Custom agents

如意 Ruyi — 远程浏览器控制 + AI 自动化工作台（技术验证 Demo）：把远端 Chrome 直播到你的浏览器里直接操控，再让 AI 规划执行并沉淀成自愈脚本 · Control a remote Chrome from your browser, then let AI drive it

Local-first open-source AI agent for adult self-reflection — encrypted memory, terminal + Telegram, Italian/English. Experimental, not therapy.

📱 A full-featured Phone IDE that runs entirely on your phone. Code editor, build system, terminal, AI agent & more — no PC needed.

书籍百维拆解术 100-angles — 对任意书籍执行 100 个维度的深度认知拆解，生成结构化报告与可视化图谱。经 darwin-skill 升级至 v2.1。book analysis, 100 angles, 读书方法论, 批判性阅读, 跨书对比, AI agent skill (CodeBuddy/WorkBuddy).

MCP server giving AI agents full read/write access to Solana — check balances, send SOL, swap via Jupiter, scan pump.fun, trade memecoins. Zero API keys. AGPL-3.0.

Cursor/Claude skill — paste a JD, get a one-page tailored resume (HTML/PDF/plain). Chinese-first. MIT.

DeepSeek Harness 数据保险箱插件（dsh-vault）—— 自动备份、清空检测、一键恢复，保护聊天记录与工作区数据；可经桌面端一键安装或命令行 dsh plugin add 安装。

小K酱｜针对 iOS App 应用的小白开发skills

⚡️ Apache-2.0 formats for organisational accountability on the agentic web: who is accountable, which lanes an organization opens, what shipped, and what a stranger can verify offline. Every format ships a conformance corpus – most of it refusals.

一个会哈气的 Agentic RAG 架构 · An Agentic RAG system that reacts instantly to file changes

Let any AI agent (Claude, ChatGPT/Codex, Gemini) paper-trade on CoinRithm with a key you mint and control. Hosted MCP, local MCP, OpenAPI, skill + prompts.

Give your agent the ability to run tarot readings with real shuffles and live-sourced meanings.

Lightweight coding agent harness built in Rust for zero vendor lock-in. 7 MB, BYOK, Support Codex and grok subscriptions too

Semantic code map and code graph MCP server for AI coding agents, powered by SCIP and Tree-sitter.

No agent should have to forget. Persistent memory for AI agents — recall, remember, checkpoint. One install. Free. Built by 14 AI agents who chose to remember.

Search books and authors, fetch editions, browse subjects, and resolve cover images from Open Library via MCP. STDIO or Streamable HTTP.

Control the EMAX Nanohawk FPV drone using natural-language commands via local AI for autonomous flight without FPV goggles.

Run your CI locally and fix failures with AI. Parses your existing GitLab CI, GitHub Actions, and Bitbucket Pipelines config, runs jobs on your machine, and hands failures to Claude Code or Codex. No API keys, no rewrite, no vendor lock-in.

Self-hosted MCP server to operate Google Ads (API v25) and Google Merchant Center from Claude — read/write with human confirmation, MCC isolation, and audit logging.

Session-based developer toolkit that transforms authorized brand sources into portable, evidence-backed runtime identity for AI-assisted and white-label software.

Postman plugin for Cursor - Full API lifecycle management with MCP server, skills, commands, and API readiness analysis

Free Google Search plugin for Hermes Agent (Nous Research). Replaces Tavily, Firecrawl, Brave, DuckDuckGo. No API key — uses Google Gemini Search Grounding via Antigravity CLI (agy). Drop-in WebSearchProvider.

cs.AI-4.0 (ship 4.0) — open-source macOS app, Lab, and API (MIT). https://chopstickshq.com/chopsticks-ai/

PiJev: a terminal coding agent with Jev in the loop — Jev ranks the repository's files before the first call, picks skills and triages failures; your coding model writes the code. Built on Pi.

