"tags":295,"targetMarket":201,"tier":202},[195,196,197,198,199,200],{"commitSha":266},{"parentExtensionId":237,"repoId":273},{"_creationTime":299,"_id":273,"identity":300,"providers":301,"workflow":1834},1778695107142.3535,{"githubOwner":231,"githubRepo":232,"sourceUrl":14},{"classify":302,"discover":1812,"extract":1815,"github":1816,"npm":1833},{"commitSha":266,"extensions":303},[304,317,324,331,338,345,352,359,366,373,379,386,393,400,406,413,420,427,434,441,448,455,462,469,476,500,516,530,544,557,572,587,597,613,629,641,656,669,680,691,702,714,725,738,749,765,779,789,799,815,825,833,841,849,857,871,885,917,931,941,951,958,968,978,992,1003,1013,1025,1035,1054,1067,1082,1098,1112,1126,1139,1152,1165,1177,1189,1202,1221,1231,1244,1257,1270,1279,1289,1298,1308,1318,1330,1343,1356,1368,1378,1388,1398,1408,1418,1430,1439,1457,1473,1483,1493,1503,1513,1527,1540,1550,1563,1575,1589,1696,1706,1743,1751,1759,1773,1787,1797],{"basePath":244,"description":251,"displayName":268,"installMethods":305,"rationale":306,"selectedPaths":307,"source":316,"sourceLanguage":18,"type":255},{"claudeCode":12},"marketplace.json at .claude-plugin/marketplace.json",[308,311,313],{"path":309,"priority":310},".claude-plugin/marketplace.json","mandatory",{"path":312,"priority":310},"README.md",{"path":314,"priority":315},"LICENSE","high","rule",{"basePath":244,"description":240,"displayName":318,"installMethods":319,"rationale":320,"selectedPaths":321,"source":316,"sourceLanguage":18,"type":245},"model-architecture",{"claudeCode":232},"inline plugin source from marketplace.json at /",[322,323],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":325,"displayName":326,"installMethods":327,"rationale":320,"selectedPaths":328,"source":316,"sourceLanguage":18,"type":245},"Text tokenization for LLMs including HuggingFace Tokenizers and SentencePiece. Use when training custom tokenizers or handling multilingual text.","tokenization",{"claudeCode":232},[329,330],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":332,"displayName":333,"installMethods":334,"rationale":320,"selectedPaths":335,"source":316,"sourceLanguage":18,"type":245},"LLM fine-tuning frameworks including Axolotl, LLaMA-Factory, PEFT, and Unsloth. Use when fine-tuning models with LoRA, QLoRA, or full fine-tuning.","fine-tuning",{"claudeCode":232},[336,337],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":339,"displayName":340,"installMethods":341,"rationale":320,"selectedPaths":342,"source":316,"sourceLanguage":18,"type":245},"Neural network interpretability tools including TransformerLens, SAELens, NNSight, and pyvene. Use when analyzing model internals, finding circuits, or understanding how models compute.","mechanistic-interpretability",{"claudeCode":232},[343,344],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":346,"displayName":347,"installMethods":348,"rationale":320,"selectedPaths":349,"source":316,"sourceLanguage":18,"type":245},"Data curation and processing at scale including NeMo Curator and Ray Data. Use when preparing training datasets or processing large-scale data.","data-processing",{"claudeCode":232},[350,351],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":353,"displayName":354,"installMethods":355,"rationale":320,"selectedPaths":356,"source":316,"sourceLanguage":18,"type":245},"RLHF and preference alignment including TRL, GRPO, OpenRLHF, SimPO, verl, slime, miles, and torchforge. Use when aligning models with human preferences, training reward models, or large-scale RL training.","post-training",{"claudeCode":232},[357,358],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":360,"displayName":361,"installMethods":362,"rationale":320,"selectedPaths":363,"source":316,"sourceLanguage":18,"type":245},"AI safety and content moderation including Constitutional AI, LlamaGuard, NeMo Guardrails, and Prompt Guard. Use when implementing safety filters, content moderation, or prompt injection detection.","safety-alignment",{"claudeCode":232},[364,365],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":367,"displayName":368,"installMethods":369,"rationale":320,"selectedPaths":370,"source":316,"sourceLanguage":18,"type":245},"Multi-GPU and multi-node training including DeepSpeed, PyTorch FSDP, Accelerate, Megatron-Core, PyTorch Lightning, and Ray Train. Use when training large models across GPUs.","distributed-training",{"claudeCode":232},[371,372],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":374,"displayName":195,"installMethods":375,"rationale":320,"selectedPaths":376,"source":316,"sourceLanguage":18,"type":245},"GPU cloud and compute orchestration including Modal, Lambda Labs, and SkyPilot. Use when deploying training jobs or managing GPU resources.",{"claudeCode":232},[377,378],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":380,"displayName":381,"installMethods":382,"rationale":320,"selectedPaths":383,"source":316,"sourceLanguage":18,"type":245},"Model optimization and quantization including Flash Attention, bitsandbytes, GPTQ, AWQ, GGUF, and HQQ. Use when reducing memory, accelerating inference, or quantizing models.","optimization",{"claudeCode":232},[384,385],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":387,"displayName":388,"installMethods":389,"rationale":320,"selectedPaths":390,"source":316,"sourceLanguage":18,"type":245},"LLM benchmarking and evaluation including lm-evaluation-harness, BigCode Evaluation Harness, and NeMo Evaluator. Use when benchmarking models or measuring performance.","evaluation",{"claudeCode":232},[391,392],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":394,"displayName":395,"installMethods":396,"rationale":320,"selectedPaths":397,"source":316,"sourceLanguage":18,"type":245},"Production LLM inference including vLLM, TensorRT-LLM, llama.cpp, and SGLang. Use when deploying models for production inference.","inference-serving",{"claudeCode":232},[398,399],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":401,"displayName":200,"installMethods":402,"rationale":320,"selectedPaths":403,"source":316,"sourceLanguage":18,"type":245},"ML experiment tracking and lifecycle including Weights & Biases, MLflow, and TensorBoard. Use when tracking experiments or managing models.",{"claudeCode":232},[404,405],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":407,"displayName":408,"installMethods":409,"rationale":320,"selectedPaths":410,"source":316,"sourceLanguage":18,"type":245},"LLM agent frameworks including LangChain, LlamaIndex, CrewAI, and AutoGPT. Use when building chatbots, autonomous agents, or tool-using systems.","agents",{"claudeCode":232},[411,412],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":414,"displayName":415,"installMethods":416,"rationale":320,"selectedPaths":417,"source":316,"sourceLanguage":18,"type":245},"Retrieval-Augmented Generation including Chroma, FAISS, Pinecone, Qdrant, and Sentence Transformers. Use when building semantic search or document retrieval systems.","rag",{"claudeCode":232},[418,419],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":421,"displayName":422,"installMethods":423,"rationale":320,"selectedPaths":424,"source":316,"sourceLanguage":18,"type":245},"Structured LLM outputs including DSPy, Instructor, Guidance, and Outlines. Use when extracting structured data or constraining LLM outputs.","prompt-engineering",{"claudeCode":232},[425,426],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":428,"displayName":429,"installMethods":430,"rationale":320,"selectedPaths":431,"source":316,"sourceLanguage":18,"type":245},"LLM application monitoring including LangSmith and Phoenix. Use when debugging LLM apps or monitoring production systems.","observability",{"claudeCode":232},[432,433],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":435,"displayName":436,"installMethods":437,"rationale":320,"selectedPaths":438,"source":316,"sourceLanguage":18,"type":245},"Vision, audio, and multimodal models including CLIP, Whisper, LLaVA, BLIP-2, Segment Anything, Stable Diffusion, AudioCraft, Cosmos Policy, OpenPI, and OpenVLA-OFT. Use when working with images, audio, multimodal tasks, or vision-language-action robot policies.","multimodal",{"claudeCode":232},[439,440],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":442,"displayName":443,"installMethods":444,"rationale":320,"selectedPaths":445,"source":316,"sourceLanguage":18,"type":245},"Advanced ML techniques including MoE Training, Model Merging, Long Context, Speculative Decoding, Knowledge Distillation, and Model Pruning. Use when implementing cutting-edge optimization or architecture techniques.","emerging-techniques",{"claudeCode":232},[446,447],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":449,"displayName":450,"installMethods":451,"rationale":320,"selectedPaths":452,"source":316,"sourceLanguage":18,"type":245},"Autonomous research orchestration using a two-loop architecture. Manages the full research lifecycle from literature survey to paper writing, routing to domain-specific skills for execution. Use when starting a research project, running autonomous experiments, or managing multi-hypothesis research.","autoresearch",{"claudeCode":232},[453,454],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":456,"displayName":457,"installMethods":458,"rationale":320,"selectedPaths":459,"source":316,"sourceLanguage":18,"type":245},"Write publication-ready ML/AI/Systems papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM, OSDI, NSDI, ASPLOS, SOSP. Includes LaTeX templates, citation verification, reviewer guidelines, publication-quality figure generation, systems paper structural blueprints, and conference presentation slides.","ml-paper-writing",{"claudeCode":232},[460,461],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":463,"displayName":464,"installMethods":465,"rationale":320,"selectedPaths":466,"source":316,"sourceLanguage":18,"type":245},"Research ideation frameworks including structured brainstorming and creative thinking. Use when exploring new research directions, generating novel ideas, or seeking fresh angles on existing work.","ideation",{"claudeCode":232},[467,468],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":244,"description":470,"displayName":471,"installMethods":472,"rationale":320,"selectedPaths":473,"source":316,"sourceLanguage":18,"type":245},"Agent-Native Research Artifact (ARA) tooling: compile any research input (paper, repo, notes) into a structured artifact, record session provenance as a post-task