趋势榜

Go 语言 Agent harness 与服务框架A Go agent harness and service framework

开源 AI 同事，每个都拥有自己的"电脑"：浏览器、文件与工具，所有动作执行前决定、执行后记录；支持接入任意 AG-UI 智能体Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

DeepSeek Harness（DSH）一只住在 DSH 界面右下角的小鲸鱼娘，帮你盯着DeepSeek账户余额。QQ弹弹，支持拖拽吸附、左吸附翻转、数字滚动动画，随界面自动启用，建议直接喊来你的dsh安装

面向机器操作的 consequence firewall。EMILIA Gate 在资金、代码、权限、基础设施或受监管状态变更前校验确切授权；其开放协议使证据可独立验证。Consequence firewall for machine actions. EMILIA Gate verifies exact authority before money, code, permissions, infrastructure, or regulated state changes; the open protocol makes the evidence independently verifiable.

一款 YouTube Music 客户端An YouTube Music Client

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

让 Codex 根据最终结果生成标题、注释、commit 和 PR，减少被否决方案在交付中的残留。

雲の音——使用 SwiftUI 构建的原生 macOS 网易云音乐客户端雲の音 — Native macOS NetEase Cloud Music client built with SwiftUI

DS4Windows 的便携式配套工具集，包含 DualShock 4 与 DualSense 的安装指南、配置说明与 Windows 构建助手Portable DS4Windows companion toolkit with setup guides, profile notes, and Windows build helpers for DualShock 4 and DualSense.

链上仍未破解的加密谜题，价值约 10 BTC、约 12 ETH 与约 1,900 AR：托管地址、作者线索以及我的逐项研究；其中少数已破解并附支付凭证Unsolved crypto puzzles worth ~10 BTC, ~12 ETH and ~1,900 AR still on-chain: escrow addresses, author clues, and my research on each. A few solved, with payout proofs.

用于机器人和 Physical AI 的多模态数据质量、处理、enrichment 与 curation pipelines 的开源 SDK。Open source SDK for building multimodal data-quality, processing, enrichment, and curation pipelines for robotics and Physical AI.

生产就绪的 FastAPI 平台，模块化组件，内置 control plane。A production-ready FastAPI platform with modular components and a built-in control plane.

ComfyUI 节点，为 MiniMax H3 而设计，可串联多个视频片段，具备运动上下文、磁盘缓存、动态图像参考、音频参考支持以及最终视频/音频的无缝解码ComfyUI node for MiniMax H3 that chains multiple video clips with motion context, disk caching, dynamic image references, audio reference support, and seamless final video/audio decoding.

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.

DeepSeek Harness 插件，支持跨预设会话迁移的可预览；固定 schema 的交接保留状态、源模型意图与未处理图像，原始会话保持不变DeepSeek Harness plugin for previewable cross-preset session migration. Fixed-schema handoffs preserve state, source-model intent, and unresolved images; the original session stays untouched.

DeepSeek Harness 换肤 / 壁纸 / 主题包插件 (dsh-plugin) — 8 套 Mirage 主题、每用户强调色、壁纸2.0、主题包导入导出/分享链接、收藏与随机，纯原生 token 系统实现。

在 DeepSeek Harness 内预览、创建并编辑办公表格、文档与幻灯片，基于 Univer 实现Preview, create, edit office spreadsheets, docs & slides inside DeepSeek Harness. Power by Univer.

智能多 agent 对话助手，采用 LangGraph 编排，支持 Human-in-the-Loop、企业级可观测性与完整 i18n 支持（6 种语言）。Smart multi-agent conversational assistant with LangGraph orchestration, Human-in-the-Loop, enterprise-grade observability, and full i18n support (6 languages)

独立的 Skill 与 Codex 插件，支持 OpenAI 兼容的图像生成、编辑、批处理工作流、QA 以及聚焦画布编辑。Standalone Skill and Codex Plugin for OpenAI-compatible image generation, editing, batch workflows, QA, and focused canvas editing.