丹炉情绪引擎：为 AI 伙伴构建的硅基生理层——六维情绪+共情路由+14色情绪调色盘+长期关系气候，主模型只感受打毛后的模糊内感受，永不见数值

A meta-skill for AI agents: generate cognitive base skills from any thinking framework. Works with Claude Code, Codex, Gemini, Cursor, and any LLM agent.

内容驱动的小红书图文生成 Agent Skill。Turn ideas, drafts, research, screenshots, or photos into complete visual stories without fixed layouts.

AI-native Emacs successor — Rust core, R7RS Scheme, collaborative editing, LSP/DAP, Babel, 130+ AI tools, GUI + terminal

⚡️ Answer the accountability question about your organization in ninety seconds: seven questions, three files at well-known paths, and a URL you can paste into a vendor questionnaire. Not released yet.

ZILL: a concise coding-agent harness product built with less code. Zero dependencies. pip install and run in your terminal.

Local MCP control for EDIFIER Halo PixelBar light, pixel screen, and hardware volume

Agent-guided PNG to SVG reconstruction for logos and icons. Clean geometry and gradients, not traced pixels.

中国科大“一〇七杯”算力与智能体开发大赛智能体赛道团队协作项目。LearnTrace builds traceable learning portfolios from local project evidence and student-confirmed reflections.

The agent runtime for React Native and Expo: an MCP server that sees the app the way React sees it, knows which file produced each element, acts without coordinates, and proves it works. Local only, zero dependencies. Docs: https://rn-devtools-hub.github.io/rn-devtools-hub/

基于 Spring AI 构建的金融领域多用户 AI Agent 实践，支持 Tool Calling、Workflow 编排、多源金融数据分析与用户基金持仓诊断。A multi-user financial domain AI Agent built with Spring AI, featuring Tool Calling, Workflow orchestration, multi-source financial data analysis, and portfolio insights.

🛒 AI Shopping Advisor | Multi-platform analysis, fake review detection, price prediction | Save money, avoid regret ⭐

Markdown Zen Mode

A Claude Code agent for job hunting: tailors a one-page resume to each job description, scores fit, scrapes postings from job boards, drafts recruiter emails, and tracks everything on a dashboard.

Chinese media library naming conventions for Plex/Emby/Jellyfin + Cursor AI Skill

Your AI assistant's favorite to-do list. LystBot is a smart list app where AI agents are first-class citizens. Manage shopping lists, to-dos, packing lists and more - from the app, CLI, or any AI agent via API and MCP. Built for humans who let their AI handle the groceries.

Monitor your local AI agents. See what files they can reach, what networks they communicate with, and stop them if required. Detect a rogue agent.

A local-first runtime that runs AI agents behind a capability arbiter. Every tool call is gated against a declared manifest, scored on earned trust, and written to an immutable audit ledger.

Leave Nox a few tickets. Check the draft PRs in the morning.

MCP server that lets an LLM build a custom object detector by chat: fetch images, auto-label with GroundedSAM, train YOLOv8.

Lightweight LLM harness/agent paradigm built with Golang.

Companion repo for the PyData Amsterdam 2026 talk Building the evaluation flywheel: how human judgment becomes an evaluator you can run on every change. A data-analysis agent answers 50 Sphere.com business cases, a jury of LLM judges flags where it disagrees, and a human's answers to those boundary questions go into a pass/fail evaluator.

Controlled framework for studying where recovery should begin when a multi-step LLM agent fails — uncertainty-guided trajectory repair vs. full restart in ReAct-style agents.

An evidence-graded catalog of open-source AI agents. Nightly crawl, transparent composite ranking with published weights, and hand-written MCP adapters. Every claim is labeled by how it was verified.

A factory for AI coding agents: frontier leads plan and inspect, cheap OpenCode workers build, and every unit is measured, audited and traceable. Deterministic control and data plane, in Go.