epilogue, and run Seal Level 2 epistemic review. Use when ingesting research into a falsifiable, agent-traversable artifact, capturing how a research project actually evolved, or auditing an ARA for evidence-claim alignment.","agent-native-research-artifact",{"claudeCode":232},[474,475],{"path":312,"priority":310},{"path":314,"priority":315},{"basePath":477,"description":478,"displayName":450,"installMethods":479,"rationale":480,"selectedPaths":481,"source":316,"sourceLanguage":18,"type":234},"0-autoresearch-skill","Orchestrates end-to-end autonomous AI research projects using a two-loop architecture. The inner loop runs rapid experiment iterations with clear optimization targets. The outer loop synthesizes results, identifies patterns, and steers research direction. Routes to domain-specific skills for execution, supports continuous agent operation via Claude Code /loop and OpenClaw heartbeat, and produces research presentations and papers. Use when starting a research project, running autonomous experiments, or managing a multi-hypothesis research effort.",{"claudeCode":12},"SKILL.md frontmatter at 0-autoresearch-skill/SKILL.md",[482,484,487,489,491,494,496,498],{"path":483,"priority":310},"SKILL.md",{"path":485,"priority":486},"references/agent-continuity.md","medium",{"path":488,"priority":486},"references/progress-reporting.md",{"path":490,"priority":486},"references/skill-routing.md",{"path":492,"priority":493},"templates/findings.md","low",{"path":495,"priority":493},"templates/progress-presentation.html",{"path":497,"priority":493},"templates/research-log.md",{"path":499,"priority":493},"templates/research-state.yaml",{"basePath":501,"description":502,"displayName":503,"installMethods":504,"rationale":505,"selectedPaths":506,"source":316,"sourceLanguage":18,"type":234},"01-model-architecture/litgpt","Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.","implementing-llms-litgpt",{"claudeCode":12},"SKILL.md frontmatter at 01-model-architecture/litgpt/SKILL.md",[507,508,510,512,514],{"path":483,"priority":310},{"path":509,"priority":486},"references/custom-models.md",{"path":511,"priority":486},"references/distributed-training.md",{"path":513,"priority":486},"references/supported-models.md",{"path":515,"priority":486},"references/training-recipes.md",{"basePath":517,"description":518,"displayName":519,"installMethods":520,"rationale":521,"selectedPaths":522,"source":316,"sourceLanguage":18,"type":234},"01-model-architecture/mamba","State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.","mamba-architecture",{"claudeCode":12},"SKILL.md frontmatter at 01-model-architecture/mamba/SKILL.md",[523,524,526,528],{"path":483,"priority":310},{"path":525,"priority":486},"references/architecture-details.md",{"path":527,"priority":486},"references/benchmarks.md",{"path":529,"priority":486},"references/training-guide.md",{"basePath":531,"description":532,"displayName":533,"installMethods":534,"rationale":535,"selectedPaths":536,"source":316,"sourceLanguage":18,"type":234},"01-model-architecture/nanogpt","Educational GPT implementation in ~300 lines. Reproduces GPT-2 (124M) on OpenWebText. Clean, hackable code for learning transformers. By Andrej Karpathy. Perfect for understanding GPT architecture from scratch. Train on Shakespeare (CPU) or OpenWebText (multi-GPU).","nanogpt",{"claudeCode":12},"SKILL.md frontmatter at 01-model-architecture/nanogpt/SKILL.md",[537,538,540,542],{"path":483,"priority":310},{"path":539,"priority":486},"references/architecture.md",{"path":541,"priority":486},"references/data.md",{"path":543,"priority":486},"references/training.md",{"basePath":545,"description":546,"displayName":547,"installMethods":548,"rationale":549,"selectedPaths":550,"source":316,"sourceLanguage":18,"type":234},"01-model-architecture/rwkv","RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.","rwkv-architecture",{"claudeCode":12},"SKILL.md frontmatter at 01-model-architecture/rwkv/SKILL.md",[551,552,553,555],{"path":483,"priority":310},{"path":525,"priority":486},{"path":554,"priority":486},"references/rwkv7.md",{"path":556,"priority":486},"references/state-management.md",{"basePath":558,"description":559,"displayName":560,"installMethods":561,"rationale":562,"selectedPaths":563,"source":316,"sourceLanguage":18,"type":234},"01-model-architecture/torchtitan","Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.","distributed-llm-pretraining-torchtitan",{"claudeCode":12},"SKILL.md frontmatter at 01-model-architecture/torchtitan/SKILL.md",[564,565,567,568,570],{"path":483,"priority":310},{"path":566,"priority":486},"references/checkpoint.md",{"path":509,"priority":486},{"path":569,"priority":486},"references/float8.md",{"path":571,"priority":486},"references/fsdp.md",{"basePath":573,"description":574,"displayName":575,"installMethods":576,"rationale":577,"selectedPaths":578,"source":316,"sourceLanguage":18,"type":234},"02-tokenization/huggingface-tokenizers","Fast tokenizers optimized for research and production. Rust-based implementation tokenizes 1GB in \u003C20 seconds. Supports BPE, WordPiece, and Unigram algorithms. Train custom vocabularies, track alignments, handle padding/truncation. Integrates seamlessly with transformers. Use when you need high-performance tokenization or custom tokenizer training.","huggingface-tokenizers",{"claudeCode":12},"SKILL.md frontmatter at 02-tokenization/huggingface-tokenizers/SKILL.md",[579,580,582,584,586],{"path":483,"priority":310},{"path":581,"priority":486},"references/algorithms.md",{"path":583,"priority":486},"references/integration.md",{"path":585,"priority":486},"references/pipeline.md",{"path":543,"priority":486},{"basePath":588,"description":589,"displayName":590,"installMethods":591,"rationale":592,"selectedPaths":593,"source":316,"sourceLanguage":18,"type":234},"02-tokenization/sentencepiece","Language-independent tokenizer treating text as raw Unicode. Supports BPE and Unigram algorithms. Fast (50k sentences/sec), lightweight (6MB memory), deterministic vocabulary. Used by T5, ALBERT, XLNet, mBART. Train on raw text without pre-tokenization. Use when you need multilingual support, CJK languages, or reproducible tokenization.","sentencepiece",{"claudeCode":12},"SKILL.md frontmatter at 02-tokenization/sentencepiece/SKILL.md",[594,595,596],{"path":483,"priority":310},{"path":581,"priority":486},{"path":543,"priority":486},{"basePath":598,"description":599,"displayName":600,"installMethods":601,"rationale":602,"selectedPaths":603,"source":316,"sourceLanguage":18,"type":234},"03-fine-tuning/axolotl","Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support","axolotl",{"claudeCode":12},"SKILL.md frontmatter at 03-fine-tuning/axolotl/SKILL.md",[604,605,607,609,611],{"path":483,"priority":310},{"path":606,"priority":486},"references/api.md",{"path":608,"priority":486},"references/dataset-formats.md",{"path":610,"priority":486},"references/index.md",{"path":612,"priority":486},"references/other.md",{"basePath":614,"description":615,"displayName":616,"installMethods":617,"rationale":618,"selectedPaths":619,"source":316,"sourceLanguage":18,"type":234},"03-fine-tuning/llama-factory","Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support","llama-factory",{"claudeCode":12},"SKILL.md frontmatter at 03-fine-tuning/llama-factory/SKILL.md",[620,621,623,625,627,628],{"path":483,"priority":310},{"path":622,"priority":486},"references/_images.md",{"path":624,"priority":486},"references/advanced.md",{"path":626,"priority":486},"references/getting_started.md",{"path":610,"priority":486},{"path":612,"priority":486},{"basePath":630,"description":631,"displayName":632,"installMethods":633,"rationale":634,"selectedPaths":635,"source":316,"sourceLanguage":18,"type":234},"03-fine-tuning/peft","Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train \u003C1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.","peft-fine-tuning",{"claudeCode":12},"SKILL.md frontmatter at 03-fine-tuning/peft/SKILL.md",[636,637,639],{"path":483,"priority":310},{"path":638,"priority":486},"references/advanced-usage.md",{"path":640,"priority":486},"references/troubleshooting.md",{"basePath":642,"description":643,"displayName":644,"installMethods":645,"rationale":646,"selectedPaths":647,"source":316,"sourceLanguage":18,"type":234},"03-fine-tuning/unsloth","Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization","unsloth",{"claudeCode":12},"SKILL.md frontmatter at 03-fine-tuning/unsloth/SKILL.md",[648,649,650,652,654],{"path":483,"priority":310},{"path":610,"priority":486},{"path":651,"priority":486},"references/llms-full.md",{"path":653,"priority":486},"references/llms-txt.md",{"path":655,"priority":486},"references/llms.md",{"basePath":657,"description":658,"displayName":659,"installMethods":660,"rationale":661,"selectedPaths":662,"source":316,"sourceLanguage":18,"type":234},"04-mechanistic-interpretability/nnsight","Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run interpretability experiments on massive models (70B+) without local GPU resources, or when working with any PyTorch architecture.","nnsight-remote-interpretability",{"claudeCode":12},"SKILL.md frontmatter at 04-mechanistic-interpretability/nnsight/SKILL.md",[663,664,666,667],{"path":483,"priority":310},{"path":665,"priority":486},"references/README.md",{"path":606,"priority":486},{"path":668,"priority":486},"references/tutorials.md",{"basePath":670,"description":671,"displayName":672,"installMethods":673,"rationale":674,"selectedPaths":675,"source":316,"sourceLanguage":18,"type":234},"04-mechanistic-interpretability/pyvene","Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.","pyvene-interventions",{"claudeCode":12},"SKILL.md frontmatter at 04-mechanistic-interpretability/pyvene/SKILL.md",[676,677,678,679],{"path":483,"priority":310},{"path":665,"priority":486},{"path":606,"priority":486},{"path":668,"priority":486},{"basePath":681,"description":682,"displayName":683,"installMethods":684,"rationale":685,"selectedPaths":686,"source":316,"sourceLanguage":18,"type":234},"04-mechanistic-interpretability/saelens","Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.","sparse-autoencoder-training",{"claudeCode":12},"SKILL.md frontmatter at 04-mechanistic-interpretability/saelens/SKILL.md",[687,688,689,690],{"path":483,"priority":310},{"path":665,"priority":486},{"path":606,"priority":486},{"path":668,"priority":486},{"basePath":692,"description":693,"displayName":694,"installMethods":695,"rationale":696,"selectedPaths":697,"source":316,"sourceLanguage":18,"type":234},"04-mechanistic-interpretability/transformer-lens","Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.","transformer-lens-interpretability",{"claudeCode":12},"SKILL.md