基于 NTFS MFT 的 Windows 超高速文件搜索工具Ultra-fast file search for Windows using NTFS MFT

开源、可自托管的 Claude Code 风格 agent 交付平台：现已支持 Slack，更多 channel 即将上线；同一 bundle 可在本地和 Kubernetes 上运行，git push 即可发布。Open-source, self-hostable delivery platform for Claude Code style agents. Connect Slack today, with more channels next. Run the same bundle locally and on Kubernetes, and ship it with git push.

grantex 是面向 AI Agent 的身份、授权与审计基础设施，堪称 Agentic 互联网的"OAuth 时刻"。提供通用 SDK 与云服务，让任意 AI Agent 可代表人类在限定、可撤销权限下执行操作，并具备加密身份与不可篡改的审计追踪。开发者可在数分钟内完成集成。grantex is the identity, authorization, and audit infrastructure for AI agents — the "OAuth moment" for the agentic internet. We provide a universal SDK and cloud service that lets any AI agent act on behalf of a human with scoped, revocable permissions, cryptographic identity, and an immutable audit trail. Developers integrate in minutes.

面向 Claude、Cursor 及 AI agent 的 GeneXus 18 MCP server，通过 Model Context Protocol 对 KB 对象（transactions、web panels、procedures、SDTs）进行读取、编辑与分析。GeneXus 18 MCP server for Claude, Cursor, and AI agents — read, edit, analyze KB objects (transactions, web panels, procedures, SDTs) over the Model Context Protocol.

面向工程团队的短期规划系统——让工程师保持专注、运行每日 Sprint 并跟踪非计划工作。构建于 GitHub Projects v2 之上（无独立数据库）：单个 Go 二进制内嵌 React UI、REST API、面向 AI Agent 的 MCP server，并通过 WebSocket 提供实时看板更新。A short-term planning system for engineering teams — keep engineers focused, run daily sprints, and track unplanned work. Built on GitHub Projects v2 (no database of its own): one Go binary with an embedded React UI, REST API, MCP server for AI agents, and live board updates over WebSocket.

隐私优先、AI 原生的 Agent，面向电商及其他场景，基于 LangGraph workflow 以低代码/无代码方式图形化创建 skill、任务和 agent。支持跨网络部署 agent。Privacy-first, AI native agents For E-Commerce (and beyond), create skill (langgraph based workflow), tasks, agents, graphically with low code or no code. Deploy agents across network.

持久化的多 agent 工作流编排——DAG 任务、git worktree、merge gate，兼具桌面端 + CLI + Slack。Persisted multi-agent workflow orchestration — DAG tasks, git worktrees, merge gates, desktop + CLI + Slack

自动化你的工作：local-first AI 任务中枢。Email、Teams、Slack 及报告 → 统一时间线 → AI 分类 → 你的编码 agent（Claude Code、Codex、Gemini）执行，你审批。Automate your job: local-first AI task hub. Email, Teams, Slack & reports -> one timeline -> AI triage -> your coding agents (Claude Code, Codex, Gemini) do the work, you approve.

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!

DebugBundle 产品源码与运行时代码。DebugBundle product source and runtime code.

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 .

AI 原生时序记忆图，采用统一 DSL——将概念、查询、智能体工作流与语音统一为可部署的原语AI-native time-series memory graph with a single DSL — unifies concepts, queries, agent workflows, and voice into deployable primitives.

通过 OAuth + MCP 将 ChatGPT Web 连接至 DSH，开放 DSH 原生工具、skills、策略与社区扩展Connect ChatGPT Web to DSH through OAuth + MCP, exposing DSH-native tools, skills, policies, and community extensions.

