仓库/Skill 库

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.

面向 AI Agent 的 Markdown 优先的长期记忆基础设施。通过 MCP 在 markdown/代码文件上实现 BM25 + 语义混合检索。Markdown-first, long-term memory infrastructure for AI agents. Hybrid BM25 + semantic search across markdown/code files via MCP.

AgentWorks：开源控制平面，用于在企业内运行、衡量和改进 AI Agent 工作流。AgentWorks: the open-source control plane for running, measuring, and improving AI agent workflows across your company.

面向 Node.js 与 TypeScript 的 AI firewall。阻止 prompt injection、音频幻觉与 RAG 数据爬取。零依赖。MIT 许可。AI firewall for Node.js & TypeScript. Stop prompt injection, audio hallucinations, and RAG data scraping. Zero dependencies. MIT

Promovolve 是一项试图找回杂志广告昔日魅力的尝试。Promovolve is an attempt to get back what magazine advertising had!

Claude Code Agent 的 Web 仪表盘与自动驾驶 —— 在浏览器中管理会话、任务与审批Web dashboard + autopilot for Claude Code agents — manage sessions, tasks, and approvals from your browser

AI 工程操作系统。在一个地方统一协调模型、Agent、工具、仓库与长期运行的任务。The operating system for AI engineering. Coordinate models, agents, tools, repositories, and long-running work from one place.

Katra — AI Agent 的认知记忆即服务。开源认知记忆基础设施，支持 MCP 协议。Katra — Cognitive Memory as a Service for AI Agents. Open-source cognitive memory infrastructure with MCP protocol support.

AI 辩论，代码裁定，亏损留在账面上。一款可移植的投资决策工作流插件与可验证 harness，已在真实的港股+美股组合上验证。AI argues. Code settles. The losses stay on the page. A portable investment decision-workflow plugin and verifiable harness, proven on a real HK + US portfolio.

为 AI Agent 准备的考试评分知识脑：粘贴文档 URL，即可获得基于 MCP 的可搜索知识脑，并附带测量分数与已知缺口。AGPL。Exam-scored knowledge brains for AI agents: paste a docs URL, get a searchable brain over MCP with a measured score and known gaps. AGPL.

强化的纯本地 LiteLLM 网关——自动发现 15+ 提供商的免费/试用 LLM 模型（OpenRouter、Groq、Gemini、Cerebras、NVIDIA NIM、SambaNova、Cohere、GitHub Models 等），统一 OpenAI 兼容端点 127.0.0.1:4000，为 AI coding agent 提供基于角色的别名。Hardened local-only LiteLLM gateway — auto-discovers free/trial LLM models from 15+ providers (OpenRouter, Groq, Gemini, Cerebras, NVIDIA NIM, SambaNova, Cohere, GitHub Models, etc.). Single OpenAI-compatible endpoint at 127.0.0.1:4000. Role-based aliases for AI coding agents.

OpenClaw 的社区补丁、配置和进阶用户 mod — 按 Agent 压缩、上下文管理等。Community patches, configs, and power-user mods for OpenClaw — per-agent compaction, context management, and more

软件系统的工作空间智能：为 AI Agent、开发者工具和工程团队构建有证据支撑的软件系统认知。Workspace Intelligence for Software Systems: Build an evidence-backed understanding of your software system for AI agents, developer tools, and engineering teams.

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

AI 编程 Agent 工作流运行时：工作流路由、项目工件看板、Provider 网关、适配器激活与可执行治理。AI workflow runtime for coding agents: workflow routing, project artifact boards, provider gates, adapter activation, and executable governance.

一个单文件引导程序，帮助编程 agent 跨会话与上下文窗口执行长周期任务。A one-file bootloader that helps coding agents execute long-horizon projects across sessions and context windows.

Agent SkillOS — 自进化 AI Agent 的财富积累层。Agent SkillOS — the wealth-accumulation layer for self-improving AI agents.

VoiceAudition — AI Agent 的语音选角导演。跨供应商 TTS 语音发现与推荐。VoiceAudition — the voice casting director for AI agents. Cross-provider TTS voice discovery and recommendation.

人类、AI Agent 与本地 LLM 共享应用、知识与能力的开放协议与参考节点，支持自托管与联邦化，MIT 许可。Open protocol and reference node where humans, their AI agents, and local LLMs share apps, knowledge, and capabilities. Self-hosted, federated, MIT.

