仓库/Skill 库

Haven 42 是一款证据门控、本地优先的 AI 工作台，支持 Windows、Linux 和 macOS 上的编码、聊天与媒体处理。Haven 42 is an evidence-gated, local-first AI workbench for coding, chat, and media on Windows, Linux, and macOS.

UMD 行星长期记忆算法、公式、基准、SDK 与可复现研究。UMD planetary long-term memory algorithms, formulas, benchmarks, SDKs, and reproducible research

针对 LLM 集成证据筛选（系统综述标题/摘要筛选）的筛选-自验证内核。Screening-and-self-validation kernel for LLM-ensemble evidence screening (systematic review title/abstract screening).

自托管 AI 新闻门户的 Agentic 控制层：一个标准库 Python CLI 加一个 MCP server（32 个工具），由 LLM agent 驱动，通过门控编辑流程撰写并发布 AI 人设文章，策展首页与作者，并完成部署。附带 prompt/skill 配方与 7×24 服务循环。Agentic control layer for the Censurado self-hosted AI news portal: a stdlib Python CLI plus an MCP server (32 tools) an LLM agent drives to write and publish AI-persona articles through a gated editorial walk, curate the front page and authors, and deploy. Ships the prompt/skill recipe and a 24/7 serve loop.

使用本地 RAG 系统增强知识库，借助混合搜索实现精准信息检索与最优结果🔍 Enhance your knowledge base with a local RAG system that leverages hybrid search for precise information retrieval and optimal results.

用于远程访问 Claude Code 的安全加固 Telegram bot。包含 24 层纵深防御、工具 allow/deny 列表、每用户成本上限、MCP 集成、本地语音转写。Security-hardened Telegram bot for remote Claude Code access. 24 defense-in-depth layers, tool allow/deny lists, per-user cost caps, MCP integration, local voice transcription

🦀 高性能 Rust 多 Agent 编排框架，内置 MCP runtime，并提供统一的 OpenAI / Open Responses LLM 网关。🦀 High-performance Rust multi-agent orchestrator with MCP runtime and unified OpenAI / Open Responses LLM gateway.

面向 AI Agent 的证据可追溯研究 API。哈希链存证可自行验证，明确拒答，并通过 x402 微型支付——自身失败时绝不收费。Evidence-grounded research API for AI agents. Hash-chained custody you can verify yourself, explicit refusal, and x402 micropayments — it never bills you for its own failure.

面向 AI Agent 的精确小数复式记账内核。通过 MCP 工具调用完成金额计算，避免在 token 中做财务运算。默认拒绝、审计哈希、完全确定性。Exact-decimal double-entry kernel for AI agents. Call MCP tools instead of doing money math in tokens. Fail-closed, audit-hashed, deterministic.

AI it yourself——一款原生 iOS 聊天应用，Agent 可在其中构建你所需的小工具，沙箱化且立即可用。无服务器、无账号：自带 API key、本地机器或在 iPhone 上运行的模型。AI it yourself — a native iOS chat where agents build the small tools you need, sandboxed and usable immediately. No server, no account: bring your own API key, your own machine, or a model running on the iPhone.

这份 AI 研究清单告诉你该"做什么不同"，而不仅仅是读什么——100+ 篇 2026 年顶会论文 → 50+ 条面向 AI Agent 构建者的生产实践要点。中英双语。The AI-research list that tells you what to DO differently, not just what to read — 100+ top-tier 2026 papers → 50+ production takeaways for AI agent builders. Bilingual EN/中文.

大语言模型情境感知相关论文精选目录。Curated bibliography of papers on situational awareness in large language models

Warzone2100 的 Skirmish AI 机器人Skirmish AI bot for Warzone2100

本人毕业学年所有天体物理课题汇总。毕业设计聚焦轨道转移优化，采用梯度下降法及 Markov Chain Monte Carlo 方法进行不确定性评估These are all the astrophysics projects in my final year. My final-year project focuses on orbital transfer optimisation using gradient descent and the Markov Chain Monte Carlo method for uncertainty evaluation.

使用 LLM 开展的迷幻治疗精神病疾病活体系统综述，涵盖标注、训练到 Web 应用全流程。Living Systematic Review on Psychedelic Treatment for Psychiatric Deseases usind LLMs - from annotation over training to web app

MeetMaxxing 是一个多 Agent AI 会议副驾，将在线会议转变为智能、协作的体验。专用 AI Agent 处理实时辅助、会议摘要、语义记忆、跟进邮件和智能日程——通过 Agent-to-Agent 通信协作，而非单一单体 LLM 工作流。MeetMaxxing is a Multi-agent AI meeting copilot that transforms online meetings into intelligent, collaborative experiences. Specialized AI agents handle live assistance, meeting summaries, semantic memory, follow-up emails, and smart scheduling — collaborating through Agent-to-Agent communication instead of one monolithic LLM workflow.

