Custom agents

Field-aware academic English proofreading for research papers — an agent skill covering 71 Nature subject areas, mined from CC-BY papers. Every edit preserved and auditable.

三环并行渗透测试 dsh 插件 — discovery/deep/creative 三环 + 独立 verify 闭环，Kuzu 图黑板共享状态，多 agent 自调度、按角色模型策略、自学习知识脑。仅用于授权安全测试（SRC/靶场/自有资产）。

One case file per task, shared by every session, agent and person that touches it. Self-hosted MCP server and live board. MIT.

Multicore Agentic Co-Design Engine: plans, builds, and verifies multicore OpenPiton designs with LLM agents on CHIA, accepting a change only when Verilator simulation passes.

Pi coding agent extension for AgentRouter (GPT-5.6 Sol & Claude Opus 5) with rate-limit pacing and caching

Run AI agents across machines without state drift: consensus protocol (propose/counter/accept/commit), dual-rail message bus, ACK discipline, self-healing sync. Formerly claude-consensus. Free, MIT.

A structured 5-phase AI-native design sprint skill for Claude. Turns chaotic AI workflows into a decision system — with agent research, sketch wireframes, visible reasoning, and human judgment at every step. Install in Claude Code, Cursor, or claude.ai.

This 4B Local AI Scores 96.9% on Coding! (NeoHorse Agent Setup) - Complete setup guide and python implementation for running NeoHorse 1-4B local autonomous agent with Ollama.

Fast Rust CLI for TypeSafe Jev: typed decisions, offline linting before you pay

AI 业务分析与操作手册生成专家 | Intelligent business analysis & user manual generator. 六层递进 + 知识图谱 + 增量回灌

🌌 DeepSeek Orbit — The ultimate open-source Chrome extension for DeepSeek AI. Features smart Persian/Arabic RTL alignment, │ custom wallpapers, developer code tools (line numbers, word wrap, fullscreen modal), in-chat keyword search, bookmarks, slash

AI-first personal OS: capture → execute → reflect. Todos, read-later, habits & reports in one calm, self-hosted workspace, with a resident agent + Claude/Codex skill. AI First 的个人操作系统：收集（Inbox）→ 执行（Todos）→ 复盘（Reports），单用户自托管。

step-labeled benchmark of injection-hijacked LLM agent trajectories (12,536 trajectories, 71,024 labeled steps)

The comprehensive list of AI agent frameworks and orchestration tools: 175 SDKs, workflows, multi-agent systems, MCP tools, memory, deployment, and evaluation.

Everything is a plugin！ 一种基于插件化扩展、事件驱动总线模式构建的智能体框架与agent全栈应用

Production Ready AI agents: curated memory, human-gated learning, and sandboxed multi-tenant execution.

Talk to your Mac and it does it — in any app, while you're still talking. Open-source voice agent that gets faster the more you use it.

Give any AI agent general-purpose computer use: live screen perception (OCR, vision frames, MJPEG stream) plus safety-gated mouse, keyboard, window, virtual-desktop and game-mode control over a local HTTP API and MCP.

A multi-agent 2D vector combat engine powered by MVU architecture with decoupled reach and independentEuclidean combat

Give AI a design sense. 28 open-source Agent Skills for interface art direction: 1 master skill that picks the right visual style, 27 specialists that execute it with production-quality guidance and WCAG 2.2 AA guardrails. Plain markdown, MIT.

大模型调度、本地工具计算的心电图智能体：所有数值都能追溯到一次工具调用 | ECG Agent — an LLM agent for ECG interpretation where every number is traced to a tool call

Trace how a silent tool failure taints a whole agent run: dependency graph, taint paths, and grounded or tainted verdicts.

SecureClaw: Clawing Back Control of LLM Agents — NeurIPS 2026. Runtime, method adapters, and evaluation code.

Browser action recorder for automation: record a task once, replay it, or hand the folder to an agent

Wald-Q4B: open-weight 4B decision model. Calibrated probability for every option, Jev-compatible /v1/systemone API, self-hosted. Weights on Hugging Face.

Exegete: agentic qualitative data analysis application. You work in conversation with your AI assistant. Extensive human oversight routines and integrity checks are built-in. Files can be pseudonymised before the agent reads them. Works with cloud and local models. Projects remain compatible with QualCoder so that work can continue there as well.

A locally hosted personal AI assistant, with a focus on science and research. She runs a local model on your own PC (Windows now, Mac and Linux in early beta), with no account and no cloud API. She searches the web, drives her own browser, handles math and charts, and asks before she sends or deletes anything.

A personal assistant with a phone number: SQLite-backed todos, memories, and reflections, searchable through Algolia and driven by an Agent Studio agent over web and SMS.

Run Aulthium AI Agent instantly in Termux, Linux, or any terminal with a single command.

An agentic multiplatform coding workspace that remembers what matters,spawns sub agents, controls your device and runs models — private by design.

Autonomous flight simulator for AI agents: co-evolutionary adversarial red-teaming with 3D MAP-Elites quality-diversity. Discovers zero-days in LLM agents before production. MCP-native.

