Skills
18,283 skills, most installed first.
Browse
simpo-trainingorchestra-researchSimple 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.skypilot-multi-cloud-orchestrationorchestra-researchMulti-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.slime-rl-trainingorchestra-researchProvides 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.sparse-autoencoder-trainingorchestra-researchProvides 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.speculative-decodingorchestra-researchAccelerate 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.stable-diffusion-image-generationorchestra-researchState-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.systems-paper-writingorchestra-researchComprehensive 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.tensorboardorchestra-researchVisualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkittensorrt-llmorchestra-researchOptimizes 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.torchforge-rl-trainingorchestra-researchProvides 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.training-llms-megatronorchestra-researchTrains 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.transformer-lens-interpretabilityorchestra-researchProvides 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.unslothorchestra-researchExpert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimizationverl-rl-trainingorchestra-researchProvides 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.weights-and-biasesorchestra-researchTrack ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platformwhisperorchestra-researchOpenAI'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.conformetry-configureorganizzoliniAdd or change a conformetry generator, and write the template it renders. Use when a needed generator does not exist yet; when editing a conformetry configuration file; when writing or changing files under a template directory; when choosing placeholders or naming-case variants; when a generated file came out with an empty value or an odd path segment; or when a conformetry command refuses to run because the emitted generator plugin is out of date.conformetry-generateorganizzoliniScaffold code with a conformetry generator instead of writing it by hand. Use when adding a module, component, service, application, or package to a workspace that has a conformetry configuration; when deciding whether a generator exists for the shape you are about to create; when a generator run overwrote or misplaced files; or when a conformance run reports a file that a generator should have produced. Covers both the Nx plugin and the conformetry command-line host, whose input contracts diffeconformetry-validateorganizzoliniRun a conformetry conformance check and act on what it reports. Use when a conformetry-validate target fails; when a composite lint run reports conformance differences; when a report names a missing file, a missing declaration, a missing comment, an ambiguous instance, or an unmatched instance; when deciding whether to edit an instance or regenerate it; or before claiming generated code is finished. Covers both the Nx plugin and the conformetry command-line host.creative-writing-craftorganvm-iv-taxisCraft compelling fiction and creative nonfiction with attention to structure, voice, prose style, and revision. Supports short stories, novel chapters, essays, and hybrid forms. Triggers on creative writing, fiction writing, story craft, prose style, or literary technique requests.orvalorval-labsGenerate type-safe API clients, TanStack Query/SWR hooks, Zod schemas, MSW mocks, Hono server handlers, MCP servers, and SolidStart actions from OpenAPI specs using Orval. Covers all clients (React/Vue/Svelte/Solid/Angular Query, Fetch, Axios), custom HTTP mutators, authentication patterns, NDJSON streaming, programmatic API, and advanced configuration.odaiorziz以“成事而不妄为”为准则处理通用任务。用户点名 odai，或任务模糊、复杂、高风险、前提可疑、易漏项时使用；简单任务直做。higgsfieldosidemediaUse this skill whenever the user asks anything about Higgsfield AI — writing or refining video/image prompts, choosing a model (Kling, Veo, Wan, Seedance, Minimax Hailuo, DoP, Soul, Nano Banana, Seedream, Flux, GPT Image, etc.), camera controls, named motion presets, Soul ID character consistency, Cinema Studio 2.5/3.0, Vibe Motion, troubleshooting failed generations, credit optimization, Photodump, or any mention of higgsfield.ai. Also trigger on generic "write me a video prompt" or "make me anreact-hook-form-zodovachieverBuild type-safe validated forms in React using React Hook Form and Zod schema validation. Single schema works on both client and server for DRY validation with full TypeScript type inference via z.infer. Use when: building forms with validation, integrating shadcn/ui Form components, implementing multi-step wizards, handling dynamic field arrays with useFieldArray, or fixing uncontrolled to controlled warnings, resolver errors, async validation issues.security-auditorovachieverContinuous security vulnerability scanning for OWASP Top 10, common vulnerabilities, and insecure patterns. Use when reviewing code, before deployments, or on file changes. Scans for SQL injection, XSS, secrets exposure, auth issues. Triggers on file changes, security mentions, deployment prep.planowainlewisTurns an approved design or decided brief into ordered tasks for separate agent runs. Use for implementation tasks, tracker tickets, or useful milestones. Do not use for one coding task or its short execution outline.reviewowainlewisUses a fresh agent to review an implementation change without editing it. Checks behavior, security, regressions, complexity, tests, docs, and missing proof. Use for code, PR, diff, security, second-opinion, or pre-merge reviews.web-security-reviewowaspReview web applications against the OWASP Top 10 for Web Applications (2021). Use when auditing web apps, reviewing server-side code, or assessing web frameworks for the classic