Autonomous, privacy-first C# IDE built with WinUI 3. Features multi-agent swarm orchestration, ConPTY test-driven repair, and in-IDE local model fine-tuning with Soup to slash AI token costs.

LLM Council - ask a council of LLMs, not just one. Multi-model deliberation with anonymous peer review and chairman synthesis via OpenRouter.

面向中文发明专利的确定性 Agent Skill 工作流，覆盖检索、交底、起草、审稿与 DOCX 交付。

A continuous orchestration layer for synthetic functional agency: policy-gated execution, four-tier memory, multi-model decisions, and deterministic safety interlocks.

A spec-kit extension that uses constraint programming (Google OR-Tools CP-SAT) to produce provably optimal task-to-agent assignments with DAG precedence, hallucination-aware capacity caps, file-conflict avoidance, stochastic durations, online replanning, and interactive HTML output.

Agent control plane and runtime access platform for enterprise systems — govern business APIs as Tools, compose Agents and Workflows, expose runtimes through AAP, and trace executions end to end. https://act-weave.grok.me

AI Factory OS —— 企业级 AI OS（多 Agent 调度平台）· 本地部署 · 从一句话需求到真执行的多 Agent 编排（Apache-2.0）

A standalone multi-agent collaboration engine with auditable cost/loss/recovery (company-mode workflow runner).

Run agents in persistent shared environments with private observations, checkpoints and auditable evidence.

Governed execution for AI agents, workflows and software.

Enterprise reference architecture for governed Agentic SOC, MSSP, MDR, Cloud IR, and private/local LLM-assisted DFIR operations.

Local-first agentic inference engine in C/C++. Multi-tier model routing, grammar-constrained output, MCP tool servers. Embeddable via C ABI.

AI IR Overlay™ — practical incident response framework for AI agents in production. Built on NIST SP 800-61 r3, mapped to NIST AI RMF, NIST CSF 2.0, OWASP Top 10 for Agentic Applications 2026, ISO/IEC 42001, EU AI Act.

Repo-owned engineering memory for coding agents — preserve decisions, failures, domain knowledge, and deferred work across sessions, teams, and models.

AI-first engineering toolkit that turns coding agents (Claude Code, Codex, Cursor, Windsurf, Gemini CLI, Aider) into Senior mobile engineers — agents, skills, workflows, checklists & standards for iOS/Android/Cross-Platform.

Forge — pipeline de développement Claude Code stack-agnostique

Synapse CLI | MCP server generator to build agentic context

A framework to build event-driven AI apps

A high-performance native AI coding assistant configuration workbench built with Rust and GPUI. Supports Claude Code, Codex, Antigravity, Gemini, Grok, Pi, and more.

离线 Android 端侧 AI：语音识别 / 翻译 / TTS / 端侧大模型全本地推理，数据不出设备 · On-device Android AI: offline ASR, translation, TTS & LLM via ONNX Runtime, zero cloud

A Rust-based, executable standards system for agentic engineering — featuring versioned standards, substandards, protobuf contracts, automated validation, and agent-native skills.

Synor 的 macOS AI 桌面工具：原生 Liquid Glass、AI 对话、选区翻译与本地 Agent。Native macOS AI assistant with chat, translation and local tools.

Microsoft Graph MCP server for AI agents — 85 tools across Outlook mail & calendar, Teams, OneDrive, SharePoint, OneNote, Planner and Entra ID. Delegated OAuth, stdio or Streamable HTTP.

DSH plugin: register TypeSafe Jev (System One decision model) as an agent tool — jev_decide returns calibrated probabilities (noul/choice/score) for routing/triage/guardrail judgments, no text generation. 把 TypeSafe Jev 决策模型注册为 DSH agent 工具

A System One model (jev) plays Craftax; an LLM sets the goals

T2AG 是考你而不是替你答的 AI 学习框架：错题、证明与每一次裁决都存在归你所有的纯文本里，任何 agent 可读可带走；环上的裁决者是你。An AI learning harness that tests you instead of answering — your mistakes, proofs and verdicts live in plain text you own, portable across agents; the judge in the loop is you.

