2026年9月4日 技术热点总结

项目简介：Reference blueprint for building shopping and merchant agents with Claude. Examples in retail, commerce, telecom, and entertainment included.

技术栈：Python

核心介绍：Two commerce agents built on Claude: a shopping agent a business embeds in its app for customers, and a merchant agent its staff use to run the back office. Each is defined once (prompt, skills, tool contracts, gates) and runs on the Messages API, the Claude Agent SDK, and Managed Agents; four runnable verticals show both over the same libraries. > Every company, brand, product, and person here is fictional; …

项目简介：DLSS 5 Swapper is a powerful, easy-to-use tool for installing, managing, and restoring DLSS 5 across games and supported emulators. It features automatic game detection, optional drive scanning, DLSS5-Feeder for compatible titles without native DLSS, emulator support, and compatibility with DirectX 9/10/11/12, Vulkan, and OpenGL.

技术栈：JavaScript

核心介绍：DLSS 5 Swapper Install and manage DLSS 5 Neural Rendering for compatible games and emulators. Windows Installer · Portable · Checksums

关键特性：Easy installation： native DLSS games, or compatible non-DLSS game…

项目简介：cn is a new engine for Tailwind class merging and conflict resolution. It replaces tailwind-merge and clsx. Same APIs. Full parity. And it is 30× faster.

技术栈：TypeScript、clsx、cn、shadcn、tailwind-merge、tailwindcss

核心介绍：cn is a new engine for Tailwind class merging and conflict resolution. It replaces tailwind-merge and clsx. Same APIs. Full parity. And it is 30× faster. import { cn } from “cn” // conditional joining (like clsx) + conflict resolution (like tailwind-merge) cn(“px-2 py-1”, isActive && “bg-blue-500”, { “text-white”: isActive }) cn has zero dependencies and i…

项目简介：Paw Work – selection-first web agent for Chrome: select on the live page, describe the outcome, take away an editable office file. BYOK, sandboxed, no server.

技术栈：JavaScript、ai-agent、browser-agent、byok、chrome-extension、llm、pptx、tldraw、univer

核心介绍：English · 中文 > Just want to use Paw Work? Do not clone this whole repository. You do not need the developer files. Clone only branch unpacked (~44 MB). That folder *is* the Chrome extension (manifest.json at the root).

项目数据：⭐ 843 Stars，🍴 5 Forks

项目简介：Anti-hallucination for AI agents that read binaries. The model proposes, deterministic tools decide: every claim is VERIFIED or REFUTED against the real bytes, with evidence, and grounded facts survive context resets. MCP server + CLI.

技术栈：Python、ai、ai-agents、anti-hallucination、binary-analysis、context-engineering、ctf、developer-tools、disassembler

核心介绍：Reverify The AI proposes. The bytes decide. Anti-hallucination for AI agents that read binaries: every claim is checked against the real bytes. Ask an AI to reverse-engineer a file and it will make things up — offsets,…

项目简介：Sketch Material 3 Expressive screens in the browser and turn them into vibe-coding prompts.

技术栈：TypeScript、design-tool、material-3-expressive、material-design、material3、nextjs、prompt、react、vibe-coding

核心介绍：M3E Canvas Sketch Material 3 Expressive screens in the browser, link them, tap through them, and copy a prompt for your AI coding tool. 日本語 · 中文 · Open the app Left: the sketch in the editor. Right: the app an AI coding tool built from the generated prompt, running on Android. (mp4)

项目数据：⭐ 472 Stars，🍴 29 Forks

项目简介：Microduck 复刻 · 从官方 MJCF 与 Rust 源码反推出的装配图、CAD 装配体与完整电控方案 | Mechanical + electronics reconstruction of Pollen Robotics’ Microduck