OWASP Top 10 risks including injection, broken auth, and XSS.system-prompt-structureowl-listenerAnatomy of effective system prompts — role, context, constraints, format.business-designowl-listenerRead financials, map competitive landscapes, and argue design decisions in the language of value. Use when defending design to commercial stakeholders. For the live negotiation itself, use `design-negotiation` (designer-toolkit).conversational-uxowl-listenerDesign voice and conversational interfaces — dialog flows, error recovery, and persona. Use when the interface speaks and listens rather than being tapped. For graphical input collection, use `form-design`.critique-affordanceowl-listenerCritique a rendered screen's affordances — what looks clickable, state visibility, CTA clarity, and action discoverability. Use when reviewing an existing screen. For sizing and positioning targets in new work, use `fitts-law` (interaction-design).interfaces-that-feelowl-listenerApply an emotional resonance lens to a UI that is technically correct but flat, prescribing changes at the copy, motion, and interaction layer. Use when a design tests fine but lands cold. For the polish-perception argument, use `aesthetic-usability` (ui-design).jakobs-lawowl-listenerApply Jakob's Law — users expect your product to work like the others they already use. Use when deciding whether to innovate on a familiar pattern. For OS-mandated conventions specifically, use `platform-conventions` (ui-design).law-of-closureowl-listenerApply the Law of Closure — the eye completes implied shapes from partial forms. Use when reducing visual weight by dropping borders or letting negative space suggest structure. For explicit containers, use `law-of-common-region`.law-of-continuityowl-listenerApply the Law of Continuity — the eye follows alignment and unbroken paths. Use when sequencing steps, aligning content, or designing carousels and timelines. For grouping rather than sequencing, use `law-of-proximity`.law-of-figure-groundowl-listenerApply the Law of Figure-Ground — establish which layer is foreground and actionable versus background. Use when designing modals, overlays, and depth. For emphasising one element among peers, use `von-restorff-effect`.law-of-similarityowl-listenerApply the Law of Similarity — shared colour, shape, or size signals that elements belong to one category. Use when signalling relationships across distance. For grouping by position, use `law-of-proximity`.peak-end-ruleowl-listenerApply the Peak-End Rule — a flow is remembered by its most intense moment and its last. Use when designing completion, celebration, or cancellation moments. For sustaining engagement mid-flow, use `zeigarnik-effect`.platform-conventionsowl-listenerDesign to iOS and Android conventions — what each OS mandates, where they diverge, and when to unify. Use when shipping native apps. For breakpoint adaptation use `responsive-design`; for matching competitor patterns use `jakobs-law` (interaction-design).serial-position-effectowl-listenerApply the Serial Position Effect — first and last items in a sequence are recalled best. Use when ordering menus, lists, and steps. For emphasising one item regardless of its position, use `von-restorff-effect` (ui-design).teslers-lawowl-listenerApply Tesler's Law — every process has irreducible complexity that someone must absorb. Use when deciding whether the product or the user carries it. For reducing apparent choice, use `hicks-law`.zeigarnik-effectowl-listenerApply the Zeigarnik Effect — incomplete tasks stay mentally active. Use when designing progress indicators, saved drafts, and return hooks. For the emotional shape of the ending, use `peak-end-rule`.migrate-oxfmtoxc-projectGuide for migrating a project from Prettier or Biome to Oxfmt. Use when asked to migrate, convert, or switch a JavaScript/TypeScript project's formatter from Prettier or Biome to Oxfmt.migrate-oxlintoxc-projectGuide for migrating a project from ESLint to Oxlint. Use when asked to migrate, convert, or switch a JavaScript/TypeScript project's linter from ESLint to Oxlint.rev-ios-dumpp4nda0sDump decrypted iOS app binaries (砸壳) from jailbroken devices using frida-ios-dump. Activate when the user wants to decrypt an iOS app, dump an IPA from a device, or extract a decrypted Mach-O binary for reverse engineering.efecto-graphic-designpablostanleyDesign graphic assets with Efecto — presentations, pitch decks, event posters, email headers, blog images, open graph cards, business cards, resumes, menus, infographics, invitations, newsletters, and documents. Use when asked to "design a poster", "create a pitch deck", "make a presentation", "design a business card", or any graphic design task. Requires Efecto MCP server.efecto-social-mediapablostanleyDesign social media assets with Efecto — Instagram posts, carousels, stories, YouTube thumbnails, TikTok covers, Twitter/X images, LinkedIn slides, Pinterest pins, and Facebook graphics. Use when asked to "design a post", "create a carousel", "make a thumbnail", "design social media", or any social content task. Requires Efecto MCP server.efecto-web-designpablostanleyDesign web pages and app UIs with Efecto — create sessions, build layouts with JSX and Tailwind CSS, manage artboards, and push real-time design changes via MCP tools. Use when asked to "design a page", "build a landing page", "create a website", "design a dashboard", "make a UI", or any visual design task. Requires Efecto MCP server.paddleocr-doc-parsingpaddlepaddleUse this skill to extract structured Markdown/JSON from PDFs and document images—tables with cell-level precision, formulas as LaTeX, figures, seals, charts, headers/footers, multi-column layout and correct reading order. Trigger terms: 文档解析, 版面分析, 版面还原, 表格提取, 公式识别, 多栏排版, 扫描件结构化, 发票, 财报, 复杂 PDF, PDF转Markdown, 图表, 阅读顺序; reading order, formula, LaTeX, layout parsing, structure extraction, PP-StructureV3, PaddleOCR-VL.