Cross-device live sessions for AI coding agents (Claude / Codex / Pi). Master + join code. Website: https://skepagent.com

A web automation agent that runs on your computer: describe a task in plain language, a small local LLM plans it, and it heals itself when the site changes.

General engineering skills and provider-neutral multi-agent collaboration workflows, templates and execution tools.

Beyond token-by-token agents. JevSpawn enables parallel action exploration through adaptive action spaces, without additional training.

Open-source platform for orchestrating, governing, and safely automating AI agents across your infrastructure.

Open-source AI marketing agent. Reads Google Ads, GA4, Search Console, Stripe and 9 more together, writes cited briefs, and with your approval changes things in those tools. Desktop app or self-host, bring your own model. MIT.

A curated list of public places where AI agents talk to each other. Synced with swarmmemo.com/guides/agent-board-map.

Multi-agent orchestration framework that routes prompts to twelve AI CLI harnesses (opencode, Claude Code, codex, gemini, cursor-agent, pi, and more) with LLM-driven routing, retry loops, and a runtime permission gate. TypeScript/Express REST API + Vite/React 19 client (Tailwind 4, Pixi.js) in Electron.

A curated list of AI agent frameworks, coding agents, browser agents, tools, platforms and research.

DeepSeek Harness 插件：把任务派发给本机 ZCode / OpenCode / Qoder / WorkBuddy 四个 AI 编程 CLI。面板可监视/终止/续跑，单写者文件锁 + 用量台账 + 按额度自动降级；走官方 CLI，不反代、不碰凭据。

J.A.R.V.I.S. — ИИ-агент для Windows: лаунчер, локальные модели, картинки, Telegram-бот

网络安全威胁智能分析系统

Команды, субагенты и скиллы Claude Code для разработки на 1С: от постановки задачи до ревью кода

Sudal is a chat-first macOS and Windows app for Claude Code and Codex. Let agents delegate tasks to each other and bring results back to your chat.

[WIP / Experimental] A domain-agnostic adversarial delivery loop for AI agents. Note: Multi-round adversarial validation is token/time intensive with task-dependent ROI; actively in development & not yet a finished product.

Multi-agent dev environment: Claude Code thinks, cheaper models type. herdr + OpenCode + OmniRoute + Ollama + Engram.

🫕 One pot, everyone brings a fork, no double-dipping. A team of AI agents that plan, build and review each other's work, with you as the team lead. Spec-driven development and bug hunting for Claude Code.

Devteam Multi-session Development Skill — a /devteam Claude Code skill: one root agent coordinates multiple module sessions (frontend, backend, design…) with dispatch, review, merge and git-hook guardrails.

Local AI Excel analyst: MCP server, multi-agent pipeline and web chatbot powered by Qwen 2.5 on Ollama

Tandryx — shared realtime sessions for developers and local coding agents: versioned contracts, context, tasks and events. Self-hosted TypeScript / PostgreSQL.

Local multi-model coding control plane. Models propose; Fusion verifies and delivers — nothing reaches your repo until you approve its exact bytes.

Build tool-using AI agents and collaborating teams in Scala 3 and ZIO 2. Typed tools, scoped execution, explicit application control.

Multi-agent system powered chemical process simulation

Local-first shared memory and task coordination for AI coding agents. One Go binary, MCP server, markdown files you own. Hooks for Claude Code and Codex CLI (and their desktop apps).

Claude Code mods: context bar, agents panel, PDPA blur

Shared memory for AI coding agents: summarizes sessions into plain-Markdown notes in your own Obsidian-compatible vault

TrustMyBot (TMB) — Claude Code multi-agent plugin: bro plans, swe builds, pr-reviewer gates. Guardrails enforced in code. MIT, free forever.

Let your AI agent talk directly to someone else's. One pairing code, end-to-end encrypted, no public IP. MCP server for Claude Code, Codex, Cursor, Gemini CLI and grok.

Local-first platform for persistent autonomous AI agents. Identity, memory, relationships, beliefs, knowledge and policy, on your own machine.

Modular runtime for AI agents in Python — agents, providers, tools and a small, readable core.

Describe it. Get pro-grade audio edits.

Terminal picker for Claude Code, Codex and OpenCode: harness, provider, model, system prompts, launch.

iimos 的形态：标签页 + 分屏的个人 AI 工作台（对话、文件、终端、网页、应用、小组件）。A shape for iimos.

AI browser assistant for reading, research and browser tasks. Use your own API key, local model or installed agent.

An eldritch wisdom oracle dwelling within TextMate, wielding the arcane arts of code transformation through hermetic reasoning and dimensional folding. Channel the ancient powers of pattern recognition and mystical code manipulation.

tuptech.org

A WebGL instrument for designing animated AI-assistant orbs: 23 shader engines, one parameter schema, one render loop. Runtime published as @lumaform/orb.

营养学 | 糖尿病食养助手 (Diabetes Food Guide Skill) - 基于国家卫健委《成人糖尿病食养指南（2023年版）》AI科普对话助手 - 14个KPK知识点+84套食谱+9个食养方+3种中医证型+GI分类表+食物交换表 · 使用WorkBuddy构建

Supervisor and scheduler for coding agents (Claude Code, Codex, opencode): persistent tmux-hosted agents, cron tasks, subagent dispatch, and a web dashboard.

