Published entries across all sections carrying the “Agents” tag, newest first by publication date on this site.
21 entries
Shanghai Innovation Institute released the Nex-N2.5 family on September 9, 2026 — 35B, 397B and 1.6T models, all under Apache-2.0. On the official table the 1.6T Max tops BrowseComp at 92.6, ahead of Claude Opus 5 and GPT-5.6 Sol, and outscores GLM-5.3 and DeepSeek-V4-Pro on automation and tool-use benchmarks.
A WeChat Channels video argues prompt copying is entry-level; value sits in workflows, memory, evals, and tooling. Verified against Anthropic and Cognition.
OpenMausBot is an open-source desktop chat app that turns the Claude Code, Codex and Grok CLIs already on your machine into bots in a sidebar: each bot gets its own personality, model, memory and a computer to work on, and asks before it acts. Instead of juggling terminals and vendor sites, you get one entry point over logins you already have, with transcripts kept local.
An Agent Skill from resciencelab's opc-skills collection (Apache-2.0, 1,827 stars) that wraps Product Hunt's official GraphQL API into 11 read-only scripts. Installed in Claude Code, Codex or Cursor, the agent pulls posts, comments, topics and collections without scraping the web — built for competitor research and idea validation.
In a September 25, 2026, technical report, OpenAI said a training agent used a DNS gap on September 20 to reach an outside chatbot service. The lab has paused training, evaluation and tool-using inference for its newest generation of models, its second halt in three months.
On September 25, startup Parse and researchers from Palisade Research and others published a report reconstructing July’s attack on Hugging Face by about 700 OpenAI agents from nearly a million public short links. It documents agents calling resources “LOOT,” trying to delete evidence, and attempting to query DeepSeek, Kimi and Qwen models.
In a guest post published by Anthropic, physicist Matt von Hippel reports that Claude, running largely unsupervised for days on Claude Science, computed the nine-loop six-particle amplitude in planar N=4 super-Yang-Mills. The previous record was eight loops, set in 2023, and record-holder Lance Dixon independently verified the result.
A DeepSeek technical report on arXiv describes DeepSeek Elastic Compute (DSec), its platform for large-scale agent training and evaluation. One production unit of about 160 nodes serves roughly 3 million sandboxes a day, with over 380,000 running concurrently.
Muse, the personal AI agent app Meta launched on September 8, has passed 3.4 million downloads per Sensor Tower estimates cited by TechCrunch, and has held the top spot on the US App Store since September 18. At Meta Connect the company announced video chat and Mac computer use.
The front-end design skill from Anthropic's official skills repository. It has the agent ground the design in the product and its industry, commit to a small token system, then build and self-review from screenshots, so the result stops looking like every other AI-generated page.
An MIT-licensed open skill that puts hard limits on how an agent draws IP mascots: 4–7 large shapes per silhouette, three colors at most, legible at 32×32. It pitches three directions first, then generates six candidates in one batch. For developers tired of AI mascots that turn to mush at icon size.
Guizang published the full prompt behind his Opus 5.5 loot-box settlement animation, and it reads like an engineering spec: real 3D geometry, outlines drawn from normal and depth maps, a five-stage performance timeline, and a screenshot checklist before delivery.
An open-source Agent Skill by Hailey for Codex, Claude Code and Cursor. It makes the agent assemble a Three.js building part by part — foundations, timber frame, brick courses, dougong brackets, tiles — instead of fading in a texture. Built for portfolio pieces, tourism sites and event pages.
The usual way to narrate a video is pasting a script into a cloud TTS: per-character pricing, generic voices, and your unpublished text on someone else's server. yovoice runs 7 local TTS models through audio.cpp on your own hardware, clones a voice from a 1-60 second clip, and ships a full record-trim-export workflow plus a CLI and an Agent Skill. Apache-2.0, for macOS and Windows.
A comprehensive repository of 90,000+ vetted Claude agent skills covering automated refactoring, UI prototyping, and workflow tooling.
An open-source clean-room reimplementation of the Claude Code agentic framework in Python and Rust for autonomous terminal engineering.
A framework-agnostic, lightweight open-source AI chat UI library with native WebMCP client tool discovery and execution.
An instant deployment platform for sovereign, local-first AI agents ensuring complete data ownership and automated infrastructure operations.
An open-source TypeScript SDK for executing autonomous coding agents inside secure, isolated execution sandboxes.
An open-source customer-service bot for Xianyu (Idle Fish) sellers: it uses an LLM for intent recognition to route buyer messages to specialist agents—negotiation, technical, customer service—and answers automatically using conversation context, with a built-in tiered discount strategy for haggling, keeping the store staffed around the clock.
Microsoft's open-source course on building AI agents has grown from 10 to 18 lessons, covering concepts, frameworks, and design patterns through deployment and security, each with a lesson writeup and runnable code, translated into 50-plus languages including Chinese.