frontmatter at 04-mechanistic-interpretability/transformer-lens/SKILL.md",[698,699,700,701],{"path":483,"priority":310},{"path":665,"priority":486},{"path":606,"priority":486},{"path":668,"priority":486},{"basePath":703,"description":704,"displayName":705,"installMethods":706,"rationale":707,"selectedPaths":708,"source":316,"sourceLanguage":18,"type":234},"05-data-processing/nemo-curator","GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristics), semantic deduplication, PII redaction, NSFW detection. Scales across GPUs with RAPIDS. Use for preparing high-quality training datasets, cleaning web data, or deduplicating large corpora.","nemo-curator",{"claudeCode":12},"SKILL.md frontmatter at 05-data-processing/nemo-curator/SKILL.md",[709,710,712],{"path":483,"priority":310},{"path":711,"priority":486},"references/deduplication.md",{"path":713,"priority":486},"references/filtering.md",{"basePath":715,"description":716,"displayName":717,"installMethods":718,"rationale":719,"selectedPaths":720,"source":316,"sourceLanguage":18,"type":234},"05-data-processing/ray-data","Scalable data processing for ML workloads. Streaming execution across CPU/GPU, supports Parquet/CSV/JSON/images. Integrates with Ray Train, PyTorch, TensorFlow. Scales from single machine to 100s of nodes. Use for batch inference, data preprocessing, multi-modal data loading, or distributed ETL pipelines.","ray-data",{"claudeCode":12},"SKILL.md frontmatter at 05-data-processing/ray-data/SKILL.md",[721,722,723],{"path":483,"priority":310},{"path":583,"priority":486},{"path":724,"priority":486},"references/transformations.md",{"basePath":726,"description":727,"displayName":728,"installMethods":729,"rationale":730,"selectedPaths":731,"source":316,"sourceLanguage":18,"type":234},"06-post-training/grpo-rl-training","Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training","grpo-rl-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/grpo-rl-training/SKILL.md",[732,733,734,736],{"path":483,"priority":310},{"path":312,"priority":315},{"path":735,"priority":493},"examples/reward_functions_library.py",{"path":737,"priority":493},"templates/basic_grpo_training.py",{"basePath":739,"description":740,"displayName":741,"installMethods":742,"rationale":743,"selectedPaths":744,"source":316,"sourceLanguage":18,"type":234},"06-post-training/miles","Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing train-inference alignment, or requiring speculative RL for maximum throughput.","miles-rl-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/miles/SKILL.md",[745,746,748],{"path":483,"priority":310},{"path":747,"priority":486},"references/api-reference.md",{"path":640,"priority":486},{"basePath":750,"description":751,"displayName":752,"installMethods":753,"rationale":754,"selectedPaths":755,"source":316,"sourceLanguage":18,"type":234},"06-post-training/openrlhf","High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.","openrlhf-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/openrlhf/SKILL.md",[756,757,759,761,763],{"path":483,"priority":310},{"path":758,"priority":486},"references/algorithm-comparison.md",{"path":760,"priority":486},"references/custom-rewards.md",{"path":762,"priority":486},"references/hybrid-engine.md",{"path":764,"priority":486},"references/multi-node-training.md",{"basePath":766,"description":767,"displayName":768,"installMethods":769,"rationale":770,"selectedPaths":771,"source":316,"sourceLanguage":18,"type":234},"06-post-training/simpo","Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.","simpo-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/simpo/SKILL.md",[772,773,775,777],{"path":483,"priority":310},{"path":774,"priority":486},"references/datasets.md",{"path":776,"priority":486},"references/hyperparameters.md",{"path":778,"priority":486},"references/loss-functions.md",{"basePath":780,"description":781,"displayName":782,"installMethods":783,"rationale":784,"selectedPaths":785,"source":316,"sourceLanguage":18,"type":234},"06-post-training/slime","Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation workflows, or needing tight Megatron-LM integration for RL scaling.","slime-rl-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/slime/SKILL.md",[786,787,788],{"path":483,"priority":310},{"path":747,"priority":486},{"path":640,"priority":486},{"basePath":790,"description":791,"displayName":792,"installMethods":793,"rationale":794,"selectedPaths":795,"source":316,"sourceLanguage":18,"type":234},"06-post-training/torchforge","Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.","torchforge-rl-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/torchforge/SKILL.md",[796,797,798],{"path":483,"priority":310},{"path":747,"priority":486},{"path":640,"priority":486},{"basePath":800,"description":801,"displayName":802,"installMethods":803,"rationale":804,"selectedPaths":805,"source":316,"sourceLanguage":18,"type":234},"06-post-training/trl-fine-tuning","Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.","fine-tuning-with-trl",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/trl-fine-tuning/SKILL.md",[806,807,809,811,813],{"path":483,"priority":310},{"path":808,"priority":486},"references/dpo-variants.md",{"path":810,"priority":486},"references/online-rl.md",{"path":812,"priority":486},"references/reward-modeling.md",{"path":814,"priority":486},"references/sft-training.md",{"basePath":816,"description":817,"displayName":818,"installMethods":819,"rationale":820,"selectedPaths":821,"source":316,"sourceLanguage":18,"type":234},"06-post-training/verl","Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.","verl-rl-training",{"claudeCode":12},"SKILL.md frontmatter at 06-post-training/verl/SKILL.md",[822,823,824],{"path":483,"priority":310},{"path":747,"priority":486},{"path":640,"priority":486},{"basePath":826,"description":827,"displayName":828,"installMethods":829,"rationale":830,"selectedPaths":831,"source":316,"sourceLanguage":18,"type":234},"07-safety-alignment/constitutional-ai","Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from AI Feedback). Use for safety alignment, reducing harmful outputs without human labels. Powers Claude's safety system.","constitutional-ai",{"claudeCode":12},"SKILL.md frontmatter at 07-safety-alignment/constitutional-ai/SKILL.md",[832],{"path":483,"priority":310},{"basePath":834,"description":835,"displayName":836,"installMethods":837,"rationale":838,"selectedPaths":839,"source":316,"sourceLanguage":18,"type":234},"07-safety-alignment/llamaguard","Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-harm, criminal planning. 94-95% accuracy. Deploy with vLLM, HuggingFace, Sagemaker. Integrates with NeMo Guardrails.","llamaguard",{"claudeCode":12},"SKILL.md frontmatter at 07-safety-alignment/llamaguard/SKILL.md",[840],{"path":483,"priority":310},{"basePath":842,"description":843,"displayName":844,"installMethods":845,"rationale":846,"selectedPaths":847,"source":316,"sourceLanguage":18,"type":234},"07-safety-alignment/nemo-guardrails","NVIDIA's runtime safety framework for LLM applications. Features jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection. Uses Colang 2.0 DSL for programmable rails. Production-ready, runs on T4 GPU.","nemo-guardrails",{"claudeCode":12},"SKILL.md frontmatter at 07-safety-alignment/nemo-guardrails/SKILL.md",[848],{"path":483,"priority":310},{"basePath":850,"description":851,"displayName":852,"installMethods":853,"rationale":854,"selectedPaths":855,"source":316,"sourceLanguage":18,"type":234},"07-safety-alignment/prompt-guard","Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, \u003C1% FPR. Fast (\u003C2ms GPU). Multilingual (8 languages). Deploy with HuggingFace or batch processing for RAG security.","prompt-guard",{"claudeCode":12},"SKILL.md frontmatter at 07-safety-alignment/prompt-guard/SKILL.md",[856],{"path":483,"priority":310},{"basePath":858,"description":859,"displayName":860,"installMethods":861,"rationale":862,"selectedPaths":863,"source":316,"sourceLanguage":18,"type":234},"08-distributed-training/accelerate","Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.","huggingface-accelerate",{"claudeCode":12},"SKILL.md frontmatter at 08-distributed-training/accelerate/SKILL.md",[864,865,867,869],{"path":483,"priority":310},{"path":866,"priority":486},"references/custom-plugins.md",{"path":868,"priority":486},"references/megatron-integration.md",{"path":870,"priority":486},"references/performance.md",{"basePath":872,"description":873,"displayName":874,"installMethods":875,"rationale":876,"selectedPaths":877,"source":316,"sourceLanguage":18,"type":234},"08-distributed-training/megatron-core","Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.","training-llms-megatron",{"claudeCode":12},"SKILL.md frontmatter at 08-distributed-training/megatron-core/SKILL.md",[878,879,880,882,884],{"path":483,"priority":310},{"path":527,"priority":486},{"path":881,"priority":486},"references/parallelism-guide.md",{"path":883,"priority":486},"references/production-examples.md",{"path":515,"priority":486},{"basePath":886,"description":887,"displayName":888,"installMethods":889,"rationale":890,"selectedPaths":891,"source":316,"sourceLanguage":18,"type":234},"08-distributed-training/pytorch-fsdp2","Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.","pytorch-fsdp2",{"claudeCode":12},"SKILL.md frontmatter at 08-distributed-training/pytorch-fsdp2/SKILL.md",[892,893,895,897,899,901,903,905,907,909,911,913,915],{"path":483,"priority":310},{"path":894,"priority":486},"references/pytorch_dcp_async_recipe.md",{"path":896,"priority":486},"references/pytorch_dcp_overview.md",{"path":898,"priority":486},"references/pytorch_dcp_recipe.md",{"path":900,"priority":486},"references/pytorch_ddp_notes.md",{"path":902,"priority":486},"references/pytorch_device_mesh_tutorial.md",{"path":904,"priority":486},"references/pytorch_examples_fsdp2.md",{"path":906,"priority":486},"references/pytorch_fsdp1_api.md",{"path":908,"priority":486},"references/pytorch_fsdp2_tutorial.md",{"path":910,"priority":486},"references/pytorch_fully_shard_api.md",{"path":912,"priority":486},"references/pytorch_tp_tutorial.md",{"path":914,"priority":486},"references/ray_train_fsdp2_example.md",{"path":916,"priority":486},"references/torchtitan_fsdp_notes.md",{"basePath":918,"description":919,"displayName":920,"installMethods":921,"rationale":922,"selectedPaths":923,"source":316,"sourceLanguage":18,"type":234},"08-distributed-training/pytorch-lightning","High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.","pytorch-lightning",{"claudeCode":12},"SKILL.md