Explore AI Agent Skills & Claude Prompts
Discover open-source agent skills for Claude Code, Codex, ChatGPT, and any tool that uses SKILL.md.
Enter through keywords, occupations, creators, and GitHub sources to see what kinds of skills are emerging across domains.
Use the same catalog through the API
Connect 381,784 public skills to your own search, analytics, or agent workflow with the REST API.
Querying local SQLite index...
nvidia-datacenter-bringup
by air-gappedBring up NVIDIA HGX/DGX datacenter GPU hosts on Ubuntu 24.04 LTS — air-gapped or connected, Secure Boot enabled. Covers B300/B200/H100/A100/L40S/L4 driver+fabricmanager+NVLSM+DOCA-OFED install order and exact package set from NVIDIA CUDA repo + DOCA repo. Triggers on B300/B200/HGX/DGX install, "fabricmanager won't start", "system not yet initialized" / cudaErrorSystemNotReady, NVLSM missing, ib_umad not loading, DOCA-OFED before NVIDIA driver, nvidia-driver-pinning-XXX, nvlink5-XXX, nvidia-open vs cuda-drivers, "Blackwell requires open kernel modules", ConnectX-7/8 bridge device, FM exact-version-match, gpu-operator cuda-validator CrashLoopBackOff, B300 PCI ID 0x3182, air-gap CUDA + DOCA mirror, three-tier DOCA GPG key, MOK enrollment, DKMS sign, Dell PowerEdge XE9780/XE9785 baseboard firmware v1.4.30, iDRAC Redfish virtual AC cycle DellOemChassis.ExtendedReset, generic "install nvidia driver ubuntu 24.04 datacenter".
nvidia-nixl
by air-gappedNVIDIA Inference Xfer Library (NIXL) operator + developer reference. Point-to-point KV-cache and tensor transport for distributed inference (Dynamo, vLLM, SGLang). Covers the agent API (full Python reference; C++/Rust via upstream pointers), all 13 backend plugins (UCX, GDS, GDS_MT, libfabric, mooncake, posix, hf3fs, obj/S3, azure_blob, gusli, uccl, gpunetio/DOCA, telemetry), build paths (pip nixl-cu12/cu13, meson+ninja from source), ETCD vs side-channel metadata, telemetry (Prometheus + cyclic shared-memory), NIXL-EP elastic MoE device kernels, and Dynamo / vLLM NixlConnector / SGLang integration patterns.
vllm-nvidia-hardware
by air-gappedNVIDIA AI-hardware + vLLM-platform reference covering Hopper (H100/H200), Blackwell (B100/B200/B300) and Blackwell Ultra, Grace-Blackwell superchips and NVL72 racks (GB200, GB300), Vera Rubin (R100/R300) with VR200 NVL144 and Kyber NVL576, Dell PowerEdge XE family and IR5000/IR7000/IR9048 racks. Per-SKU HBM, FP4/FP8/FP16 TFLOPs, NVLink5, TDP, rack power/cooling (135 kW GB300, 180-220 kW NVL144, 600 kW Kyber), DLC vs RDHx, 800 VDC HVDC. Memory-wall roofline, HBM3E→HBM4 supply 2026. vLLM attention-backend × SM matrix, FP4/FP8 paths, KV connectors, Blackwell gotchas (SM103 TRTLLM hang, 270 vs 288 GB B300 bin split).
jira-best-practices
by air-gappedAdvise on USING Jira well, not operating it: make the structural call — is this an epic, a story, a task, or a sub-task? — and diagnose why a Jira is a dread, then recommend the lean fix. Adapt to the organisation's OWN hierarchy names, conventions, and working language instead of imposing a methodology. Self-hosted-first: Jira Data Center 10.3/11.x (no Cloud AI; dual Epic Link + Parent Link). Built for an agent that ACTS on Jira through the jira-cli tool or the mcp-atlassian MCP server while advising the user; Jira web-UI and admin-schema guidance is secondary. Covers ALL project types — software AND non-software (operations, engineering, services, business).
lmcache-mp
by air-gappedLMCache multiprocess (MP) mode — standalone LMCache server in its own pod/process that vLLM connects to over ZMQ. Gives process isolation, no GIL contention on the inference path, one cache shared by multiple vLLM pods per node, and CPU-memory scaling independent of GPU memory. Covers the `LMCacheMPConnector` path (vs the in-process `LMCacheConnectorV1`), the DaemonSet+Deployment K8s pattern and LMCache Operator, the L1 (CPU DRAM) + L2 (NIXL, fs, mooncake_store, s3, Redis) cascade, the `lmcache/standalone` + `lmcache/vllm-openai` image pair, and the production gotchas (`--no-enable-prefix-caching`, `--disable-hybrid-kv-cache-manager`, vLLM/lmcache version pins, hybrid models unsupported, cache_salt fallback bug).
