Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 79 404 files from 1 741 authors, of which 61 763 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
> Guidelines for writing and running tests in the Agent Framework Python codebase. Use this when creating, modifying, or running tests.
> Coding standards, conventions, and patterns for developing Python code in the Agent Framework repository. Use this when writing or modifying Python source files in the python/ directory.
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded project_endpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
> Step-by-step process for validating a Python Foundry hosted agent sample (under python/samples/04-hosting/foundry-hosted-agents/) end to end — running it locally (native runtime and `azd ai agent run`) and after deploying it to an Azure AI Foundry project with `azd`. Use this when asked to validate a hosted agent sample.
> Guidance for package and feature lifecycle in the Agent Framework Python codebase, including stage meanings, feature-stage decorators, feature enums, and how to move APIs from one stage to the next.
Use this skill to fuzz open source JVM projects (Java, Kotlin, Scala, etc.) using Jazzer.
Use this skill to fuzz open source Python software projects using Atheris.
Use this skill to fuzz open source Go software projects.
Use this skill to fuzz open source C/C++ software projects.
Use this skill to fuzz open source Rust software projects.
Use this skill to interact with the OSS-Fuzz infrastructure.
Reproduce a public GitHub issue or PR in the Mapbox GL JS repo as a minimal focused debug page under `./debug/`. Trigger when the user pastes a GitHub issue/PR URL (github.com/mapbox/mapbox-gl-js/issues/N or /pull/N), or says "repro this issue", "reproduce #N", "make a repro page for", "debug page for issue", "recreate this bug", "build a minimal repro", or provides a bare issue number like `#12345` in the context of investigating a bug. Use this skill whenever the user wants to investigate a bug report, regression, or reported behavior — even if they don't say the word "repro" — since a working debug page is almost always the first step before fixing.
Add or extend Redis commands in the Jedis client API — a new core command, a family of new commands, an extension to an existing command's options, or a module command (Search/TimeSeries/JSON/Bloom). Gathers evidence (Redis server PR, HLD document), plans the full implementation matrix in plan mode, then implements with unit and integration tests following Jedis maintainer conventions.
Generate a clear, concise GitHub PR title and description from the diff between two local git branches, and save it to prDescription.md in the repo root. Use this whenever the user asks to write, generate, draft, or update a PR description or PR title from local branch changes — including phrasing like "summarize this diff into a PR description," "write a PR description for my current branch," "create a PR title and description," or any request to compare a base and target branch for PR purposes. Trigger even if the user doesn't name specific branches; this skill knows how to default them.
Development workflows for the playwright-cli repository. Use when the user asks about rolling dependencies, releasing, or other repo maintenance tasks.
Audits and optimizes Convex application performance across hot-path reads, write contention, subscription cost, and function limits. Use this skill when a Convex feature is slow or expensive, npx convex insights shows high bytes or documents read, OCC conflict errors or mutation retries appear, subscriptions or UI updates are costly, functions hit execution or transaction limits, or the user mentions performance, latency, read amplification, or invalidation problems in a Convex app.
Designs and builds Convex components with isolated tables, clear boundaries, and app-facing wrappers. Use this skill when creating a new Convex component, extracting reusable backend logic into a component, building a third-party integration that owns its own tables, packaging Convex functionality for reuse, or when the user mentions defineComponent, app.use, ComponentApi, ctx.runQuery/runMutation across component boundaries, or wants to separate concerns into isolated Convex modules.
Initializes a new Convex project from scratch or adds Convex to an existing app. Use this skill when starting a new project with Convex, scaffolding with npm create convex@latest, adding Convex to an existing React, Next.js, Vue, Svelte, or other frontend, wiring up ConvexProvider, configuring environment variables for the deployment URL, or running npx convex dev for the first time, even if the user just says "set up Convex" or "add a backend."
Plans and executes safe Convex schema and data migrations using the widen-migrate-narrow workflow and the @convex-dev/migrations component. Use this skill when a deployment fails schema validation, existing documents need backfilling, fields need adding or removing or changing type, tables need splitting or merging, or a zero-downtime migration strategy is needed. Also use when the user mentions breaking schema changes, multi-deploy rollouts, or data transformations on existing Convex tables.
Audits and optimizes Convex application performance across hot-path reads, write contention, subscription cost, and function limits. Use this skill when a Convex feature is slow or expensive, npx convex insights shows high bytes or documents read, OCC conflict errors or mutation retries appear, subscriptions or UI updates are costly, functions hit execution or transaction limits, or the user mentions performance, latency, read amplification, or invalidation problems in a Convex app.
Sets up Convex authentication with user management, identity mapping, and access control. Use this skill when adding login or signup to a Convex app, configuring Convex Auth, Clerk, WorkOS AuthKit, Auth0, or custom JWT providers, wiring auth.config.ts, protecting queries and mutations with ctx.auth.getUserIdentity(), creating a users table with identity mapping, or setting up role-based access control, even if the user just says "add auth" or "make it require login."
Finalizes any PR for merge by verifying title/description match implementation AND performing code review for best practices. Use when asked to "finalize PR", "check PR description", "review commit message", before merging any PR, or when PR implementation changed during review. Do NOT use for extracting lessons or investigating build failures.
Adds a new built-in RESP command to Garnet end-to-end. Covers enum registration, parsing, dispatch, RESP handler, API surface, storage session, RMW callbacks, command metadata JSON, ACL tests, and integration tests. Use when asked to "add a command", "implement RI.SET", "add RESP command", or any new server command. Do NOT use for custom extension commands (CustomRawStringFunctions) or object-type sub-operations.
Use when writing, running, or debugging tests in the Gutenberg repository — JavaScript unit tests (Jest), PHP tests (PHPUnit), or end-to-end tests (Playwright).
quiche draft GitHub release automation from a release commit hash or existing tag. Use when creating draft releases for the quiche crate from a quiche/Cargo.toml version bump.
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.
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.
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).
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.
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.
Fast tokenizers optimized for research and production. Rust-based implementation tokenizes 1GB in <20 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.
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.
Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support
Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support
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 <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.
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.
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.
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.
Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, <1% FPR. Fast (<2ms GPU). Multilingual (8 languages). Deploy with HuggingFace or batch processing for RAG security.
Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention
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.
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.
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.
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.
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.
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.
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.
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.
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.
Claude Skills — questions
Answers built from the skills we actually parsed.