frontmatter at 08-distributed-training/pytorch-lightning/SKILL.md",[924,925,927,929],{"path":483,"priority":310},{"path":926,"priority":486},"references/callbacks.md",{"path":928,"priority":486},"references/distributed.md",{"path":930,"priority":486},"references/hyperparameter-tuning.md",{"basePath":932,"description":933,"displayName":934,"installMethods":935,"rationale":936,"selectedPaths":937,"source":316,"sourceLanguage":18,"type":234},"08-distributed-training/ray-train","Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.","ray-train",{"claudeCode":12},"SKILL.md frontmatter at 08-distributed-training/ray-train/SKILL.md",[938,939],{"path":483,"priority":310},{"path":940,"priority":486},"references/multi-node.md",{"basePath":942,"description":943,"displayName":944,"installMethods":945,"rationale":946,"selectedPaths":947,"source":316,"sourceLanguage":18,"type":234},"09-infrastructure/lambda-labs","Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.","lambda-labs-gpu-cloud",{"claudeCode":12},"SKILL.md frontmatter at 09-infrastructure/lambda-labs/SKILL.md",[948,949,950],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":230,"description":10,"displayName":13,"installMethods":952,"rationale":953,"selectedPaths":954,"source":316,"sourceLanguage":18,"type":234},{"claudeCode":12},"SKILL.md frontmatter at 09-infrastructure/modal/SKILL.md",[955,956,957],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":959,"description":960,"displayName":961,"installMethods":962,"rationale":963,"selectedPaths":964,"source":316,"sourceLanguage":18,"type":234},"09-infrastructure/skypilot","Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.","skypilot-multi-cloud-orchestration",{"claudeCode":12},"SKILL.md frontmatter at 09-infrastructure/skypilot/SKILL.md",[965,966,967],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":969,"description":970,"displayName":971,"installMethods":972,"rationale":973,"selectedPaths":974,"source":316,"sourceLanguage":18,"type":234},"10-optimization/awq","Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.","awq-quantization",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/awq/SKILL.md",[975,976,977],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":979,"description":980,"displayName":981,"installMethods":982,"rationale":983,"selectedPaths":984,"source":316,"sourceLanguage":18,"type":234},"10-optimization/bitsandbytes","Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.","quantizing-models-bitsandbytes",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/bitsandbytes/SKILL.md",[985,986,988,990],{"path":483,"priority":310},{"path":987,"priority":486},"references/memory-optimization.md",{"path":989,"priority":486},"references/qlora-training.md",{"path":991,"priority":486},"references/quantization-formats.md",{"basePath":993,"description":994,"displayName":995,"installMethods":996,"rationale":997,"selectedPaths":998,"source":316,"sourceLanguage":18,"type":234},"10-optimization/flash-attention","Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.","optimizing-attention-flash",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/flash-attention/SKILL.md",[999,1000,1001],{"path":483,"priority":310},{"path":527,"priority":486},{"path":1002,"priority":486},"references/transformers-integration.md",{"basePath":1004,"description":1005,"displayName":1006,"installMethods":1007,"rationale":1008,"selectedPaths":1009,"source":316,"sourceLanguage":18,"type":234},"10-optimization/gguf","GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.","gguf-quantization",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/gguf/SKILL.md",[1010,1011,1012],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1014,"description":1015,"displayName":1016,"installMethods":1017,"rationale":1018,"selectedPaths":1019,"source":316,"sourceLanguage":18,"type":234},"10-optimization/gptq","Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with \u003C2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.","gptq",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/gptq/SKILL.md",[1020,1021,1023,1024],{"path":483,"priority":310},{"path":1022,"priority":486},"references/calibration.md",{"path":583,"priority":486},{"path":640,"priority":486},{"basePath":1026,"description":1027,"displayName":1028,"installMethods":1029,"rationale":1030,"selectedPaths":1031,"source":316,"sourceLanguage":18,"type":234},"10-optimization/hqq","Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.","hqq-quantization",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/hqq/SKILL.md",[1032,1033,1034],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1036,"description":1037,"displayName":1038,"installMethods":1039,"rationale":1040,"selectedPaths":1041,"source":316,"sourceLanguage":18,"type":234},"10-optimization/ml-training-recipes","Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning neural networks, debugging loss spikes or OOM, choosing architectures, or optimizing GPU throughput.","ml-training-recipes",{"claudeCode":12},"SKILL.md frontmatter at 10-optimization/ml-training-recipes/SKILL.md",[1042,1043,1044,1046,1048,1050,1052],{"path":483,"priority":310},{"path":539,"priority":486},{"path":1045,"priority":486},"references/biomedical.md",{"path":1047,"priority":486},"references/domain-specific.md",{"path":1049,"priority":486},"references/experiment-loop.md",{"path":1051,"priority":486},"references/optimizers.md",{"path":1053,"priority":486},"references/scaling-and-selection.md",{"basePath":1055,"description":1056,"displayName":1057,"installMethods":1058,"rationale":1059,"selectedPaths":1060,"source":316,"sourceLanguage":18,"type":234},"11-evaluation/bigcode-evaluation-harness","Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing coding abilities, testing multi-language support, or measuring code generation quality. Industry standard from BigCode Project used by HuggingFace leaderboards.","evaluating-code-models",{"claudeCode":12},"SKILL.md frontmatter at 11-evaluation/bigcode-evaluation-harness/SKILL.md",[1061,1062,1063,1065],{"path":483,"priority":310},{"path":527,"priority":486},{"path":1064,"priority":486},"references/custom-tasks.md",{"path":1066,"priority":486},"references/issues.md",{"basePath":1068,"description":1069,"displayName":1070,"installMethods":1071,"rationale":1072,"selectedPaths":1073,"source":316,"sourceLanguage":18,"type":234},"11-evaluation/lm-evaluation-harness","Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.","evaluating-llms-harness",{"claudeCode":12},"SKILL.md frontmatter at 11-evaluation/lm-evaluation-harness/SKILL.md",[1074,1075,1077,1079,1080],{"path":483,"priority":310},{"path":1076,"priority":486},"references/api-evaluation.md",{"path":1078,"priority":486},"references/benchmark-guide.md",{"path":1064,"priority":486},{"path":1081,"priority":486},"references/distributed-eval.md",{"basePath":1083,"description":1084,"displayName":1085,"installMethods":1086,"rationale":1087,"selectedPaths":1088,"source":316,"sourceLanguage":18,"type":234},"11-evaluation/nemo-evaluator","Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.","nemo-evaluator-sdk",{"claudeCode":12},"SKILL.md frontmatter at 11-evaluation/nemo-evaluator/SKILL.md",[1089,1090,1092,1094,1096],{"path":483,"priority":310},{"path":1091,"priority":486},"references/adapter-system.md",{"path":1093,"priority":486},"references/configuration.md",{"path":1095,"priority":486},"references/custom-benchmarks.md",{"path":1097,"priority":486},"references/execution-backends.md",{"basePath":1099,"description":1100,"displayName":1101,"installMethods":1102,"rationale":1103,"selectedPaths":1104,"source":316,"sourceLanguage":18,"type":234},"12-inference-serving/llama-cpp","Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.","llama-cpp",{"claudeCode":12},"SKILL.md frontmatter at 12-inference-serving/llama-cpp/SKILL.md",[1105,1106,1108,1110],{"path":483,"priority":310},{"path":1107,"priority":486},"references/optimization.md",{"path":1109,"priority":486},"references/quantization.md",{"path":1111,"priority":486},"references/server.md",{"basePath":1113,"description":1114,"displayName":1115,"installMethods":1116,"rationale":1117,"selectedPaths":1118,"source":316,"sourceLanguage":18,"type":234},"12-inference-serving/sglang","Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.","sglang",{"claudeCode":12},"SKILL.md frontmatter at 12-inference-serving/sglang/SKILL.md",[1119,1120,1122,1124],{"path":483,"priority":310},{"path":1121,"priority":486},"references/deployment.md",{"path":1123,"priority":486},"references/radix-attention.md",{"path":1125,"priority":486},"references/structured-generation.md",{"basePath":1127,"description":1128,"displayName":1129,"installMethods":1130,"rationale":1131,"selectedPaths":1132,"source":316,"sourceLanguage":18,"type":234},"12-inference-serving/tensorrt-llm","Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4), in-flight batching, and multi-GPU scaling.","tensorrt-llm",{"claudeCode":12},"SKILL.md frontmatter at 12-inference-serving/tensorrt-llm/SKILL.md",[1133,1134,1136,1137],{"path":483,"priority":310},{"path":1135,"priority":486},"references/multi-gpu.md",{"path":1107,"priority":486},{"path":1138,"priority":486},"references/serving.md",{"basePath":1140,"description":1141,"displayName":1142,"installMethods":1143,"rationale":1144,"selectedPaths":1145,"source":316,"sourceLanguage":18,"type":234},"12-inference-serving/vllm","Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.","serving-llms-vllm",{"claudeCode":12},"SKILL.md frontmatter at 12-inference-serving/vllm/SKILL.md",[1146,1147,1148,1149,1151],{"path":483,"priority":310},{"path":1107,"priority":486},{"path":1109,"priority":486},{"path":1150,"priority":486},"references/server-deployment.md",{"path":640,"priority":486},{"basePath":1153,"description":1154,"displayName":1155,"installMethods":1156,"rationale":1157,"selectedPaths":1158,"source":316,"sourceLanguage":18,"type":234},"13-mlops/mlflow","Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform","mlflow",{"claudeCode":12},"SKILL.md frontmatter at 