仓库/Skill 库

面向 LLM 和 LVLM 的强大模型压缩框架，适配 NVIDIA GPU 和华为昇腾 NPUA powerful model compression framework for LLMs and LVLMs, adapted for NVIDIA GPUs and Huawei Ascend NPUs.

面向治理、风险与合规（GRC）的 Claude Skills：针对 ISO 27001、SOC 2、FedRAMP、GDPR、HIPAA、NIST CSF、PCI DSS、EU AI Act、ISO 42001、ISO 27701、DORA、CSRD、印度 DPDPA、CMMC 2.0、NIST AI Risk、SWIFT、澳大利亚 ISM、EU NIS2、CCPA/CPRA 等的专家级合规指导。使用 skills 基准 97%，不使用 81%。Claude Skills for Governance, Risk, & Compliance (GRC): Expert-level compliance guidance for ISO 27001, SOC 2, FedRAMP, GDPR, HIPAA, NIST CSF, PCI DSS, EU AI Act, ISO 42001, ISO 27701, DORA, CSRD, India's DPDPA, CMMC 2.0, NIST AI Risk, SWIFT, Australia's ISM, EU NIS2, CCPA/CPRA, and others. Benchmark 97% (with skills) vs 81% (without skills).

基于 Google PaperOrchestra 论文实现的全自动 AI 研究论文写作器，通过技能-基准测试 + 自动评分器，配合任意编码 Agent（Claude Code、Cursor、Antigravity、Cline、Aider）。无需 API Key，无需 LLM SDK。An automated AI research-paper writer based off Google's PaperOrchestra paper's implementation through a skills - benchmark + autoraters using any coding agent (Claude Code, Cursor, Antigravity, Cline, Aider). No API keys, no LLM SDKs.

Qwen3.8-27B 在单卡 RTX 3090 上使用 vLLM 部署：64 并发下约 1,000 tok/s（int8 张量核心 GEMM、fp16 DeltaNet 状态），默认采样下单用户约 114 tok/s/贪心约 124 tok/s（MTP 草稿、自输出草稿词表、校准 int4 lm_head、split-KV 校验注意力），150k–262k 上下文；附带补丁、重新量化脚本与基准测试Qwen3.8-27B on a single RTX 3090 with vLLM: ~1,000 tok/s at 64 concurrent (int8 tensor-core GEMMs, fp16 DeltaNet state), ~114 tok/s single-user at default sampling / ~124 greedy (MTP drafts, own-output draft vocab, calibrated int4 lm_head, split-KV verify attention), 150k-262k context; patches, requant scripts, benchmarks

BioPhi 是开源抗体设计平台，提供自动化抗体人源化方法（Sapiens）、人源性评估（OASis）及计算机辅助抗体序列设计界面BioPhi is an open-source antibody design platform. It features methods for automated antibody humanization (Sapiens), humanness evaluation (OASis) and an interface for computer-assisted antibody sequence design.

适用于 OpenAI Codex 的研究类 skills 精选清单，覆盖写作、文献综述、评估与研究工作流A curated list of research-oriented skills usable in OpenAI Codex, covering writing, literature review, evaluation, and research workflows.

一门实战课程，用 PyTorch 从零构建现代 LLM，包含 26 个可运行的 Jupyter Notebook，涵盖 tokenizer、attention、MoE、RLHF、推理、评估和蒸馏。A hands-on course for building modern LLMs from scratch in PyTorch, with 26 runnable Jupyter Notebooks covering tokenizers, attention, MoE, RLHF, inference, evaluation, and distillation.

Claude Fable 5 终极指南 2026：使用场景、集成与基准测试Ultimate Claude Fable 5 Guide 2026: Use Cases, Integrations & Benchmarks

面向 Godot 的开源游戏编程 Agent 环境与基准An open-source game coding agent environment and benchmark for Godot.

一个 LLM 相关论文、学位论文、工具、数据集、课程与基准的合集。A collection of LLM related papers, thesis, tools, datasets, courses, benchmarks

用于可复现 AI Agent 评测的确定性、可分叉、有状态 MCP 测试世界——不对生产环境产生副作用。Deterministic, forkable, stateful MCP test worlds for reproducible AI agent evaluation—without production side effects.

[ISPRS 2024] 卫星视频单目标跟踪：系统综述与定向目标跟踪基准[ISPRS 2024] Satellite Video Single Object Tracking: A Systematic Review and An Oriented Object Tracking Benchmark

面向长时对话记忆层的综合基准测试框架A Comprehensive Benchmarking Framework for Long-Term Conversational Memory Layers

热门 claude-skills 与 AI Agent 仓库的自动更新排行榜，每 15 分钟刷新一次。Auto-updated leaderboard of trending claude-skills and AI agent repos, refreshed every 15 minutes

AI 原生的结构化数据 wire 格式。在每个前沿模型上实现 100% 理解，比 JSON 减少 50-92% token，跨 17 种格式完成 43B+ 无损往返。Spec v3.4 Stable。The AI-native wire format for structured data. 100% comprehension on every frontier model. 50-92% fewer tokens than JSON. 43B+ lossless round-trips across 17 formats. Spec v3.4 Stable.