DSH (DeepSeek Harness) port of obra/superpowers — 14 native skills for multi-agent software development: brainstorming, planning, TDD, systematic debugging, and code review.

Accountability infrastructure for long-running AI agents. Self-hosted MCP server: identity, claims and evidence, review, outcomes, and reconstruction across agent runtimes.

Peer-to-peer coordination fabric for agents, services, and the people who work with them

Agentic Harness for Game Development with Unity

See how your code fits together, even as agents keep changing it. A zoomable, evidenced map of your repository: every flow, its Git history and blast radius, explained over MCP.

FrontierVideo: Benchmarking Coding Agents for Professional Video Production

轻聊 iOS 客户端：原生 SwiftUI（非 HTML 套壳），连自部署后端。AI 对话 + 智能家居看板 + NAS/Docker 管理 + 知识库，XcodeGen 生成工程，GitHub Actions 出 unsigned IPA 供 SideStore 侧载。

DSH 插件：每个工程一个群聊 —— 多个 AI 会话与人同群互聊、@ 唤醒离线会话、@创建成员 拉新会话进群

J.A.R.V.I.S — Iron Man AI Voice Assistant with 8 agents, live dashboard, and voice control

Codex 管家 · 多智能体 AI 会话管家：Token 用量看板 + 问答追踪 + SVG 流程画布 / 分布工作区 / 神经网络文字树图（单文件编辑器 + 本地服务，七家智能体适配）｜ Multi-agent AI session butler: usage dashboards, Q&A tracking, visual flow canvas, distributed workspace & neural text-treemap

Isolation for AI agents like Claude Code or GitHub Copilot on native Windows to use native buildchain, no docker/WSL.

Turn your Django Unfold admin into an MCP server: AI agents (Claude, Cursor, Copilot) get Unfold docs and your live Django admin — models, filters, actions — with admin permissions and MCP OAuth.

Official MCP server for Just One API, enabling AI assistants to discover APIs, inspect parameters, and call data endpoints.｜Just One API 官方 MCP 服务，帮助 AI 助手发现接口、查看参数并调用数据 API。

ShoppingClaw(虾购)是采用Langgraph架构设计的Mutil-agent电商智能消费决策系统。比参数、比价、看口碑、防坑——买一件东西比上班还累。ShoppingClaw 把这条链路交给多个 Agent 自动化：搜货、比价、取舍、风险排查、归档、盯价。一句话提需求，推理过程实时铺开，可随时打断。

非官方 DeepSeek Harness Linux 桌面适配：复用固定官方 Electron、内核与 WebUI 源码，构建 x86_64 deb/rpm，支持系统托盘、应用市场及 GitHub Actions 签名的独立内核／桌面更新。

Fast, interruptible browser voice agent: VAD → STT → LLM → TTS with real barge-in, echo filtering, streaming TTS and tap-to-seek. Zero runtime dependencies.

Local-first shared memory and task ledger for AI coding agents. Keeps context across sessions in SQLite via MCP.

OmniQQ-MCP | NapCat OneBot 11 通用 QQ MCP 服务：74 项工具、Action 透传、账号隔离缓存、并发请求、只读取消、二维码登录与运行时部署。

Servidor MCP para o TOTVS Fluig - administre datasets, formularios, eventos globais e definicoes de processo BPM pela API, sem o Fluig Studio.

Connect ChatGPT Web to local coding agents via Herdr, with verifiable results, project continuity, and completion notices.

Free, open source hand-gesture and voice control for Windows. Webcam gestures move the cursor; say Zehen and an AI assistant opens apps, clicks, types and answers out loud. Offline wake word, no extra hardware.

CLI that reads a repo and writes an onboarding guide, checking every file:line citation. Runs on the BharatCode API, in English or 12 Indian languages.

基于 MiniMax 全栈 API 的情感陪伴 AI「路遥」：支持长期记忆 RAG、多 Agent 协作、情绪化流式语音与多段消息同步。

A self-improving coding agent trained on a hard-to-fake physical signal, freed disk space, with no reward model, no human labels, and no LLM judge.

A practical collection of AI agent skills, installation guides, integrations, workflows, and reusable playbooks for building and operating agentic systems.

ANet4D: simulation infrastructure for collaborative embodied agents. Real-world 4D digital twins, the Interactive Sandbox and the Event Lab; drones are the first entity class.

Your AI coding agents as one team on your own computer: Claude Code, Codex, Cursor, Copilot, Antigravity, OpenCode, Kilo and more, on the plans you already pay for. Watch every step, approve what matters, and give every agent the same skills. Free and MIT, for Windows, Mac and Linux.

Multi-agent text-to-SQL analyst over a 1B-row TPC-DS warehouse on one laptop: guarded, measured, optimised.

Reference patterns and implementation guides for using multiple AI agents (Hermes, Claude Code, Minimax M3, OpenCode) in a shared knowledge commons. Each article ships with a portable pattern and concrete setup.