13-mlops/mlflow/SKILL.md",[1159,1160,1161,1163],{"path":483,"priority":310},{"path":1121,"priority":486},{"path":1162,"priority":486},"references/model-registry.md",{"path":1164,"priority":486},"references/tracking.md",{"basePath":1166,"description":1167,"displayName":1168,"installMethods":1169,"rationale":1170,"selectedPaths":1171,"source":316,"sourceLanguage":18,"type":234},"13-mlops/swanlab","Provides guidance for experiment tracking with SwanLab. Use when you need open-source run tracking, local or self-hosted dashboards, and lightweight media logging for ML workflows.","experiment-tracking-swanlab",{"claudeCode":12},"SKILL.md frontmatter at 13-mlops/swanlab/SKILL.md",[1172,1173,1175],{"path":483,"priority":310},{"path":1174,"priority":486},"references/integrations.md",{"path":1176,"priority":486},"references/visualization.md",{"basePath":1178,"description":1179,"displayName":1180,"installMethods":1181,"rationale":1182,"selectedPaths":1183,"source":316,"sourceLanguage":18,"type":234},"13-mlops/tensorboard","Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit","tensorboard",{"claudeCode":12},"SKILL.md frontmatter at 13-mlops/tensorboard/SKILL.md",[1184,1185,1186,1188],{"path":483,"priority":310},{"path":1174,"priority":486},{"path":1187,"priority":486},"references/profiling.md",{"path":1176,"priority":486},{"basePath":1190,"description":1191,"displayName":1192,"installMethods":1193,"rationale":1194,"selectedPaths":1195,"source":316,"sourceLanguage":18,"type":234},"13-mlops/weights-and-biases","Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform","weights-and-biases",{"claudeCode":12},"SKILL.md frontmatter at 13-mlops/weights-and-biases/SKILL.md",[1196,1197,1199,1200],{"path":483,"priority":310},{"path":1198,"priority":486},"references/artifacts.md",{"path":1174,"priority":486},{"path":1201,"priority":486},"references/sweeps.md",{"basePath":1203,"description":1204,"displayName":1205,"installMethods":1206,"rationale":1207,"selectedPaths":1208,"source":316,"sourceLanguage":18,"type":234},"14-agents/a-evolve","Provides guidance for automatically evolving and optimizing AI agents across any domain using LLM-driven evolution algorithms. Use when building self-improving agents, optimizing agent prompts and skills against benchmarks, or implementing automated agent evaluation loops.","evolving-ai-agents",{"claudeCode":12},"SKILL.md frontmatter at 14-agents/a-evolve/SKILL.md",[1209,1210,1211,1212,1213,1215,1217,1218,1220],{"path":483,"priority":310},{"path":665,"priority":486},{"path":606,"priority":486},{"path":539,"priority":486},{"path":1214,"priority":486},"references/design-patterns.md",{"path":1216,"priority":486},"references/examples.md",{"path":1066,"priority":486},{"path":1219,"priority":486},"references/releases.md",{"path":668,"priority":486},{"basePath":1222,"description":1223,"displayName":1224,"installMethods":1225,"rationale":1226,"selectedPaths":1227,"source":316,"sourceLanguage":18,"type":234},"14-agents/autogpt","Autonomous AI agent platform for building and deploying continuous agents. Use when creating visual workflow agents, deploying persistent autonomous agents, or building complex multi-step AI automation systems.","autogpt-agents",{"claudeCode":12},"SKILL.md frontmatter at 14-agents/autogpt/SKILL.md",[1228,1229,1230],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1232,"description":1233,"displayName":1234,"installMethods":1235,"rationale":1236,"selectedPaths":1237,"source":316,"sourceLanguage":18,"type":234},"14-agents/crewai","Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when you need role-based agent collaboration with memory, or for production workflows requiring sequential/hierarchical execution. Built without LangChain dependencies for lean, fast execution.","crewai-multi-agent",{"claudeCode":12},"SKILL.md frontmatter at 14-agents/crewai/SKILL.md",[1238,1239,1241,1243],{"path":483,"priority":310},{"path":1240,"priority":486},"references/flows.md",{"path":1242,"priority":486},"references/tools.md",{"path":640,"priority":486},{"basePath":1245,"description":1246,"displayName":1247,"installMethods":1248,"rationale":1249,"selectedPaths":1250,"source":316,"sourceLanguage":18,"type":234},"14-agents/langchain","Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, ReAct agents, tool calling, memory management, and vector store retrieval. Use for building chatbots, question-answering systems, autonomous agents, or RAG applications. Best for rapid prototyping and production deployments.","langchain",{"claudeCode":12},"SKILL.md frontmatter at 14-agents/langchain/SKILL.md",[1251,1252,1254,1255],{"path":483,"priority":310},{"path":1253,"priority":486},"references/agents.md",{"path":583,"priority":486},{"path":1256,"priority":486},"references/rag.md",{"basePath":1258,"description":1259,"displayName":1260,"installMethods":1261,"rationale":1262,"selectedPaths":1263,"source":316,"sourceLanguage":18,"type":234},"14-agents/llamaindex","Data framework for building LLM applications with RAG. Specializes in document ingestion (300+ connectors), indexing, and querying. Features vector indices, query engines, agents, and multi-modal support. Use for document Q&A, chatbots, knowledge retrieval, or building RAG pipelines. Best for data-centric LLM applications.","llamaindex",{"claudeCode":12},"SKILL.md frontmatter at 14-agents/llamaindex/SKILL.md",[1264,1265,1266,1268],{"path":483,"priority":310},{"path":1253,"priority":486},{"path":1267,"priority":486},"references/data_connectors.md",{"path":1269,"priority":486},"references/query_engines.md",{"basePath":1271,"description":1272,"displayName":1273,"installMethods":1274,"rationale":1275,"selectedPaths":1276,"source":316,"sourceLanguage":18,"type":234},"15-rag/chroma","Open-source embedding database for AI applications. Store embeddings and metadata, perform vector and full-text search, filter by metadata. Simple 4-function API. Scales from notebooks to production clusters. Use for semantic search, RAG applications, or document retrieval. Best for local development and open-source projects.","chroma",{"claudeCode":12},"SKILL.md frontmatter at 15-rag/chroma/SKILL.md",[1277,1278],{"path":483,"priority":310},{"path":583,"priority":486},{"basePath":1280,"description":1281,"displayName":1282,"installMethods":1283,"rationale":1284,"selectedPaths":1285,"source":316,"sourceLanguage":18,"type":234},"15-rag/faiss","Facebook's library for efficient similarity search and clustering of dense vectors. Supports billions of vectors, GPU acceleration, and various index types (Flat, IVF, HNSW). Use for fast k-NN search, large-scale vector retrieval, or when you need pure similarity search without metadata. Best for high-performance applications.","faiss",{"claudeCode":12},"SKILL.md frontmatter at 15-rag/faiss/SKILL.md",[1286,1287],{"path":483,"priority":310},{"path":1288,"priority":486},"references/index_types.md",{"basePath":1290,"description":1291,"displayName":1292,"installMethods":1293,"rationale":1294,"selectedPaths":1295,"source":316,"sourceLanguage":18,"type":234},"15-rag/pinecone","Managed vector database for production AI applications. Fully managed, auto-scaling, with hybrid search (dense + sparse), metadata filtering, and namespaces. Low latency (\u003C100ms p95). Use for production RAG, recommendation systems, or semantic search at scale. Best for serverless, managed infrastructure.","pinecone",{"claudeCode":12},"SKILL.md frontmatter at 15-rag/pinecone/SKILL.md",[1296,1297],{"path":483,"priority":310},{"path":1121,"priority":486},{"basePath":1299,"description":1300,"displayName":1301,"installMethods":1302,"rationale":1303,"selectedPaths":1304,"source":316,"sourceLanguage":18,"type":234},"15-rag/qdrant","High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor search, hybrid search with filtering, or scalable vector storage with Rust-powered performance.","qdrant-vector-search",{"claudeCode":12},"SKILL.md frontmatter at 15-rag/qdrant/SKILL.md",[1305,1306,1307],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1309,"description":1310,"displayName":1311,"installMethods":1312,"rationale":1313,"selectedPaths":1314,"source":316,"sourceLanguage":18,"type":234},"15-rag/sentence-transformers","Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval. Supports multilingual, domain-specific, and multimodal models. Use for generating embeddings for RAG, semantic search, or similarity tasks. Best for production embedding generation.","sentence-transformers",{"claudeCode":12},"SKILL.md frontmatter at 15-rag/sentence-transformers/SKILL.md",[1315,1316],{"path":483,"priority":310},{"path":1317,"priority":486},"references/models.md",{"basePath":1319,"description":1320,"displayName":1321,"installMethods":1322,"rationale":1323,"selectedPaths":1324,"source":316,"sourceLanguage":18,"type":234},"16-prompt-engineering/dspy","Build complex AI systems with declarative programming, optimize prompts automatically, create modular RAG systems and agents with DSPy - Stanford NLP's framework for systematic LM programming","dspy",{"claudeCode":12},"SKILL.md frontmatter at 16-prompt-engineering/dspy/SKILL.md",[1325,1326,1327,1329],{"path":483,"priority":310},{"path":1216,"priority":486},{"path":1328,"priority":486},"references/modules.md",{"path":1051,"priority":486},{"basePath":1331,"description":1332,"displayName":1333,"installMethods":1334,"rationale":1335,"selectedPaths":1336,"source":316,"sourceLanguage":18,"type":234},"16-prompt-engineering/guidance","Control LLM output with regex and grammars, guarantee valid JSON/XML/code generation, enforce structured formats, and build multi-step workflows with Guidance - Microsoft Research's constrained generation framework","guidance",{"claudeCode":12},"SKILL.md frontmatter at 16-prompt-engineering/guidance/SKILL.md",[1337,1338,1340,1342],{"path":483,"priority":310},{"path":1339,"priority":486},"references/backends.md",{"path":1341,"priority":486},"references/constraints.md",{"path":1216,"priority":486},{"basePath":1344,"description":1345,"displayName":1346,"installMethods":1347,"rationale":1348,"selectedPaths":1349,"source":316,"sourceLanguage":18,"type":234},"16-prompt-engineering/instructor","Extract structured data from LLM responses with Pydantic validation, retry failed extractions automatically, parse complex JSON with type safety, and stream partial results with Instructor - battle-tested structured output library","instructor",{"claudeCode":12},"SKILL.md frontmatter at 