本地 Agent 个人 AI，本地 AI Runtime，专注于完全离线在设备端运行，将普通小型 chatbot 升级为强大的 AI，轻量而强大，用于构建自适应的个人 AI。支持多种 Backend。Local Agent Personal Artificial Intelligence, Local AI Runtime, That Focus work on device entirely offline and enchant tiny chatbot regular AI to become powerful, a lightweight yet powerful for building personal, adaptive AI. Work with multiple Backend

DSH 完整移植版 DietrichGebert/ponytail — 懒惰 senior 模式，6 个中文 Skill，无空 tool

确定性的 MCP 安全架构。以 FrozenNamespace 作为 Model Context Protocol 工具验证的可信根（Root of Trust）。Deterministic MCP Security Architecture. FrozenNamespace as Root of Trust for Model Context Protocol tool verification

Commercient Data Lake / Data Hub CLI（dlake）——为 SQL Server 数据平台即时提供 REST + GraphQL API 以及面向 AI Agent 的 MCP 访问能力。提供官方二进制发行版。Commercient Data Lake / Data Hub CLI (dlake) - instant REST + GraphQL APIs and MCP access for AI agents over a SQL Server data platform. Official binaries.

面向殡葬/保险/遗产服务机构的 To B 多租户 AI 平台。包含案件管理、审计日志、知识库、团队 RBAC、许可证及数据导出。基于 FastAPI + LangGraph 构建。To B multi-tenant AI platform for funeral/insurance/estate service organizations. Case management, audit logs, knowledge base, team RBAC, license & data export. Built on FastAPI + LangGraph.

ResearchTrack 是面向大学实验室、导师与学生研究者的学术研究实验室与文献管理平台，集中管理文献综述、独立学生论文综合以及多范围论文分配（个人、实验室级、子组级）ResearchTrack is an academic research laboratory and literature management platform designed for university labs, faculty advisors, and student researchers. It centralizes literature reviews, isolated student paper syntheses, multi-scope paper assignments (individual, lab-wide, and sub-group-wise)

面向 Claude 等 LLM agent 的 MCP server，对 CSV、Parquet、Excel、JSON 等表格数据进行 profiling。MCP server that profiles tabular data files (CSV, Parquet, Excel, JSON) for LLM agents like Claude.

opencode 的持久化记忆 MCP server：Redis 存储事实、ChromaDB 语义搜索、文档存储、幂律记忆衰减、agent 身份；内置官方 Claude.app OAuth，可使用 Claude Max 套餐且规避第三方检测；一键安装。Persistent memory MCP server for opencode — Redis facts, ChromaDB semantic search, document storage, power-law memory decay, agent identity. Plus first-party Claude.app OAuth: use your Claude Max plan without third-party detection. One-command install.

基于 RAG、LangGraph 多 agent workflow、FastAPI 与 PostgreSQL（pgvector）构建的 AI 教育内容生成平台。生产级架构，集成 LangSmith tracing。AI-powered educational content generation platform using RAG, LangGraph multi-agent workflows, FastAPI, and PostgreSQL with pgvector. Production-ready architecture with LangSmith tracing.

开放式数据库工程课程 · 14 个模块共 64 节课 · 210 小时 · 涵盖概念建模、分布式架构、运维及面向 AI 的数据恢复 · 109 项参考文献（带 ISBN、DOI 或标准号），逐课引用并在 CI 中校验 · 5 个零依赖可执行实验 · 256 题自测 📚🗄️ Programa abierto de ingeniería de bases de datos · 64 clases en 14 partes · 210 horas · Del modelado conceptual a la arquitectura distribuida, la operación y la recuperación para IA · 109 fuentes con ISBN, DOI o norma, citadas clase a clase y verificadas en CI · 5 laboratorios ejecutables sin dependencias · Autoevaluación de 256 preguntas 📚

Rust 可嵌入的混合搜索原语：BM25、HNSW、reciprocal-rank fusion、UTF-8 安全的 chunking，零依赖。Embeddable hybrid search primitives for Rust: BM25, HNSW, reciprocal-rank fusion, UTF-8-safe chunking. Zero dependencies.

用于系统文献综述的 Web 应用，类似于 Rayyan 与 Parsifal。A web app for systematic literature reviews, similar to Rayyan and Parsifal.