🕸️ 工程化组织，而不仅仅是 Agent。562 项精选资源 · 9 个设计层 · 11 个章节 · 252 篇论文与预印本 — 一本面向图结构多 Agent 系统（角色、拓扑、交接、工作图、状态、闸门、可靠性、可观测性）的实战指南、CC0 开源数据集与交互式地图册。🕸️ Engineer the organization, not just the agent. 562 curated resources · 9 design layers · 11 sections · 252 papers & preprints — a field guide, CC0 open dataset, and interactive atlas for graph-structured multi-agent systems: roles, topologies, handoffs, work graphs, state, gates, reliability, observability.

用于评估 Agentic AI 应用的 AI 红队工具与 LLM 安全框架，测试 prompt injection，支持漏洞评估、SBOM 生成与静态分析。AI red-teaming tool and LLM security framework to evaluate agentic AI applications. Tests prompt injections, handles vulnerability assessment, SBOM generation, and static analysis.

面向 AI agent 的基于证据的评估——将每条断言与 agent 真实工具输出进行核对（受约束、基于证据的模型判断，而非整体式 LLM 评判的猜测），并附带置信区间。Evidence-grounded evaluation for AI agents — verifies each claim against the agent's real tool outputs (constrained, evidence-grounded model judgment, not holistic LLM-judge guesswork), with confidence intervals.

面向 ECG-语言模型（ELM）的研究型训练与评估框架A research-oriented training and evaluation framework for ECG-Language Models (ELMs)

FlexEval 是一个面向实际量化分析的 LLM 评估工具。FlexEval is an LLM evaluation tool designed for practical quantitative analysis.

每日 LLM 价值排行榜——基于智能、速度、价格对比 300+ 模型。OpenRouter + Artificial Analysis。大模型性价比排行榜Daily LLM value rankings - compare 300+ models by intelligence, speed and price. OpenRouter + Artificial Analysis. 大模型性价比排行榜

RAG 管道的回归测试与配置扫描，附带统计学指标以判断变更是否真正带来改进。Regression testing and configuration sweeps for RAG pipelines, with the statistics to know whether a change actually helped.

微调、评估、提示工程、开源模型Fine-tuning, evaluation, prompting, open-source models

MobileMem：面向持续演进 Agent 的设备端记忆MobileMem: On-Device Memory for Continually Evolving Agents

面向 MCP 的 Agent 评估标准——对输出质量打分、捕获安全失败、强制成本预算。The agent eval standard for MCP — score output quality, catch safety failures, enforce cost budgets

Proof Gradient 是 agent 演化协议：每次运行都会留下 proof，每个 proof 都会筛选 intelligence，每个被选中的 artifact 都会推动网络演化。Proof Gradient is the agent evolution protocol where every run leaves proof, every proof selects intelligence, and every selected artifact evolves the network.

🔍 使用 DeepSearch 自动化数字足迹分析，借助高级 Google 搜索操作符高效收集多平台洞察。🔍 Automate digital footprint analysis with DeepSearch, using advanced Google search operators to gather insights from multiple platforms efficiently.

builderr.ai 交易 Agent 模板——fork 后实现 decide()，即可提交。Trading agent template for builderr.ai — fork this, implement decide(), submit.

Galactus 在低于规格的 Mac 上运行 744B 与 235B MoE 模型，具备与 llama.cpp 的逐位精度一致、以 RAM 作为缓存的执行模式，并提供完整的本地应用，涵盖 Agent、权限门控、代码编辑器、已认证的服务器模式、定时无人值守运行以及完整公开的测评数据Galactus executes 744B and 235B MoE models on undersized Macs with bit‑perfect llama.cpp parity, RAM-as-cache execution, and a full local app featuring an agent, permission gate, code editor, authenticated server mode, scheduled unattended runs, and fully published measurements.

自动化并规模化 "LLMs as a participant"，将 LLM 作为研究参与者Automates and scales "LLMs as a participant."

实时更新的向量数据库项目、集成和基准评测全景图——每……刷新。Live-updating landscape of vector database projects, integrations, and benchmarks — refreshed every

用于生成基于 Transformer 的 LLM 完整注意力头热力图的一组脚本。A set of scripts to generate full attention-head heatmaps for transformer-based LLMs

硕士论文（UiB，计算语言学，2023）及扩展工作。针对系统综述筛选的文本分类。原始 11 工作流分析结合 NEO ontology 集成，以及 Cohen 等人（2006）药物类别基准的扩展，比较 BoW 与 BiomedBERT。Master's thesis (UiB, Computational Linguistics, 2023) and extension work. Text classification for systematic review screening. Original 11-workflow analysis with NEO ontology integration plus Cohen et al. (2006) drug-class benchmark extension comparing BoW and BiomedBERT.

AI agent memory 与基础设施全景——912 个系统 × 68 列的对比目录，覆盖记忆层、agent 框架、运行时、vector store、知识图谱、MCP server、benchmark。支持按类型化边、谱系、引用进行检索。AI agent memory & infrastructure landscape — comparative catalog of 912 systems × 68 columns covering memory layers, agent frameworks, runtimes, vector stores, knowledge graphs, MCP servers, benchmarks. Searchable with typed edges, lineages, citations.

🖥️ 通过终端中键盘优先的 Kanban board 简化工作流，实现快速、专注的任务管理。🖥️ Streamline your workflow with a keyboard-first Kanban board in your terminal for fast, focused task management.

绵羊遗传评估 CLI 与 MCP server——基于 NSIP 数据库检索个体、对比 EBV、规划配种、排序羊群。Sheep genetic evaluation CLI & MCP server -- search animals, compare EBVs, plan matings, rank flocks via the NSIP database