sglang-hicache
by air-gappedSGLang HiCache (hierarchical KV cache) — three-tier prefix cache: GPU HBM (L1) → pinned host DRAM (L2) → distributed L3 (Mooncake / 3FS / NIXL / AIBrix / EIC / SiMM / file / LMCache). Covers `--enable-hierarchical-cache`, all `--hicache-*` flags, write policies, page_first* layouts, prefetch policy (best_effort / wait_complete / timeout), per-rank sizing, MHA / MLA / DSA / Mamba / SWA support matrix (SWA + 3FS hybrid shipped in v0.5.11), runtime attach/detach HTTP admin, and auto-rewrite startup log lines that silently substitute layout × IO × storage combinations.
vllm-caching
by air-gappedvLLM tiered KV cache configuration for production H100/H200 clusters. Native CPU offload, LMCache (CPU+NVMe+GDS), NixlConnector (disaggregated prefill), MooncakeConnector (RDMA), MultiConnector composition. Version gates, sizing math (flag total across TP, not per-GPU — opposite of SGLang), KV-vs-weights offload distinction operators most often get wrong.
vllm-gemma-4-31b
by air-gappedOperating-point reference for serving Gemma 4 31B on vLLM — TP sizing, max_model_len, max_num_seqs, gpu_memory_utilization, kv_cache_dtype, EAGLE3 spec-dec, chat_template choice.
vllm-observability
by air-gappedObserve production vLLM — `/metrics` Prometheus surface (V1 engine), SLO-driven alerting on TTFT/ITL/queue/KV/preemption/aborts/corrupted-logits, shipping Grafana dashboards in `examples/observability/`, OTLP tracing with `--otlp-traces-endpoint` and `--collect-detailed-traces={model,worker,all}`, diagnostic rules to triage from /metrics alone — queue-grows + TPOT-stable means capacity, queue-stable + TPOT-grows means context/model, DCGM `SM_OCCUPANCY` is the real GPU-saturation signal not `GPU_UTIL`. V1 metric names (kv_cache_usage_perc), gpu_→kv_ rename saga, DCGM-exporter pairing, dashboard-lying pitfalls.
vllm-quantization
by air-gappedvLLM datacenter-GPU quantization — picking, configuring, troubleshooting NVFP4, FP8, MXFP4, MXFP8, AWQ, GPTQ, INT8, compressed-tensors, modelopt, quark on H100/H200/B200/B300/GB200/GB300. 29 `--quantization` flag values, KV-cache dtypes (fp8_e4m3, nvfp4, per-token-head, turboquant), MoE backend selection (CUTLASS, TRTLLM, FlashInfer, DeepGEMM, Marlin, Qutlass), producing checkpoints with llm-compressor and NVIDIA ModelOpt (NVFP4_DEFAULT_CFG, FP8_DEFAULT_CFG, W4A16, SmoothQuant+GPTQ), online quantization (`fp8_per_tensor`, `fp8_per_block`), training EAGLE-3/dflash drafters on BF16 targets before PTQ, version gates per vLLM release (v0.14 → v0.21).
openshift-app
by air-gappedPackage applications for OpenShift deployment: container images (UBI, arbitrary UID, multi-stage builds), packaging formats (Helm, Kustomize, Operators, OLM v1), CI/CD (Tekton, ArgoCD, Shipwright, Conforma), security (SCC, PSA, supply chain, image signing, secrets), operations (Routes, probes, scaling, monitoring, storage), disconnected/air-gapped patterns, and critical gotchas. Also when an app "works on Kubernetes but fails on OpenShift" (SCC denied, random/arbitrary UID, permission errors). Covers OCP 4.14-4.21. NOT for cluster installation or infrastructure management.
sglang-model-gateway
by air-gappedSGLang Model Gateway (`sgl-model-gateway`, formerly `sgl-router`) — Rust router fronting vLLM and SGLang inference workers on Kubernetes. Covers first-class vLLM gRPC backend plus HTTP transparent-proxy for vanilla vLLM, the policy set (six `--policy` values, `cache_aware` default), tokenizer-format dispatch (`tokenizer.json` HF-fast vs `tiktoken.model` BPE — including when neither is required because `cache_aware` is text-based), air-gapped recipe (gateway ignores `HF_ENDPOINT`, mount tokenizer files on PVC only when actually needed), K8s manifests with `model_id` labels and per-model RBAC, three HA mitigations (single + PDB, `sessionAffinity: ClientIP`, `--enable-mesh` CRDT sync), and a pitfall catalog covering the Dec 2025 `sgl-router` → `sgl-model-gateway` rename and over-engineered tokenizer init-container traps.