16-prompt-engineering/instructor/SKILL.md",[1350,1351,1352,1354],{"path":483,"priority":310},{"path":1216,"priority":486},{"path":1353,"priority":486},"references/providers.md",{"path":1355,"priority":486},"references/validation.md",{"basePath":1357,"description":1358,"displayName":1359,"installMethods":1360,"rationale":1361,"selectedPaths":1362,"source":316,"sourceLanguage":18,"type":234},"16-prompt-engineering/outlines","Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize inference speed with Outlines - dottxt.ai's structured generation library","outlines",{"claudeCode":12},"SKILL.md frontmatter at 16-prompt-engineering/outlines/SKILL.md",[1363,1364,1365,1366],{"path":483,"priority":310},{"path":1339,"priority":486},{"path":1216,"priority":486},{"path":1367,"priority":486},"references/json_generation.md",{"basePath":1369,"description":1370,"displayName":1371,"installMethods":1372,"rationale":1373,"selectedPaths":1374,"source":316,"sourceLanguage":18,"type":234},"17-observability/langsmith","LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.","langsmith-observability",{"claudeCode":12},"SKILL.md frontmatter at 17-observability/langsmith/SKILL.md",[1375,1376,1377],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1379,"description":1380,"displayName":1381,"installMethods":1382,"rationale":1383,"selectedPaths":1384,"source":316,"sourceLanguage":18,"type":234},"17-observability/phoenix","Open-source AI observability platform for LLM tracing, evaluation, and monitoring. Use when debugging LLM applications with detailed traces, running evaluations on datasets, or monitoring production AI systems with real-time insights.","phoenix-observability",{"claudeCode":12},"SKILL.md frontmatter at 17-observability/phoenix/SKILL.md",[1385,1386,1387],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1389,"description":1390,"displayName":1391,"installMethods":1392,"rationale":1393,"selectedPaths":1394,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/audiocraft","PyTorch library for audio generation including text-to-music (MusicGen) and text-to-sound (AudioGen). Use when you need to generate music from text descriptions, create sound effects, or perform melody-conditioned music generation.","audiocraft-audio-generation",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/audiocraft/SKILL.md",[1395,1396,1397],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1399,"description":1400,"displayName":1401,"installMethods":1402,"rationale":1403,"selectedPaths":1404,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/blip-2","Vision-language pre-training framework bridging frozen image encoders and LLMs. Use when you need image captioning, visual question answering, image-text retrieval, or multimodal chat with state-of-the-art zero-shot performance.","blip-2-vision-language",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/blip-2/SKILL.md",[1405,1406,1407],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1409,"description":1410,"displayName":1411,"installMethods":1412,"rationale":1413,"selectedPaths":1414,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/clip","OpenAI's model connecting vision and language. Enables zero-shot image classification, image-text matching, and cross-modal retrieval. Trained on 400M image-text pairs. Use for image search, content moderation, or vision-language tasks without fine-tuning. Best for general-purpose image understanding.","clip",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/clip/SKILL.md",[1415,1416],{"path":483,"priority":310},{"path":1417,"priority":486},"references/applications.md",{"basePath":1419,"description":1420,"displayName":1421,"installMethods":1422,"rationale":1423,"selectedPaths":1424,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/cosmos-policy","Evaluates NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments. Use when setting up cosmos-policy for robot manipulation evaluation, running headless GPU evaluations with EGL rendering, or profiling inference latency on cluster or local GPU machines.","evaluating-cosmos-policy",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/cosmos-policy/SKILL.md",[1425,1426,1428],{"path":483,"priority":310},{"path":1427,"priority":486},"references/libero-commands.md",{"path":1429,"priority":486},"references/robocasa-commands.md",{"basePath":1431,"description":1432,"displayName":1433,"installMethods":1434,"rationale":1435,"selectedPaths":1436,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/llava","Large Language and Vision Assistant. Enables visual instruction tuning and image-based conversations. Combines CLIP vision encoder with Vicuna/LLaMA language models. Supports multi-turn image chat, visual question answering, and instruction following. Use for vision-language chatbots or image understanding tasks. Best for conversational image analysis.","llava",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/llava/SKILL.md",[1437,1438],{"path":483,"priority":310},{"path":543,"priority":486},{"basePath":1440,"description":1441,"displayName":1442,"installMethods":1443,"rationale":1444,"selectedPaths":1445,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/openpi","Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID, and LIBERO environments. Use when adapting pi0 models to custom datasets, converting JAX checkpoints to PyTorch, running policy inference servers, or debugging norm stats and GPU memory issues.","fine-tuning-serving-openpi",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/openpi/SKILL.md",[1446,1447,1449,1451,1453,1455],{"path":483,"priority":310},{"path":1448,"priority":486},"references/checkpoints-and-env-map.md",{"path":1450,"priority":486},"references/config-recipes.md",{"path":1452,"priority":486},"references/pytorch-gotchas.md",{"path":1454,"priority":486},"references/remote-client-pattern.md",{"path":1456,"priority":486},"references/training-debugging.md",{"basePath":1458,"description":1459,"displayName":1460,"installMethods":1461,"rationale":1462,"selectedPaths":1463,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/openvla-oft","Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditioning on LIBERO simulation and ALOHA real-world setups. Use when reproducing OpenVLA-OFT paper results, training custom VLA action heads (L1 or diffusion), deploying server-client inference for ALOHA, or debugging normalization, LoRA merge, and cross-GPU issues.","fine-tuning-openvla-oft",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/openvla-oft/SKILL.md",[1464,1465,1467,1469,1471],{"path":483,"priority":310},{"path":1466,"priority":486},"references/aloha-workflow.md",{"path":1468,"priority":486},"references/config-troubleshooting.md",{"path":1470,"priority":486},"references/libero-workflow.md",{"path":1472,"priority":486},"references/paper-and-checkpoints.md",{"basePath":1474,"description":1475,"displayName":1476,"installMethods":1477,"rationale":1478,"selectedPaths":1479,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/segment-anything","Foundation model for image segmentation with zero-shot transfer. Use when you need to segment any object in images using points, boxes, or masks as prompts, or automatically generate all object masks in an image.","segment-anything-model",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/segment-anything/SKILL.md",[1480,1481,1482],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1484,"description":1485,"displayName":1486,"installMethods":1487,"rationale":1488,"selectedPaths":1489,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/stable-diffusion","State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.","stable-diffusion-image-generation",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/stable-diffusion/SKILL.md",[1490,1491,1492],{"path":483,"priority":310},{"path":638,"priority":486},{"path":640,"priority":486},{"basePath":1494,"description":1495,"displayName":1496,"installMethods":1497,"rationale":1498,"selectedPaths":1499,"source":316,"sourceLanguage":18,"type":234},"18-multimodal/whisper","OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual ASR.","whisper",{"claudeCode":12},"SKILL.md frontmatter at 18-multimodal/whisper/SKILL.md",[1500,1501],{"path":483,"priority":310},{"path":1502,"priority":486},"references/languages.md",{"basePath":1504,"description":1505,"displayName":1506,"installMethods":1507,"rationale":1508,"selectedPaths":1509,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/knowledge-distillation","Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.","knowledge-distillation",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/knowledge-distillation/SKILL.md",[1510,1511],{"path":483,"priority":310},{"path":1512,"priority":486},"references/minillm.md",{"basePath":1514,"description":1515,"displayName":1516,"installMethods":1517,"rationale":1518,"selectedPaths":1519,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/long-context","Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+ tokens), extending pre-trained models beyond original context limits, or implementing efficient positional encodings. Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs.","long-context",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/long-context/SKILL.md",[1520,1521,1523,1525],{"path":483,"priority":310},{"path":1522,"priority":486},"references/extension_methods.md",{"path":1524,"priority":486},"references/fine_tuning.md",{"path":1526,"priority":486},"references/rope.md",{"basePath":1528,"description":1529,"displayName":1530,"installMethods":1531,"rationale":1532,"selectedPaths":1533,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/model-merging","Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.","model-merging",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/model-merging/SKILL.md",[1534,1535,1537,1538],{"path":483,"priority":310},{"path":1536,"priority":486},"references/evaluation.md",{"path":1216,"priority":486},{"path":1539,"priority":486},"references/methods.md",{"basePath":1541,"description":1542,"displayName":1543,"installMethods":1544,"rationale":1545,"selectedPaths":1546,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/model-pruning","Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% sparsity with minimal accuracy loss, or enabling faster inference on hardware accelerators. Covers unstructured pruning, structured pruning, N:M sparsity, magnitude pruning, and one-shot methods.","model-pruning",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/model-pruning/SKILL.md",[1547,1548],{"path":483,"priority":310},{"path":1549,"priority":486},"references/wanda.md",{"basePath":1551,"description":1552,"displayName":1553,"installMethods":1554,"rationale":1555,"selectedPaths":1556,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/moe-training","Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.","moe-training",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/moe-training/SKILL.md",[1557,1558,1560,1562],{"path":483,"priority":310},{"path":1559,"priority":486},"references/architectures.md",{"path":1561,"priority":486},"references/inference.md",{"path":543,"priority":486},{"basePath":1564,"description":1565,"displayName":1566,"installMethods":1567,"rationale":1568,"selectedPaths":1569,"source":316,"sourceLanguage":18,"type":234},"19-emerging-techniques/speculative-decoding","Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6× speedup), reducing latency for real-time applications, or deploying models with limited compute. Covers draft models, tree-based attention, Jacobi iteration, parallel token generation, and production deployment strategies.","speculative-decoding",{"claudeCode":12},"SKILL.md frontmatter at 19-emerging-techniques/speculative-decoding/SKILL.md",[1570,1571,1573],{"path":483,"priority":310},{"path":1572,"priority":486},"references/lookahead.md",{"path":1574,"priority":486},"references/medusa.md",{"basePath":1576,"description":1577,"displayName":1578,"installMethods":1579,"rationale":1580,"selectedPaths":1581,"source":316,"sourceLanguage":18,"type":234},"20-ml-paper-writing/academic-plotting","Generates publication-quality figures for ML papers from research context. Given a paper section or description, extracts system components and relationships to generate architecture diagrams via Gemini. Given experiment results or data, auto-selects chart type and generates data-driven figures via matplotlib/seaborn. Use when creating any figure for a conference paper.","academic-plotting",{"claudeCode":12},"SKILL.md frontmatter at 20-ml-paper-writing/academic-plotting/SKILL.md",[1582,1583,1585,1587],{"path":483,"priority":310},{"path":1584,"priority":486},"references/data-visualization.md",{"path":1586,"priority":486},"references/diagram-generation.md",{"path":1588,"priority":486},"references/style-guide.md",{"basePath":1590,"description":1591,"displayName":457,"installMethods":1592,"rationale":1593,"selectedPaths":1594,"source":316,"sourceLanguage":18,"type":234},"20-ml-paper-writing/ml-paper-writing","Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. For systems venues (OSDI, NSDI, ASPLOS, SOSP), use systems-paper-writing instead.",{"claudeCode":12},"SKILL.md frontmatter at 20-ml-paper-writing/ml-paper-writing/SKILL.md",[1595,1596,1598,1600,1602,1604,1606,1608,1610,1612,1614,1616,1618,1620,1622,1624,1626,1628,1630,1632,1634,1636,1638,1640,1642,1644,1646,1648,1650,1652,1654,1656,1658,1660,1662,1664,1666,1668,1670,1672,1674,1676,1678,1680,1682,1684,1686,1688,1690,1692,1694],{"path":483,"priority":310},{"path":1597,"priority":486},"references/checklists.md",{"path":1599,"priority":486},"references/citation-workflow.md",{"path":1601,"priority":486},"references/reviewer-guidelines.md",{"path":1603,"priority":486},"references/sources.md",{"path":1605,"priority":486},"references/writing-guide.md",{"path":1607,"priority":493},"templates/README.md",{"path":1609,"priority":493},"templates/aaai2026/README.md",{"path":1611,"priority":493},"templates/aaai2026/aaai2026-unified-supp.tex",{"path":1613,"priority":493},"templates/aaai2026/aaai2026-unified-template.tex",{"path":1615,"priority":493},"templates/aaai2026/aaai2026.bib",{"path":1617,"priority":493},"templates/aaai2026/aaai2026.bst",{"path":1619,"priority":493},"templates/aaai2026/aaai2026.sty",{"path":1621,"priority":493},"templates/acl/README.md",{"path":1623,"priority":493},"templates/acl/acl.sty",{"path":1625,"priority":493},"templates/acl/acl_latex.tex",{"path":1627,"priority":493},"templates/acl/acl_lualatex.tex",{"path":1629,"priority":493},"templates/acl/acl_natbib.bst",{"path":1631,"priority":493},"templates/acl/anthology.bib.txt",{"path":1633,"priority":493},"templates/acl/custom.bib",{"path":1635,"priority":493},"templates/acl/formatting.md",{"path":1637,"priority":493},"templates/colm2025/README.md",{"path":1639,"priority":493},"templates/colm2025/colm2025_conference.bib",{"path":1641,"priority":493},"templates/colm2025/colm2025_conference.bst",{"path":1643,"priority":493},"templates/colm2025/colm2025_conference.pdf",{"path":1645,"priority":493},"templates/colm2025/colm2025_conference.sty",{"path":1647,"priority":493},"templates/colm2025/colm2025_conference.tex",{"path":1649,"priority":493},"templates/colm2025/fancyhdr.sty",{"path":1651,"priority":493},"templates/colm2025/math_commands.tex",{"path":1653,"priority":493},"templates/colm2025/natbib.sty",{"path":1655,"priority":493},"templates/iclr2026/fancyhdr.sty",{"path":1657,"priority":493},"templates/iclr2026/iclr2026_conference.bib",{"path":1659,"priority":493},"templates/iclr2026/iclr2026_conference.bst",{"path":1661,"priority":493},"templates/iclr2026/iclr2026_conference.pdf",{"path":1663,"priority":493},"templates/iclr2026/iclr2026_conference.sty",{"path":1665,"priority":493},"templates/iclr2026/iclr2026_conference.tex",{"path":1667,"priority":493},"templates/iclr2026/math_commands.tex",{"path":1669,"priority":493},"templates/iclr2026/natbib.sty",{"path":1671,"priority":493},"templates/icml2026/algorithm.sty",{"path":1673,"priority":493},"templates/icml2026/algorithmic.sty",{"path":1675,"priority":493},"templates/icml2026/example_paper.bib",{"path":1677,"priority":493},"templates/icml2026/example_paper.pdf",{"path":1679,"priority":493},"templates/icml2026/example_paper.tex",{"path":1681,"priority":493},"templates/icml2026/fancyhdr.sty",{"path":1683,"priority":493},"templates/icml2026/icml2026.bst",{"path":1685,"priority":493},"templates/icml2026/icml2026.sty",{"path":1687,"priority":493},"templates/icml2026/icml_numpapers.pdf",{"path":1689,"priority":493},"templates/neurips2025/Makefile",{"path":1691,"priority":493},"templates/neurips2025/extra_pkgs.tex",{"path":1693,"priority":493},"templates/neurips2025/main.tex",{"path":1695,"priority":493},"templates/neurips2025/neurips.sty",{"basePath":1697,"description":1698,"displayName":1699,"installMethods":1700,"rationale":1701,"selectedPaths":1702,"source":316,"sourceLanguage":18,"type":234},"20-ml-paper-writing/presenting-conference-talks","Generates conference presentation slides (Beamer LaTeX PDF and editable PPTX) from a compiled paper with speaker notes and talk script. Use when preparing oral talks, spotlight presentations, or invited talks for ML and systems conferences.","presenting-conference-talks",{"claudeCode":12},"SKILL.md frontmatter at 20-ml-paper-writing/presenting-conference-talks/SKILL.md",[1703,1704],{"path":483,"priority":310},{"path":1705,"priority":486},"references/slide-templates.md",{"basePath":1707,"description":1708,"displayName":1709,"installMethods":1710,"rationale":1711,"selectedPaths":1712,"source":316,"sourceLanguage":18,"type":234},"20-ml-paper-writing/systems-paper-writing","Comprehensive guide for writing systems papers targeting OSDI, SOSP, ASPLOS, NSDI, and EuroSys. Provides paragraph-level structural blueprints, writing patterns, venue-specific checklists, reviewer guidelines, LaTeX templates, and conference deadlines. Use this skill for all systems conference paper writing.","systems-paper-writing",{"claudeCode":12},"SKILL.md frontmatter at 20-ml-paper-writing/systems-paper-writing/SKILL.md",[1713,1714,1716,1717,1719,1721,1723,1725,1727,1729,1731,1733,1735,1737,1739,1741],{"path":483,"priority":310},{"path":1715,"priority":486},"references/checklist.md",{"path":1601,"priority":486},{"path":1718,"priority":486},"references/section-blueprints.md",{"path":1720,"priority":486},"references/systems-conferences.md",{"path":1722,"priority":486},"references/writing-patterns.md",{"path":1724,"priority":493},"templates/asplos2027/main.tex",{"path":1726,"priority":493},"templates/asplos2027/references.bib",{"path":1728,"priority":493},"templates/nsdi2027/main.tex",{"path":1730,"priority":493},"templates/nsdi2027/references.bib",{"path":1732,"priority":493},"templates/nsdi2027/usenix-2020-09.sty",{"path":1734,"priority":493},"templates/osdi2026/main.tex",{"path":1736,"priority":493},"templates/osdi2026/references.bib",{"path":1738,"priority":493},"templates/osdi2026/usenix-2020-09.sty",{"path":1740,"priority":493},"templates/sosp2026/main.tex",{"path":1742,"priority":493},"templates/sosp2026/references.bib",{"basePath":1744,"description":1745,"displayName":1746,"installMethods":1747,"rationale":1748,"selectedPaths":1749,"source":316,"sourceLanguage":18,"type":234},"21-research-ideation/brainstorming-research-ideas","Guides researchers through structured ideation frameworks to discover high-impact research directions. Use when exploring new problem spaces, pivoting between projects, or seeking novel angles on existing work.","brainstorming-research-ideas",{"claudeCode":12},"SKILL.md frontmatter at 21-research-ideation/brainstorming-research-ideas/SKILL.md",[1750],{"path":483,"priority":310},{"basePath":1752,"description":1753,"displayName":1754,"installMethods":1755,"rationale":1756,"selectedPaths":1757,"source":316,"sourceLanguage":18,"type":234},"21-research-ideation/creative-thinking-for-research","Applies cognitive science frameworks for creative thinking to CS and AI research ideation. Use when seeking genuinely novel research directions by leveraging combinatorial creativity, analogical reasoning, constraint manipulation, and other empirically grounded creative strategies.","creative-thinking-for-research",{"claudeCode":12},"SKILL.md frontmatter at 21-research-ideation/creative-thinking-for-research/SKILL.md",[1758],{"path":483,"priority":310},{"basePath":1760,"description":1761,"displayName":1762,"installMethods":1763,"rationale":1764,"selectedPaths":1765,"source":316,"sourceLanguage":18,"type":234},"22-agent-native-research-artifact/compiler","Compiles any research input — PDF papers, GitHub repositories, experiment logs, code directories, or raw notes — into a complete Agent-Native Research Artifact (ARA) with cognitive layer (claims, concepts, heuristics), physical layer (configs, code stubs), exploration graph, and grounded evidence. Use when ingesting a paper or codebase into a structured, machine-executable knowledge package, building an ARA from scratch, or converting research outputs into a