精简 2026 多 Agent 开发工作流：使用 SwarmUX CLI 编排 tmux 会话。Streamline Multi-Agent Dev Workflows 2026: Orchestrate tmux Sessions with SwarmUX CLI

原生 macOS 学术写作助手，通过 OpenAI 或 DeepSeek 提供 AI 驱动的语法、标点与风格检查。A native macOS academic writing assistant with AI-powered grammar, punctuation, and style checks via OpenAI or DeepSeek.

本仓库包含用于复现论文《Pharmacological treatments for ADHD and comorbid substance use disorder: a systematic review and network meta-analysis》中数据库与分析的 R 代码。This repository holds the R codes for reproducing the databases and analyses in my paper "Pharmacological treatments for ADHD and comorbid substance use disorder: a systematic review and network meta-analysis".

Swift Essay 是基于 Next.js 与 MongoDB 构建的学术写作平台，管理员可发布学术写作任务，writer 可被分配并完成。Swift Essay is an academic writing platform built with NextJs and MongoDB, where admins can post academic writing tasks and writers can be assigned and complete them.

AI 构建，你决定是否上线 —— AI agent 工作的审核关卡。AI builds it. You decide if it ships. — review gates for AI-agent work.

面向互联医疗 IoT 中可审计的两级本地—边缘 ECG PVC 风险服务回路的可复现研究代码Reproducible research code for an auditable two-tier local-edge ECG PVC risk service loop in connected healthcare IoT.

从首次检索到完成的证据表，一站式文献综述。面向需要可复现系统综述的团队，以及在同一机制上完成论文检索的学生。A literature review, from the first search to the finished evidence table. For teams running a systematic review that has to be reproducible, and for students running a thesis search on the same machinery.

免费、无需注册的顺序电子签名服务，~30 秒即可发送 NDA 或合同Free, no-signup sequential e-signatures — send an NDA or contract in ~30 seconds

基于 Claude Code 的学术写作工作台：论文、实习报告、文献综述与课程作业。包含交互式引导、引用核验、风格规则与字数预算。A Claude Code workspace for academic writing: theses, internship reports, literature reviews, and course papers. Interactive onboarding, verified citations, style rules, character budgets.

Episciences 引文管理器Episciences citation manager

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.

可复现的研究问题与预注册Reproducible research question,preregistration

prompt工具箱

面向 Claude Code / Codex 的 Agent Skill：为系统评价与 Meta 分析构建可复现、可按 PRISMA-S 报告的检索策略 —— 通过 NCBI API 获取 MeSH、主题词+自由词检索式、单行记录数、召回率校验、多数据库语法。Agent Skill for Claude Code / Codex: build reproducible, PRISMA-S-reportable search strategies for systematic reviews & meta-analyses — MeSH via NCBI API, subject-heading + free-text lines, per-line record counts, recall check, multi-database syntax. 系统评价/Meta分析检索策略构建 skill

基于 Google ADK 的多智能体文献综述系统，并行扫描预印本/同行评审/实现仓库，自评估覆盖循环，代码层面强制引文不变性Multi-agent literature review system on Google ADK — parallel preprint/peer-reviewed/implementation sweep, self-evaluating coverage loop, and a citation invariant enforced in code.

基于 Clean Architecture 的模块化网络爬虫——wreq TLS 指纹识别、RAG 导出、MCP server、TUI、AI 语义清洗，配备 1,337 项测试并通过 Miri 验证。Modular web scraper with Clean Architecture — wreq TLS fingerprinting, RAG export, MCP server, TUI, AI semantic cleaning. 1,337 tests, Miri verified.

Cloudflare Code Mode MCP 的 global bundle，面向 DeepSeek Harness（DSH）——包含 mcp__cloudflare__* 工具、cfbridge skill 与 wrangler passthrough。Cloudflare Code Mode MCP global bundle for DeepSeek Harness (DSH) — mcp__cloudflare__* tools + cfbridge skill + wrangler passthrough

每日课程：面向从未打开过终端的用户的 Claude Code 和 Codex 教程。面向企业主而非程序员。内容来源于实际运行的 AI agent 集群，而非理论。Daily lessons: Claude Code and Codex for people who have never opened a terminal. For business owners, not programmers. Written from a running fleet of AI agents, not from theory.

