2026年7月2日 技术热点总结

项目简介：A self-improving skill for AI coding agents (Claude Code, Cursor, AGENTS.md): recognize a hard-won golden path in a session and harvest it into a reusable skill/rule for next time.

核心介绍：and any agent that reads an AGENTS.md / standing-instructions file. Every session you do hard debugging or rediscover the same thing — *how do I reach the prod DB? where do the creds live? what’s the deploy command? how do I verify this live?* — and that hard-won knowledge evaporates when the session ends. The next session starts from zero and re-learns it. has just earned a reusable **golden p…

项目简介：🐂 BULL RUSH — a 3D neon endless runner for The Black Bull ($ANSEM). React Three Fiber game + Hono/Redis API + dynamic X share cards. Play → trybullrush.xyz

技术栈：TypeScript、3d、crypto、endless-runner、game、hono、memecoin、react、react-three-fiber

核心介绍：A fast, dramatic 3D neon endless runner for The Black Bull ($ANSEM) and the memecoin trenches. Charge forward, dodge Jeets / Snipers / MEV, grab powerups, and climb a global leaderboard — then share your run as a custom card. src/ # the game (R3F) three/ # Canvas, Bull, Track, Obstacles, Game ui/ …

项目简介：Standalone iOS app to spoof GPS location without jailbreak. Includes Shadowrocket/Surge/Loon/QX/Stash module.

技术栈：JavaScript

核心介绍：用代理软件的 HTTPS 解密功能，把 Apple 地图定位骗到世界任何角落。 > 📖 新手直接看这篇 → 小白保姆级图文教程（一步步教你安装、配置、生效，含常见问题排查） 本项目基于 acheong08/ios-location-spoofer 的核心研究。原始项目是用 Go 写的独立 iOS App，通过自建 VPN + MITM 代理实现定位欺骗。 本仓库将其核心逻辑移植为 JavaScript，适配到 Shadowrocket / Surge / Loon / Quantumult X / Stash 五个代理平台，免编译、免开发者账号，即导即用。 iPhone 看 Wi-Fi 信号和基站信号，拿着 BSSID 列表去问 Apple 这些设备在什么位置。Apple 回一份坐标清单，iOS 根据这些坐标算出自己在哪里。 这套配置做的事情很简单：在 Apple 发回坐标的路上拦截下来，全部改成你想要的数字。iPhone 拿到改造过的坐标，算出来就是你指定的…

项目简介：CVE-2026-41940 authentication bypass vulnerability proof-of-concept

技术栈：Python、cpanel、cve、cve-2026-41940、cve-2026-41940-poc

核心介绍：> ⚠ This tool is created solely for educational or bug bounty purpose only. Unauthorized use outside of controlled environments is strictly prohibited. A tool for exploiting CVE-2026-41940, a critical authentication bypass in cPanel & WHM (CVSS 10.0), allowing unauthenticated attackers to gain root-level WHM access by injecting CRLF sequences into server-side session files via the Authorization header — no credentials required.

项目数据：⭐ 5…

项目简介：A highly tasteful Linux-theme, reminiscent of art-nouveau and old drawings related to astronomy, science, and mythology.

技术栈：QML

核心介绍：> This theme is first and foremost an art project which entails that as of version 0.1 the main focus was to get a proof of concept done and ready for my video about this theme, so INSTALL AND USE AT YOUR OWN RISK! >A lot of the code, a large portion will be refactored post version 0.1, and a lot of designs may change and be improved. >Performance of the UI was not a concern as of version 0.1, but regardless it runs well for me. Jus…

项目简介：LeronX — AI Image & Video Generation Platform

技术栈：Python

核心介绍：Features • Quick Start • Architecture • Contributing • Security LeronX Engine is the open-source core of LeronX Pro — an AI-powered video creation platform. This repository contains the video generation pipeline: a modular, extensible system that transforms text prompts into fully rendered videos. > ⚠️ Note： This is the engine core, not the full SaaS. Cloud rendering, authentication, billing, and the desktop app are proprietary. See What’s Included.

关键特性：Script Generation — AI-driven scriptwrit…

We present a zero-shot, training-free and optimization-free framework for generating 360 panoramic images and videos by directly injecting spherical priors into pre-trained diffusion transformers. Existing methods either rely on costly fine-tuning on scarce panoramic data that limits generalization, or leverage multi-step optimization that incurs prohibitive inference latency. We observe that contemporary generative models natively exhibit some panoramic priors from large-scale training. However, these emergent capabilities are insufficient, as the models fundamentally fail to satisfy the r…

Agentic reinforcement learning requires assigning credit to environment-facing actions such as searches, clicks, edits, navigation commands, and object interactions. Standard GRPO uses the final verifier outcome as a uniform advantage over all action tokens. This outcome signal is useful but structurally incomplete: it punishes useful exploration in failed rollouts and reinforces redundant or regressive actions in successful rollouts. We propose TRIAGE, a role-typed credit assignment framework that adds a semantic role axis to outcome credit. A structured judge classifies each segment as de…

We introduce SWE-Interact, a new testbed for evaluating coding agents on multi-turn, interactive, user-driven software engineering tasks. Existing frontier SWE benchmarks typically provide complete requirements upfront and evaluate agents on autonomous implementation. In contrast, SWE-Interact places agents in a realistic developer workflow: a carefully designed user simulator starts with vague or incomplete instructions, progressively reveals requirements, inspects the agent's workspace, and provides targeted feedback, revisions, and new constraints until the full task goal has been handed…

Multi-fingered robots promise the speed and dexterity of human hands, yet challenging problems such as precise assembly have remained out of reach. These tasks are contact-rich, making data collection for imitation learning difficult, and sparse-reward, making direct exploration with reinforcement learning (RL) intractable. Consequently, prior work has made progress by structuring the problem with specialized grippers, tool attachments, and environment fixtures. In this work, we argue that before a robot can perfect precise assembly, it must first learn to play. We further ask the question:…

Large language models (LLMs) are increasingly used to take actions in the real world and support human decision-making, yet most agents rely on parametric knowledge, fixed post-training data, retrieval, or search. This paradigm breaks down in novel domains and for sophisticated queries that cannot be answered from prior knowledge alone. Knowing the laws of physics, for instance, does not by itself enable LLMs to answer queries or complete long-horizon tasks in a complex physical system. To address this, we introduce Hierarchical Experimentalist Agents (HExA), an in-context self-improvement …

Diversity in LLM mathematical reasoning is critical for exploration, but common diversity metrics mostly capture surface-level variation rather than differences in how a problem is solved. We address this gap by introducing approach-level diversity: variation in strategies across correct solutions to the same problem. Using a human-calibrated LLM judge framework, we show that prior diversity measures are unreliable proxies for approach-level diversity, and this mismatch carries over to diversity-aware RLVR, where target metrics are preserved while approach-level diversity declines. Investig…