falsifiable, agent-traversable form.","ara-compiler",{"claudeCode":12},"SKILL.md frontmatter at 22-agent-native-research-artifact/compiler/SKILL.md",[1766,1767,1769,1771],{"path":483,"priority":310},{"path":1768,"priority":486},"references/ara-schema.md",{"path":1770,"priority":486},"references/exploration-tree-spec.md",{"path":1772,"priority":486},"references/validation-checklist.md",{"basePath":1774,"description":1775,"displayName":1776,"installMethods":1777,"rationale":1778,"selectedPaths":1779,"source":316,"sourceLanguage":18,"type":234},"22-agent-native-research-artifact/research-manager","Records research provenance as a post-task epilogue, scanning conversation history at the end of a coding or research session to extract decisions, experiments, dead ends, claims, heuristics, and pivots, and writing them into the ara/ directory with user-vs-AI provenance tags. Use as a session epilogue — never during execution — to maintain a faithful, auditable trace of how a research project actually evolved.","ara-research-manager",{"claudeCode":12},"SKILL.md frontmatter at 22-agent-native-research-artifact/research-manager/SKILL.md",[1780,1781,1783,1785],{"path":483,"priority":310},{"path":1782,"priority":486},"references/event-taxonomy.md",{"path":1784,"priority":486},"references/provenance-tags.md",{"path":1786,"priority":486},"references/session-protocol.md",{"basePath":1788,"description":1789,"displayName":1790,"installMethods":1791,"rationale":1792,"selectedPaths":1793,"source":316,"sourceLanguage":18,"type":234},"22-agent-native-research-artifact/rigor-reviewer","Performs ARA Seal Level 2 semantic epistemic review on Agent-Native Research Artifacts, scoring six dimensions (evidence relevance, falsifiability, scope calibration, argument coherence, exploration integrity, methodological rigor) and producing a constructive, severity-ranked report with a Strong Accept-to-Reject recommendation. Use after Level 1 structural validation passes, when an ARA needs an objective epistemic critique before publication or release.","ara-rigor-reviewer",{"claudeCode":12},"SKILL.md frontmatter at 22-agent-native-research-artifact/rigor-reviewer/SKILL.md",[1794,1795],{"path":483,"priority":310},{"path":1796,"priority":486},"references/review-dimensions.md",{"basePath":1798,"description":1799,"displayName":1800,"installMethods":1801,"license":224,"rationale":1802,"selectedPaths":1803,"source":316,"sourceLanguage":18,"type":1811},"packages/ai-research-skills","Install AI research engineering skills to your coding agents (Claude Code, OpenCode, Cursor, Gemini CLI, Hermes Agent, and more)","@orchestra-research/ai-research-skills",{"npm":1800},"cli ecosystem detected at packages/ai-research-skills",[1804,1806,1807,1809],{"path":1805,"priority":310},"package.json",{"path":312,"priority":310},{"path":1808,"priority":486},"bin/cli.js",{"path":1810,"priority":493},"src/index.js","cli",{"sources":1813},[1814],"manual",{"npmPackage":268},{"closedIssues90d":216,"description":1817,"forks":217,"homepage":1818,"license":224,"openIssues90d":219,"pushedAt":220,"readmeSize":214,"stars":221,"topics":1819},"Comprehensive open-source library of AI research and engineering skills for any AI model. Package the skills and your claude code/codex/gemini agent will be an AI research agent with full horsepower. Maintained by Orchestra Research.","http://orchestra-research.com",[1820,261,1821,1822,1823,1824,1825,1826,1827,1828,1829,1830,1831,1832],"ai","claude","claude-code","claude-skills","codex","gemini","gpt-5","grpo","huggingface","machine-leanring","megatron","skills","vllm",{"downloads":8},{"classifiedAt":1835,"discoverAt":1836,"extractAt":1837,"githubAt":1837,"npmAt":1838,"updatedAt":1835},1778695115942,1778695107142,1778695112108,1778695113836,[198,199,197,195,200,196],{"evaluatedAt":228,"extractAt":277,"updatedAt":228},[],[1843,1870,1894,1923,1941,1965],{"_creationTime":1844,"_id":1845,"community":1846,"display":1847,"identity":1853,"providers":1856,"relations":1864,"tags":1866,"workflow":1867},1778695021936.5552,"k172b680yjc5dekp4dw1ny02q186nnyk",{"reviewCount":8},{"description":1848,"installMethods":1849,"name":1851,"sourceUrl":1852},"Deploy applications and infrastructure to Cloudflare using Workers, Pages, and related platform services. Use when the user asks to deploy, host, publish, or set up a project on Cloudflare.",{"claudeCode":1850},"openai/skills","cloudflare-deploy","https://github.com/openai/skills",{"basePath":1854,"githubOwner":1855,"githubRepo":1831,"locale":18,"slug":1851,"type":234},"skills/.curated/cloudflare-deploy","openai",{"evaluate":1857,"extract":1863},{"promptVersionExtension":188,"promptVersionScoring":189,"score":259,"tags":1858,"targetMarket":201,"tier":202},[198,199,1859,1860,1861,195,1862],"cloudflare","workers","pages","developer-tools",{"commitSha":266},{"repoId":1865},"kd75n2zj3yh472p25zffgycved86mnpx",[198,1859,199,1862,195,1861,1860],{"evaluatedAt":1868,"extractAt":1869,"updatedAt":1868},1778695077986,1778695021936,{"_creationTime":1871,"_id":1872,"community":1873,"display":1874,"identity":1878,"providers":1881,"relations":1890,"tags":1891,"workflow":1892},1778695021936.561,"k1703ngx380xc9697a003dtgy186m90t",{"reviewCount":8},{"description":1875,"installMethods":1876,"name":1877,"sourceUrl":1852},"Deploy applications to Render by analyzing codebases, generating render.yaml Blueprints, and providing Dashboard deeplinks. Use when the user wants to deploy, host, publish, or set up their application on Render's cloud platform.",{"claudeCode":1850},"Render Deploy",{"basePath":1879,"githubOwner":1855,"githubRepo":1831,"locale":18,"slug":1880,"type":234},"skills/.curated/render-deploy","render-deploy",{"evaluate":1882,"extract":1888},{"promptVersionExtension":188,"promptVersionScoring":189,"score":259,"tags":1883,"targetMarket":201,"tier":202},[1884,199,1885,198,1886,1887],"render","iac","devops","ci-cd",{"commitSha":266,"license":1889},"Apache-2.0",{"repoId":1865},[1887,198,199,1886,1885,1884],{"evaluatedAt":1893,"extractAt":1869,"updatedAt":1893},1778695535578,{"_creationTime":1895,"_id":1896,"community":1897,"display":1898,"identity":1904,"providers":1908,"relations":1916,"tags":1919,"workflow":1920},1778699018122.793,"k1734tnjy7pr4we3yqg8rd5zyh86njgt",{"reviewCount":8},{"description":1899,"installMethods":1900,"name":1902,"sourceUrl":1903},"Optimize cloud costs across AWS, Azure, GCP, and OCI through resource rightsizing, tagging strategies, reserved instances, and spending analysis. Use when reducing cloud expenses, analyzing infrastructure costs, or implementing cost governance policies.",{"claudeCode":1901},"wshobson/agents","Cost Optimization","https://github.com/wshobson/agents",{"basePath":1905,"githubOwner":1906,"githubRepo":408,"locale":18,"slug":1907,"type":234},"plugins/cloud-infrastructure/skills/cost-optimization","wshobson","cost-optimization",{"evaluate":1909,"extract":1915},{"promptVersionExtension":188,"promptVersionScoring":189,"score":280,"tags":1910,"targetMarket":201,"tier":202},[198,1907,1911,1912,1913,1914,195],"aws","azure","gcp","oci",{"commitSha":266,"license":224},{"parentExtensionId":1917,"repoId":1918},"k177fdvvq05pdbpee0rz61me2h86mahe","kd74de64zj0axtg5b8t7eqqe2x86nske",[1911,1912,198,1907,1913,195,1914],{"evaluatedAt":1921,"extractAt":1922,"updatedAt":1921},1778700805333,1778699018122,{"_creationTime":1924,"_id":1925,"community":1926,"display":1927,"identity":1929,"providers":1931,"relations":1937,"tags":1938,"workflow":1939},1778695116697.1848,"k17cn5wpxpdj7qzt0hbg76ft3h86ma5f",{"reviewCount":8},{"description":960,"installMethods":1928,"name":961,"sourceUrl":14},{"claudeCode":12},{"basePath":959,"githubOwner":231,"githubRepo":232,"locale":18,"slug":1930,"type":234},"skypilot",{"evaluate":1932,"extract":1936},{"promptVersionExtension":188,"promptVersionScoring":189,"score":280,"tags":1933,"targetMarket":201,"tier":202},[195,1934,1935,197,1907,1930,200],"multi-cloud","orchestration",{"commitSha":266},{"parentExtensionId":237,"repoId":273},[1907,197,195,200,1934,1935,1930],{"evaluatedAt":1940,"extractAt":277,"updatedAt":1940},1778696405741,{"_creationTime":1942,"_id":1943,"community":1944,"display":1945,"identity":1949,"providers":1954,"relations":1959,"tags":1961,"workflow":1962},1778685991755.714,"k17axjt1t8jb33pgs6wy1htxed86mp2k",{"reviewCount":8},{"description":10,"installMethods":1946,"name":13,"sourceUrl":1948},{"claudeCode":1947},"davila7/claude-code-templates","https://github.com/davila7/claude-code-templates",{"basePath":1950,"githubOwner":1951,"githubRepo":1952,"locale":18,"slug":1953,"type":234},"cli-tool/components/skills/ai-research/infrastructure-modal","davila7","claude-code-templates","infrastructure-modal",{"evaluate":1955,"extract":1958},{"promptVersionExtension":188,"promptVersionScoring":189,"score":280,"tags":1956,"targetMarket":201,"tier":1957},[195,196,197,198,199,200],"community",{"commitSha":266},{"repoId":1960},"kd71fzn4s7r0269fkw47wt670n86ndz0",[198,199,197,195,200,196],{"evaluatedAt":1963,"extractAt":1964,"updatedAt":1963},1778687424295,1778685991755,{"_creationTime":1966,"_id":1967,"community":1968,"display":1969,"identity":1975,"providers":1981,"relations":1987,"tags":1990,"workflow":1991},1778686637545.601,"k17cgy0grzvan5bp2b4emm1s3s86nkcb",{"reviewCount":8},{"description":1970,"installMethods":1971,"name":1973,"sourceUrl":1974},"Cloud-GPU-Verarbeitung über RunPod Serverless. Verwenden Sie dies beim Einrichten von RunPod-Endpunkten, beim Bereitstellen von Docker-Images, beim Verwalten von GPU-Ressourcen, beim Beheben von Endpunktproblemen oder beim Verstehen von Kosten. Beinhaltet alle 5 Toolkit-Images (qwen-edit, realesrgan, propainter, sadtalker, qwen3-tts).",{"claudeCode":1972},"digitalsamba/claude-code-video-toolkit","RunPod Cloud GPU","https://github.com/digitalsamba/claude-code-video-toolkit",{"basePath":1976,"githubOwner":1977,"githubRepo":1978,"locale":1979,"slug":1980,"type":234},".claude/skills/runpod","digitalsamba","claude-code-video-toolkit","de","runpod",{"evaluate":1982,"extract":1986},{"promptVersionExtension":188,"promptVersionScoring":189,"score":280,"tags":1983,"targetMarket":201,"tier":202},[198,197,1980,196,1984,1820,1985],"docker","machine-learning",{"commitSha":266,"license":224},{"repoId":1988,"translatedFrom":1989},"kd70r97eght58pp9f1x8scdagd86n32q","k17ezctet9954yj232ppmq7b5d86mrx7",[1820,198,1984,197,1985,1980,196],{"evaluatedAt":1992,"extractAt":1993,"updatedAt":1994},1778686491299,1778686219732,1778686637545]