数据驱动、自动验证的机器人研究文献综述语料库（manipulation、locomotion、perception、planning、learning、HRI、multi-robot、simulation）。Robotics Research Corpus: Data-driven, auto-validated literature review for robotics research (manipulation, locomotion, perception, planning, learning, HRI, multi-robot, simulation)

音乐研究语料库：面向音乐生成、MIR、感知与认知、AI 音乐系统、评估的 agentic 文献综述 — 覆盖符号、音频、多模态与 agentic 方向Music research corpus: agentic literature review of music generation, MIR, perception & cognition, AI music systems, evaluation — symbolic, audio, multimodal, agentic

贝叶斯统计研究语料库：面向贝叶斯统计（推断、计算、先验、模型选择、层次模型、非参数、深度贝叶斯、应用）的数据驱动、自动化校验文献综述Bayesian Statistics Research Corpus: Data-driven, auto-validated literature review for Bayesian statistics (inference, computation, priors, model selection, hierarchical, nonparametric, deep Bayesian, applications)

一个面向 Codex custom subagent 的纪律严明、跨平台的编排系统。A disciplined, cross-platform orchestration system for Codex custom subagents

一个 Claude Managed Agent，可回答 SlideRule 研究问题并返回独立的 HTML 仪表板；每个用户需在其自己的 Anthropic 账号中部署专属 AgentA Claude Managed Agent that answers SlideRule research questions and returns a self-contained HTML dashboard. Each user provisions their own agent in their own Anthropic account.

研究问题：在脓毒症发生前检测患者 — 使用分类算法实现脓毒症的早期检测Research Question: Detect Sepsis in Patients Before it Emerges - Early Detection of Sepsis using Classification algorithms

将混乱的需求转化为 coding agent 一次就能做对的 prompt。一个适用于 Claude Code、Cursor 以及任何能阅读 markdown 的 agent 的 skill。Turn a messy request into a prompt your coding agent gets right the first time. A skill for Claude Code, Cursor, and any agent that reads markdown.

论文一：NL2SQL 歧义检测——M1.5 分类法、确定性基线、人工金标评估与可复现研究制品Paper 1: NL2SQL Ambiguity Detection — M1.5 taxonomy, deterministic baseline, human-gold evaluation, reproducible research artifact

面向 CBSE、ICSE、AP SSC（1-10 年级）的 AI 备考平台 —— 基于知识点先决关系 DAG 的自适应练习、行为分析、实时竞技房间，以及减少绝大多数 AI 调用的 7 层缓存。AI-powered exam prep for CBSE, ICSE and AP SSC (Grades 1-10) — adaptive practice over a topic-prerequisite DAG, behavioural analysis, live competition rooms, and a 7-layer cache that keeps most AI calls from ever firing.

常态化跨领域研究智库：系统评价与方法学文档，特色方法是对零结果与阴性结果的取证式处理。A standing cross-domain research think tank: systematic reviews and methodology documentation, with forensic treatment of null and negative results as its distinctive method.

针对任意 GitHub 仓库的自然语言问答 —— 构建 Neo4j 依赖图与 Qdrant 向量索引，用自然语言即可提问结构性与行为性问题。Natural-language Q&A over any GitHub repo — builds a Neo4j dependency graph + Qdrant vector index so you can ask structural and behavioral questions in plain English

企业级 Java AI Agent 任务编排、安全审批、RAG 与全链路审计平台

只读型 Hyperliquid SP500 反转扫描器，配套 Discord 告警与可复现研究工具Read-only Hyperliquid SP500 reversal scanner with Discord alerts and reproducible research tooling.

计算两位独立评分者在系统评价所用 5 项方法学严谨性标准（检测性能、理论基础、计算效率、可解释性、生态相关性）上的 Cohen's kappa（非加权与二次加权）、一致率、平均绝对差及 Spearman 相关系数。Computes Cohen's kappa (unweighted & quadratic-weighted), percent agreement, mean absolute difference, and Spearman correlation between two independent rators' scores across 5 methodological rigour criteria (detection performance, theoretical foundation, computational efficiency,interpretability, ecological relevance) used in the systematic review.