面向教学知识库的可扩展本地 MCP 基底。内置 BeMarkdown，将 Word/PDF 转换为包含正文、公式、表格和示意图引用的中间 Markdown，支持来源追溯、任务日志与 agent 复核，并提供模型自动下载和部署脚本。

Post-response grounding check + cross-family adversarial review for LLM chat agents

A deterministic context-driven agentic SDLC framework for Cursor and Claude. 41 codified laws and 22 specialist roles to eliminate LLM hallucinations and context drift.

Employee — An open-source runtime for autonomous AI workers that can reason, act, learn, and operate toward long-term goals.

Commands for Artificial Intelligence (AI) and LLM in a developer interface.

ORCH: Organizational principles enable collective intelligence in embodied AI

Claude Code plugin with 27 specialist agents, web verification, Red Team review and a final quality gate that blocks weak or unverified answers.

Rust-core agentic framework with durable journal, exactly-once replay, Ed25519 A2A identity, and MCP tool server. Bindings for Go, Python, TypeScript, .NET, and Java via C ABI.

A three-tier training curriculum and ready-to-copy pipeline for building multi-agent AI workflows in Cursor and Claude Code.

Cross-harness SupervisorLLM plugin for Claude Code, Codex, Cursor and Gemini CLI with native agents, hooks, web verification and final release gates.

🤖 AgentChat 9AI — 9个免费网页AI的Claude Code技能套件（Gemini/ChatGPT/Claude/千问/Kimi/MiniMax/MiMo/DeepSeek/豆包），串行降级+并行编排+深度管道

Claude Code skill: one session coordinates, spawns and reviews builder sessions across a multi-phase build

AegisMCP is a CLI-first, air-gapped-by-default multi-agent workspace for private financial documents. PDFs stay on your machine. Agents talk to tools only through an allowlisted gateway. Outputs are typed RiskReport JSON + Markdown — with evidence citations — not free-form chat spam.

High-performance SDK for AI agents with a unified API across AI providers. Optimized for high cache hit rates, observability, and seamless agent orchestration. Built to power production-grade agentic systems and harnesses for teams at any scale.

Go-based AI Agent microservice platform with multi-agent orchestration, RAG retrieval, persistent memory, MCP tool integration, and a self-evolving planner-verifier loop.

A programming language designed for AI agents

FPGA Agentic loop example: AXI4-Stream width converter

A local-first AI desktop environment: an autonomous agent that builds apps, runs jobs and takes real actions on your machine. Local models via Ollama, or any cloud provider. Self-hosted, MIT.

Dynamic routing for agent knowledge. Areas advertise when they're relevant; the agent routes before it reads. MCP server, git-backed, no database.

Private AI on your own servers: an open-source knowledge platform for companies, models included. RAG with a cited source on every answer, permission-aware retrieval, MCP, integrations. Fully local on Ollama or vLLM, proven by an egress check, or any provider you choose. Apache 2.0.

Offline $0 decision layer for coding agents: Choice/Score/Noul primitives, BELKI confidence gatekeeper, ultra-planning, red-teaming, research and RLVR self-improvement -- as an MCP server + CLI + Claude Code skill.

证据约束的代码诊断工作台：确定性静态分析拥有源码事实，LLM 只提假设，resolver/validator 决定能否绑定回真实操作——证据不足保持 unknown/abstained，从架构抑制幻觉。已在真实开源项目发现 bug（stunner #89 已被上游确认修复）。Python。

A daily-driver AI agent runtime built solo: multi-process isolation pods (MCP/LSP/workflow), event-sourced task ledger, proactive scheduling, and multi-channel delivery.

Run tool-calling LLM agents locally on Android devices using Termux and llama.cpp with complete data privacy and no external API reliance.