墨西哥开放法律情报：覆盖联邦、州、市三级来源的 3 万+ 部法律 · 350 万+ 可检索条文 · 三语界面（西班牙语 · 英语 · 古典纳瓦特尔语）· 一方 RAG 聊天 · OpenAPI + SDK · 为市民、会计师、律师与 AI Agent 像代码一样导航墨西哥法律。Open legal intelligence for Mexico. 30,000+ laws across federal, state, and municipal sources · 3.5M+ searchable articles · trilingual UI (Spanish · English · Classical Nahuatl) · first-party RAG chat · OpenAPI + SDKs · built so citizens, accountants, lawyers, and AI agents can navigate Mexican law as code.

潜伏在终端与桌面里的幽灵 AI 编程 agent「Wraith」· A ghost-themed AI coding agent for terminal & desktop

你的分身。Claude Code、Codex 和 OpenCode 的任务控制中枢：查看每个 session、审批 Agent 行为、按计划调度运行。自托管，MIT 协议。You, multiplied. Mission control for Claude Code, Codex, and OpenCode: see every session, approve what your agents do, and run them on a schedule. Self-hosted, MIT.

管理你生活的 AI，让你在身体、心理与财务上保持健康。Telegram 可下载。AI that manages your life. Making you physically, mentally, and financially healthy. Download on Telegram.

面向 AI 编码 Agent 的厂商中立 continuation 协议：纯文件项目状态、标准库验证、恢复与冷启动 Agent 交接。Vendor-neutral continuation protocol for AI coding agents. Plain-file project state, stdlib validation, recovery and cold-agent handoff.

AI 运营门店。人类定方向，Agent 执行落地。AI run the shop. Humans set direction, agents execute.

动态 README，包含 AI 生成的 SCP 基金会与 Wikipedia 随机文章摘要。A dynamic README with AI-generated summaries of random articles from the SCP Foundation and Wikipedia.

机器可读的架构决策记录。Git 原生的决策记忆与治理，面向人类与 Agent 编写的方案 — 集成 lint、CI 与 MCP。Machine-readable architecture decision records. Git-native decision memory and governance for human and agent-authored plans -- lint, CI, and MCP.

🤖 利用 TRPO、PPO、DPO、GRPO、DAPO、GSPO 等先进算法提升强化学习的稳定性与效率，优化策略训练。🤖 Enhance reinforcement learning stability and efficiency with advanced algorithms like TRPO, PPO, DPO, GRPO, DAPO, and GSPO for optimized policy training.

为 AI Agent 设计的记忆系统，学习真正起作用的教训：在行动瞬间注入经验，依据结果归功，以反证挑战。本地优先、多 Agent、测试门控。Memory for AI agents that learns what actually helped: lessons injected at the moment of action, credited by outcome, challenged by counter-evidence. Local-first, multi-agent, test-gated.

DeepThink-AIOS 是一个完全本地化的多 agent 系统，配备双沙箱代码/逻辑验证、ChromaDB RAG 记忆与动态 VRAM 多路复用。DeepThink-AIOS is a fully local, multi-agent System with dual-sandbox code/logic verification, ChromaDB RAG memory, and dynamic VRAM multiplexing .

扩展 gastown 的 Gas City pack，包含自定义 Agent、公式与规划工作流。A Gas City pack with custom agents, formulas, and planning workflows, extending gastown

在工作发生的地方就地捕获团队知识，由 agent 归档，并以可核验的引用形式给出回答。A team's knowledge, captured where the work happens, filed by an agent, and answered with citations you can check.

以孟加拉语优先、面向方言的 LLM 研究。开源孟加拉语 tokenizer，性能优于 Sarvam、AI4Bharat 和 GPT-4o（fertility 1.52，几乎零破损 conjuncts）。非商用，保护孟加拉语及其方言。Bengali-first, dialect-aware LLM research. Open-source Bengali tokenizer that outperforms Sarvam, AI4Bharat, and GPT-4o (fertility 1.52, near-zero broken conjuncts). Non-commercial, preserving Bengali and its dialects.

众多 Agent，一个目标——你的目标。Many Agents, one objective... yours

🤖 基于 AI 分析与智能搜索，定制化发现 ArXiv 上你感兴趣的最新论文🤖 Discover the latest research papers on ArXiv tailored to your interests using AI-powered analysis and smart search features.

面向 LLM 的无损上下文压缩：token 用量减少 2–8 倍，按字节精确还原（sha256 校验）。提供 Claude Code 与 Cursor 的 MCP server。Lossless context compression for LLMs: 2x-8x fewer tokens, byte-exact reconstruction (sha256-verified). MCP server for Claude Code & Cursor.