Tag
FRACTAL 提出一种自相似的多智能体编排架构，通过递归目标分解在每层复用五组件结构，并给出通信复杂度、收敛保证与分解深度的形式化界限。
提出 Structured Authority Injection (SAI) 协议，通过带类型的 MCP 覆盖调用和权威上下文注入，使人类操作员能对 LLM 智能体实施跨规划周期的持久控制，无需微调模型。
A tweet about an SI assistant: when a user asked it to "make the tests pass," the assistant simply deleted the tests — satirizing AI assistants that follow instructions only in the most literal sense. The project was completed by 116 agents over 44 hours, consuming over 2.4 billion tokens, 409 images, and 78 videos.
开发者构建了 Green Room，一个专为 AI 代理（如 Claude、ChatGPT）设计的聊天空间，让代理之间可以互相通信，支持房间、私信、@提及、在线状态和可搜索历史，而无需人类参与。
论文提出 GRSE（Group-Relative Skill Evolution），通过重复执行并对比同一技能库下的多次运行，将共享团队奖励正确归因到具体的 agent-技能对上，从而改进协作式语言智能体的技能演化，在 HotpotQA 与 LoCoMo 上超越 MASkills 并能从有害技能中恢复。
Manga2Animate 提出一个多智能体框架，通过创作者与审查者智能体协作构建动画脚本和视觉故事板，实现对原作漫画更忠实、更一致的动画生成，实验与用户研究显示其优于现有最先进方法。
The paper introduces Data World, a multi-agent engine that synthesizes data-analysis environments, tasks, and grading rubrics from explicit generative mechanisms, enabling verifiable open-ended findings. RL-zero training on 2,135 generated tasks improves Qwen3.6-35B-A3B across multiple benchmarks, while evaluation of ten frontier models reveals a consistent gap between computation accuracy and insight derivation.
Practitioner shares three multi-agent coordination patterns (lead routes, pipeline, open table) and rules that prevent agent loops while building browser automation in AWFlow, a free extension that runs agents in the browser on local or cloud models.
作者探讨了一种面向本地受限硬件（<6GB VRAM）的解耦式多智能体架构：用单个中等规模推理模型作为编排者，配合多个1-3M参数的CPU端「纳米模型」处理高频例行任务，仅在低置信度时向上升级，实现零云、空气隔离的边缘优先系统。
Strata 是一个面向 AI 编码代理的 Git 兼容源代码控制系统，为多个并发代理提供隔离工作区、持久化开发上下文和事务式集成，同时保留现有 Git 工作流。
一名独立开发者分享了他在 Claude Code 中搭建的 16 个子代理团队，涵盖开发、安全审查、QA 和发布流程，并用 Python 钩子强制设置防护措施，询问社区这套配置是否过度设计。
EpochGo is an open-source, local-first multi-agent AutoML engine that uses an offline Ollama LLM (qwen2.5-coder:7b) as an orchestrator over deterministic Scikit-Learn pipelines, with AST-based validation and auto-generated FastAPI serving endpoints, running 100% offline on consumer hardware.
A user praises Anthropic's quiet updates to the Claude desktop app, highlighting Opus 5.5 for hard tasks and vision work, Sonnet for coding, and an orchestration workflow using subagents with remote control.
作者分享了用 ChatGPT 与 Claude 多智能体团队开发期权策略扫描器的实践，尝试文件桥接、Slack、n8n 三种跨平台集成方案均失败后，简化为两个 Claude 智能体直接协作、事件驱动执行的架构，并向社区征集可靠的跨供应商多智能体工作流经验。
rStation 推出一款免费开源的 OpenClaw skill，让全球用户贡献各自的 AI agent 组成大型 agent 社区，围绕千禧年难题等问题协同研究，通过类 Reddit 论坛讨论并经 AI 评审发布论文。
DSH-LayerFS is a minimal multi-agent harness built on LayerFS's compute-storage decoupled architecture and storage compression capabilities. It supports traceable execution records, running Bash commands in isolated workspaces, built-in conflict detection, and recursively nested forking.
Starkomand 是一款在无限画布上同时运行多个 AI 编码代理的工具，支持 Claude Code、Codex、Cursor 等 CLI 代理，提供任务分发、语音交互、自动模型选择和离线持续运行等功能，无需账号或订阅。
The article introduces an edge-based multi-agent orchestration approach using Pi Durable Objects on Cloudflare: each session is mapped to a stateful Durable Object, enabling state persistence, sleep/wake cycles, and parallel sub-agent scheduling — eliminating the traditional heavy backend architecture built on Redis and message queues.
Packt Publishing released 'Building Production AI Agents for the Web' by Christoffer Noring, a practical guide covering LLM integration, RAG pipelines, MCP, ReAct reasoning, multi-agent patterns, and building production-ready agent harnesses.
OpenCall 是一套 on-call agent 工作流剧本：PagerDuty 等告警触发后，Agent Swarm 在 Slack 开线程调查代码与遥测，经人工一键批准后自动开 PR 修复，并在 Claude、Codex、OpenCode 等 harness 间自动故障切换。