Browse Agent Skills by Occupation
23 major groups · 867 SOC occupations
Browse by Category
Explore agent skills organized by their primary use case
Explore the agent skills ecosystem by occupation and creator
SkillMD is not just a keyword search box. It is an open map that organizes public skills by occupation, creator, and repository, helping you see which workflows, judgment criteria, and domain habits people are writing for AI agents.
Then follow creators and GitHub repositories back to the source: compare the skills a team maintains, whether the repo is active, and how the README frames the work before you open, install, or reuse anything.
Use it three ways: learn an unfamiliar field by occupation, study how creators organize skills, then use source context to decide what is worth opening or reusing.
01 Map a field
Browse 23 occupation groups and 867 SOC roles to learn what skills exist in adjacent domains and how they break down real work.
02 Follow creators
Use creator and repository pages to inspect maintained skill collections, recent updates, and source context before trusting a result.
03 Search with sources
Search 1.7M+ collected skills, then use occupation tags, creators, and GitHub source context to decide what is worth opening.
Start with the occupation map, then follow creators and repositories back to real code. SkillMD helps explain why a skill is worth opening, not only what it is named.
Standardizing Agent Capabilities with SKILL.md and Model Context Protocol (MCP)
In the rapidly evolving landscape of artificial intelligence, LLM agents (Large Language Model agents) have transitioned from simple text predictors to autonomous problem solvers. To orchestrate complex, multi-step agentic workflows, developers require a standardized format to specify agent capabilities, prompt instructions, system rules, and database bindings. This is where SKILL.md and the Model Context Protocol (MCP) have emerged as standard developer paradigms. SkillMD serves as the central directory for indexing, exploring, and sharing these critical agent configurations.
Our open-source registry currently tracks over 1.7 million collected SKILL.md configurations and system prompts. By compiling agent configurations from active developers on GitHub, we bridge the gap between prompt engineering research and production execution. Whether you are building agents with Anthropic's Claude Code, OpenAI's GPT-4, Google's Gemini, or local models using Ollama and LlamaIndex, standardized skill definitions ensure your agents behave predictably across different runtime environments.
What is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is an open-source standard designed to connect LLMs to data sources, developer tools, and external environments. MCP establishes a bidirectional communication channel between client applications (like Cursor, Claude Desktop, or custom agent systems) and servers hosting data or capabilities. Standardizing instructions via SKILL.md enables LLMs to query databases, read local files, execute terminal commands, and integrate third-party APIs. SkillMD allows you to find ready-to-run MCP servers and prompt instructions for various occupations and technical tasks.
The Structure of a Professional SKILL.md File
A valid SKILL.md configuration is designed to be easily read by humans and parsed by LLMs. It contains precise system instructions, trigger conditions, required parameters, and execution examples. Below is the typical architectural blueprint of a professional agent skill:
- Metadata & Core Scope: Declares the name of the skill, author details, target models, and a description of the capability.
- Triggers & Intent Detection: Details semantic triggers that help the agent decide when to invoke this skill.
- System Prompts: Explicit system-level instructions that direct the agent's behavior, personality, safety guardrails, and formatting preferences.
- Capabilities & Tools: Lists the files, databases, or APIs the agent must access to complete the tasks.
- Few-Shot Examples: Demonstrates real inputs and outputs, helping the model generalize behavior through in-context learning.
Optimizing Agent Workflows for Modern LLMs
Writing effective agent skills requires deep knowledge of prompt engineering. With the release of advanced reasoning models like Claude 3.5 Sonnet, ChatGPT o1, and DeepSeek-V3, prompt templates must focus on structured thinking. Developers are encouraged to use XML tags (e.g., <thought>, <context>, and <rules>) to isolate execution boundaries. Standardized prompts prevent agents from suffering from context drift, ensuring that long-running tasks remain aligned with the initial system parameters.
Exploring by SOC Occupations and Creator Profiles
What makes SkillMD unique is its taxonomy. Instead of simple text search, we parse and organize files according to the Standard Occupational Classification (SOC) system. This means you can discover skills written for Computer and Mathematical roles, Business and Financial operations, Legal, Design, and and Educational Instruction fields. By tracking creator profiles, developers can study how different teams organize their custom instructions, compare version updates, and fork public configs for specialized enterprise use cases.
SkillMD operates as a high-performance index running on a fast Go backend and a highly responsive Astro SSR frontend. All search queries execute in milliseconds, featuring smart debouncing to prevent multiple API requests while keeping user data secure. Join our community of developers to standardize your AI agent instructions and optimize your LLM prompting workflows today.
Frequently Asked Questions
A practical guide to agent skills: what they are, how to inspect them, and how SkillMD helps you explore the ecosystem.