本地优先、隐私至上的 AI 驱动学术写作人性化工具。AI-Powered Academic Writing Humanizer - Local-First, Privacy-Centric Solution

关于备灾基础设施、离线连接、开源硬件与分布式制造的持续陪伴式资料：框架、案例研究、扩展文献综述与建议。Living companion on preparedness infrastructure, offline connectivity, open hardware, and distributed manufacturing: framework, case studies, extended literature review, and recommendations.

Kiki — 一个面向 Google Campaign Manager 360 的开源对话式 trafficking 助手。集成 70 个 CM360 工具，具备写入安全确认、内置 QA 以及零凭证 MCP demo。Kiki — an open-source conversational trafficking assistant for Google Campaign Manager 360. 70 CM360 tools, write-safety confirmation, built-in QA, zero-credential MCP demo.

CryptoMindAI 是一个面向加密货币市场量化分析的开源研究框架，结合系统化策略、统计分析、回测、样本外验证与 AI 辅助研究；项目聚焦可复现实验与方法学稳健性，而非单纯生成交易信号CryptoMindAI is an open-source research framework for quantitative analysis of cryptocurrency markets, combining systematic strategies, statistical analysis, backtesting, out-of-sample validation and AI-assisted research. The project focuses on reproducible experimentation and methodological robustness rather than simply generating trading signals.

农业机器人成像系统综述与 meta-analysis 的数据与分析代码。Data and analysis code for a systematic review and meta-analysis of agricultural robotic imaging

一篇文献综述（Quarto 书），针对依赖效应量下的贝叶斯元分析稳健方差估计，论证 Cholesky 校准修正（CCC）A literature review (Quarto book) of robust variance estimation for Bayesian meta-analysis with dependent effect sizes — the case for Cholesky Calibration Correction (CCC)

一个可复现的研究原型，针对漂移感知欺诈检测，采用 actor-critic 强化学习、动态特征重加权与自适应回放，应对交易分布的持续演化A reproducible research prototype for drift-aware fraud detection using actor-critic reinforcement learning, dynamic feature reweighting, and adaptive replay under evolving transaction distributions.

领域系统性调研体系：基于 bibliometrix 的文献计量分析 + 思想流派演变/冲突缺口/未来方向 + 两轮验证系统性报告（R/Shiny）

可审计的 Codex skill 与 Python 项目包，用于证据可追溯的研究、系统综述、投稿与修订。Auditable Codex skill and Python project pack for evidence-traceable research, systematic reviews, submission, and revision.

面向解耦归纳式时序链接预测的可复现研究，附带校验和闭环的证据Reproducible research for decoupled inductive temporal link prediction, with checksum-closed evidence.

Research Loom 是 profile 驱动的 research harness，用于维持从研究问题与证据到发现、写作与发表的全链路可追溯性Research Loom is a profile-driven research harness for maintaining traceability from research questions and evidence through findings, writing, and publication.

面向老版视觉小说的可复现研究、解包工具与归档清单Reproducible research, unpacking tools, and archival manifests for legacy visual novels.

一个开源的 Agentic 开发环境：在终端网格中并行运行 Claude Code、cursor-agent、Codex 与 shell，配合用于任务分派的看板、共享 Agent 记忆以及多 Agent 编排能力。An open-source agentic development environment: run Claude Code, cursor-agent, Codex and shells side by side in a terminal grid, with a kanban board that dispatches them, shared agent memory, and multi-agent orchestration.

面向 LLM agent 的 MCP server，用于加拿大调整后成本基础（ACB）与资本利得：支持平均成本法与 superficial-loss 检测。MCP server for Canadian adjusted cost base (ACB) and capital gains: average-cost, superficial-loss detection, for LLM agents.

一个独立开发的板块轮动预测项目，2026 年重构为含回顾性方法学审计的可复现研究工具包An independently developed sector-rotation forecasting project, rebuilt in 2026 as a reproducible research toolkit with a retrospective methodological audit.