Aqua — 研究与学习助手。测验、知识捕获、文献综述。本地部署，隐私优先。Aqua — research & study assistant. Quizzes, knowledge capture, literature review. Locally hosted, privacy-first.

2026 终极 AI 记忆栈：面向 LLM 的本地优先 Rust MCP2026 Ultimate AI Memory Stack: Local-First Rust MCP for LLMs

🔒 模拟对抗行为以测试和强化 MCP 防御，无需真实利用或风险，确保稳健的安全评估🔒 Simulate adversarial behaviors to test and strengthen MCP defenses without real exploitation or risk, ensuring robust security evaluations.

精选的评分量表、检查清单、评分标准集、原则和评分指南汇总，用于对现代生成模型进行评分、排名、验证、过滤或训练。A curated list of rubrics, checklists, criteria sets, principles, and scoring guides used to score, rank, verify, filter, or train modern generative models.

证据门禁式 Coding Agent 编排器：冻结任务、独立验证、绑定摘要的评审、仅草稿可发布。Evidence-gated orchestration for coding agents: frozen tasks, independent verification, digest-bound review, draft-only publication. 证据门禁式 Coding Agent 编排器。

开源的托管型 agent harness：持久化的 LLM agent session、每 session 一个 sandbox、支持无人值守运行——基础设施由你拥有，model key 由你掌握。An open-source managed agent harness: durable LLM agent sessions, a sandbox per session, and unattended runs — on infrastructure you own, with model keys you hold.

为 AI 编程 agent 构建 Claude Code prompt 模板，包含系统 prompt、工具 prompt、记忆管理与多 agent 协调。Build Claude Code prompt templates for AI coding agents, with system prompts, tool prompts, memory, and multi-agent coordination

Mistral Studio 产品营销策略、品类叙事、定位与信息框架、从专业消费者到企业的分层、销售作战卡、品牌资产与市场分析。Mistral Studio Product Marketing Strategy, Category Narrative, Positioning & Messaging Framework, Prosumer-to-Enterprise Segmentation, Sales Battlecards, Brand Assets & Market Analysis.

🤖 为 Alrouf Lighting 提供 AI 驱动的 RFQ 自动化处理平台，集成 CRM 流程以提升运营效率🤖 Automate RFQ handling with our AI-powered platform, integrating CRM processes to streamline operations and enhance efficiency for Alrouf Lighting.

为 Ramone 提供的容器化本地 RAG 服务，使用 atlas-corpus 检索、ChromaDB 会话记忆和 Ollama 生成。Containerised local RAG service for Ramone using atlas-corpus retrieval, ChromaDB session memory and Ollama generation.

解锁 1000+ 通用 Agentic AI Skills 2026 — Claude、Gemini、Cursor 脚本Unlock 1000+ Universal Agentic AI Skills 2026 - Claude, Gemini, Cursor Scripts

AI agent 沙箱的时间旅行调试——基于 Firecracker microVM，支持 record、snapshot、restore 与 diff，非容器方案。Time-travel debugging for AI agent sandboxes — Firecracker microVMs with record, snapshot, restore, and diff. Not containers.

📄 面向 AI Agent、RAG、知识图谱的每日论文发现 —— 覆盖 arXiv + HuggingFace + Papers With Code📄 Daily paper discovery for AI Agent, RAG, Knowledge Graph - arXiv + HuggingFace + Papers With Code

📄 从任意网站提取面向 AI Agent 和 LLM 的文档，通过便捷访问 llms.txt 和 install.md 文件简化自动化工作流。📄 Extract documentation for AI agents and LLMs from any website, simplifying automation workflows with easy access to llms.txt and install.md files.

一款将模型命令沙箱化，并使用可编辑状态机处理长时间任务的编码 AgentA coding agent that jails model commands and uses editable state machines for long-running tasks

构建安全、有治理、可观测、成本可控的云与 AI 平台能力的实用参考框架。A practical reference framework for building secure, governed, observable, and cost-aware Cloud & AI platform capabilities.

一个通过自博弈掌握 LLM 工具调用的 Python 库A Python library for LLM tool call mastery through self play.

面向飞书/Lark 的开源 AI 辅助入职方案：可配置工作流、文档 OCR 与审阅、Bitable 同步、提醒功能，以及零凭证 Demo。Open-source AI-assisted HR onboarding for Feishu/Lark: configurable workflows, document OCR and review, Bitable sync, reminders, and a zero-credential demo.

CI 验证的科学领域 Agent Skill——69 个研究级 Skill，覆盖科学写作、机器学习、数据分析、量子计算与研究数据库。无死链、无虚假 Skill、无厂商广告。兼容 Claude Code / Cursor / Codex。CI-validated Agent Skills for science - 69 research-grade skills for scientific writing, ML, data analysis, quantum computing, and research databases. No dead links, no phantom skills, no vendor ads. Claude Code / Cursor / Codex compatible.

Same Targets, Different Computation 的代码与 artifact：后训练如何在模型各层之间划分计算Code and artifacts for Same Targets, Different Computation: how post-training divides work across model layers.