技术栈：Python

核心介绍：> 对 Pollen Robotics Microduck 的第三方复刻研究。 > 从官方公开的 MJCF 仿真模型反推出装配图、爆炸图和可直接导入 CAD 的装配体。 Microduck 是一只 25cm 高、737g 的双足机器鸭，15 个 Dynamixel XL330 舵机（14 个受策略控制），用强化学习走路。 它的软件开源（Apache-2.0）。硬件是部分开源： KiCad 9 工程、Gerber、BOM、贴片坐标、STEP 一应俱全。这块板不需要自己画。 > 📌 勘误（2026-09-03）：本文档此前写作「硬件不开源、没有 PCB 原理图」，这是错的。 > 当时只检索了 pollen-robotics/microduck 主仓，未翻查该组织下以 elec_ 开头的硬件仓库。 但官方在 microduck_rl 里发布了完整的 MJCF 仿真模型 + 47 个 STL 网格。 MJCF…

项目简介：Open-source SEO + GEO skills for Claude — keyword research, rank tracking, site audits, backlinks, competitor gaps, AI visibility. Runs on your real Search Console / GA4 / ads data via MCP, with DataForSEO built in. Free, MIT.

技术栈：Shell、ai-seo、backlinks、claude、claude-skills、dataforseo、generative-engine-optimization、geo、keyword-research

核心介绍：Open-source SEO + GEO skills for Claude — keyword research, rank tracking, site audits, backlinks, competitor gaps, AI visibility — running on your own Search Console, Analytics and ads data through the Ryze MCP, with DataForSEO bu…

Conversational Recommender Systems (CRS) typically require domain-specific dialogue data, which is costly, scarce, and often unavailable in new domains. We conduct a systematic empirical study of zero-data CRS bootstrapping: generating synthetic conversational supervision from non-conversational signals—item reviews, metadata, and user-item interactions—without any in-domain dialogue corpus. We compare two information-theoretic selection strategies, Jensen-Shannon diversity and Fisher information, across domain signals, model architectures, datasets, and fine-tuning paradigms. Our resul…

A language model's prediction of its next token develops across layers, and lens methods track this process by decoding intermediate hidden states into tokens. But a lens reading reflects both the hidden state and the readout (the unembedding matrix) used to decode it. Many lenses are fit on a corpus, and we show that two lenses differing only in their fitting corpus can report different tokens for the same hidden states. We call this dependence corpus conditionality. To examine readout structure independently of the fitting corpus, we introduce Sparse Readout Prism (SRP), which decomposes …

Agent performance depends jointly on the model parameters and the executable harness code that manages context and control flow. Optimizing either component in isolation can leave the system bottlenecked by its frozen counterpart: weight updates can change which harness is effective, while harness updates can change which model capabilities are exposed. Existing joint-adaptation methods optimize weights and textual prompts but leave the broader harness fixed. We propose Weight-Harness Alternating LEarning (WHALE), a simple recipe that alternates two phases: updating the model under the curr…

Rubric-based reinforcement learning extends RL beyond tasks with exact answers or rule-based verifiers by scoring responses against instance-specific criteria. However, this makes reward computation expensive: training requires repeated rubric judging, often with proprietary APIs or local generative LLM judges with 7B parameters or more. We study whether smaller language models can serve as efficient and reliable rubric-based judges. To make this question measurable, we construct PointRubric and RaR-Science-Static, two pointwise rubric-based evaluation datasets with instance-specific criter…

Model compression techniques such as pruning and quantization facilitate the efficient deployment and acceleration of Large Language Models (LLMs). However, recent studies show that weight sparsification methods, such as SparseGPT, can amplify existing biases in models, with outputs varying significantly depending on persona cues in the prompt. In this paper, we introduce Debias-SparseGPT, a post-training pruning method incorporating representational debiasing using a second-order term defined over demographically contrasting inputs. We perform empirical validation of our method over a wide…

Portfolio risk assessment ordinarily relies on reliable estimates of cross-asset return covariances, which are difficult to obtain in short, high-dimensional panels. We show that firm-level distribution-valued characteristics can instead provide one-sided certificates of portfolio risk. Under maintained links from characteristics to systematic exposures and from exposures to returns, multi-firm Wasserstein-2 dispersion yields a sharp upper bound on systematic portfolio variance and a corresponding bound for standardized returns. A weighted pairwise relaxation produces an objective that is c…