GMS 6803 临床研究数据科学课程材料，含幻灯片、代码与实操实验，内容涵盖真实世界健康数据、基于 EHR 的队列构建、SQL、Python、统计分析与临床研究中的机器学习Course materials for GMS 6803 Data Science for Clinical Research, including slides, code, and hands-on labs on real-world health data, EHR-based cohort development, SQL, Python, statistical analysis, and machine learning for clinical research.

为 AI agent 搭建团队知识仓库——集中领域知识、工作流与规范，供 agent 消费、基于其开发，并通过 PR 循环持续保鲜。Scaffold team knowledge repos for AI agents — centralize domain knowledge, workflows, and specs that agents can consume, develop with, and keep fresh via PR loops.

面向精确 MCP 能力发现、有界调用与提供者效应验证的 Agent Skill。Agent Skill for exact MCP capability discovery, bounded invocation, and provider-effect verification

研究问题：流动性金融资产的相关性与波动率随时间有多稳定？常用的高斯性质在真实市场数据中是否成立？Research Questions: How stable are correlations and volatility over time in liquid financial assets? Do commonly assumed Gaussian properties hold for real market data?

分析葡萄牙公共养老金体系历史融资的可复现研究包Reproducible research package analyzing the historical financing of Portugal's public pension system

MMLA 在发展中经济体与 LMIC 中的综合系统综述。A Comprehensive Systematic review of MMLA in developing economics and LMIC.

论文 "Composite endpoints sharing a name do not always share a definition — a registry-based audit of five systematic reviews, and why heterogeneity statistics do not detect the problem" 的配图与补充材料。Figures and supplementary materials accompanying the manuscript “Composite endpoints sharing a name do not always share a definition — a registry-based audit of five systematic reviews, and why heterogeneity statistics do not detect the problem.

分配一个 AI 编码 agent 处理真实 issue，审阅其提出的方案，获取签名且可审计的 pull request。当前为单人模式，设计上支持扩展为多人协作模式。Assign an AI coding agent to a real issue, review the plan it proposes, and get a signed, auditable pull request back. Single-player today, designed to grow into shared, multiplayer use.

面向 MetaTrader 5 的可配置开盘区间突破 EA，附带可复现的研究 harness；所有参数均为输入项，兼容 Windows 与 Linux/WineConfigurable opening-range breakout EA for MetaTrader 5, with a reproducible research harness. Every parameter is an input; runs on Windows and Linux/Wine.

Claude Code Skill，可将论文与对话转化为交叉链接的 Obsidian 知识库：十段式论文笔记含内嵌图表，归档于主题枢纽下Claude Code skills that turn papers and conversations into a cross-linked Obsidian knowledge base — ten-section paper notes with inline figures, filed under topic hubs.

一个 agentic、循证的研究系统：通过工具调用、RAG 与显式的论据—来源核验从研究问题生成结构化、带引用的报告，避免盲目的 LLM 摘要；能识别来源冲突并如实报告不确定性An agentic, evidence-grounded research system that generates a structured, citation-backed report from a research question — using tool-calling, RAG, and explicit claim-to-source verification instead of blind LLM summarization. Detects conflicting sources and reports uncertainty rather than hiding it.

本仓库收录了我关于超透镜（metalens）的研究日志与学习笔记，主要用作个人记录空间，用于整理讲义、代码与文献综述；任何对平面光学感兴趣的人也可使用。This repository contains my research logs and study notes on metalenses. It serves primarily as a personal bookkeeping space to document lecture notes, code, and literature reviews, though anyone interested in flat optics is welcome to use these resources.

个人 AI 工程博客与作品集——12 篇文章，覆盖 agentic AI、RAG、embeddings、chunking、token 及生产实践。基于 Astro 构建，部署于 Cloudflare Pages。Personal AI engineering blog and portfolio — 12 posts covering agentic AI, RAG, embeddings, chunking, tokens, and production patterns. Built with Astro and deployed on Cloudflare Pages.