2026年9月24日 技术热点总结

项目简介：Async-first agent harness

技术栈：Go

核心介绍：An async-first agent harness from Unreal Labs. stable across redeliveries. operations. It runs synchronously on the coordinator’s event loop and must not perform I/O or suspend the loop. to submitted operations. Operation execution state is tracked separately; the translator formats these into a model-facing result. for asynchronous execution. Implementations are encouraged to use the available primitives.

项目数据：⭐ 1,706 Stars，🍴 87 Forks

项目简介：An open-source AI platform for knowledge work. Connect your apps, find answers, and get work done.

技术栈：TypeScript

核心介绍：Open Glean is the AI workspace over Hydra DB. Ask a question across your memories, files, and connected apps. Open Glean retrieves the context, writes the answer, and cites its sources. This is the open-source code for Open Glean. The Next.js server proxies requests to the official @hydradb/sdk. Keys are held in an encrypted, httpOnly session cookie and used server-side, never in the browser. You can also run it yourself with your own Hydra DB key.

**关键特…

项目简介：Suitable for Android APK reverse engineering analysis

技术栈：Python

核心介绍：An Agent Skill for Android APK reverse engineering, debloating, ad removal, surgical dex patching, repacking, and runtime/server analysis. It is a skill, not a tutorial: it is written to be loaded by an agent (Claude Code, Codex, or any harness that supports the Agent Skills format) while it works, so it is organized for progressive disclosure — a short decision-oriented SKILL.md, detailed references loaded only when a step needs them, and parameterized scripts you can run

项目数据：⭐ 1,270 Stars，🍴 …

技术栈：Rust

核心介绍：> Apple Wallet Card Skinner & Lockscreen Passcode Themer for iOS 18+ (No Jailbreak Required) > Native Windows client written in Rust. Powered by the airlift AirTraffic sync exploit. > iPhone not detected, AirTraffic sync hangs, or operation fails? > Corrupted or conflicting Apple USB drivers on Windows are the #1 root cause. > 1. Download and install 3uTools. > 2. Disconnect your iPhone from your PC. > 3. In 3uTools, go to Toolbox ➔ Repair Driver.

关键特性：🎨 Custom Card Skins： Assign custom artwork, textures, or bank logos to Apple Pay and App…

项目简介：A verified, community-maintained catalog of 834 open-source projects built with Jev.

技术栈：Astro、astro、awesome、awesome-jev、awesome-list、jev、typesafe

核心介绍：English · 简体中文 · 日本語 · 한국어 · Español · Português (Brasil) 🌐 Explore the website 📜 Changelog 🧠 Jev-like models Awesome Jev is an independent, community-maintained catalog of 834 open-source projects built with Jev, TypeSafe AI’s System One model for typed decisions inside software. It is not affiliated with or endorsed by TypeSafe AI. > What makes this catalog useful?

项目数据：⭐ 765 Stars，🍴 54 Forks

Omni-modal large language models (OLLMs) jointly process vision, audio, and text, yet their modality bias under cross-modal conflict remains underexplored. Existing benchmarks conflate two distinct forms of evidence within a single modality: perceptual signals (e.g., a photograph or recording of a dog) and propositional signals (e.g., the declarative claim "this is a dog"), such that any measured modality bias is inherently confounded with evidence-form bias, precluding clean attribution to either source. To address this, we introduce Tri-PvP, an 8,000-sample tri-modal conflict benchmark cr…

The rapid progress of artificial intelligence is reshaping robotics and accelerating the adoption of learning-based approaches. While purely data-driven methods have achieved remarkable success in computer vision and natural language processing, robotics remains constrained by limited data, complex real-world interactions, and the need for reliable operation. These challenges have motivated the exploration of physics-embedded robot learning, which embeds physics priors into learning algorithms. By encoding the underlying physical laws and constraints, physics priors can complement limited d…

A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness frameworks support multi-agent systems. However, the scalability of current multi-agent harnesses is often constrained by a central orchestrator's capacity to allocate tasks and coordinate workers. To address this limitation, we introduce Agensh, a scalable self-organized multi-agent harness without a central orchestrator: concurrent workers execute a multi-agent cooperation loop, continuously gathering context, claiming and self-assigning sub-tasks, taking action and sharing …

LLM-as-a-judge enables evaluation across diverse tasks, but inference cost and confidence reliability become critical at scale. We study whether a decision-only judge can provide an economical first pass and identify when stronger evaluation is needed. Comparing jev-as-a-judge with sixteen generative and reward-model judges, with blinded human adjudication, we find it within three percentage points of a state-of-the-art LLM judge, our strongest comparator, on ordinary preference and evidence-grounded factuality at 0.36% of the comparator's fee. Larger gaps arise when judgments require check…

Can one language model hand its live memory to another without the receiver rereading the context? We demonstrate useful persistent hybrid-state transfer across one architecture-matched Qwen3.5 4B-to-9B sibling pair. To our knowledge, this is the first demonstrated cross-model handoff of persistent recurrent inference state between differently sized hybrid language models without target prefix replay. Translated attention KV alone leaves a large gap; adding the Gated DeltaNet (GDN) persistent-state package lowers teacher-forced negative log-likelihood (NLL), the average next-token log-loss,…

Modern embodied agents achieve impressive success rates, yet their actual instruction-following ability is far weaker than these numbers suggest. We trace this illusion to a structural property we term low scene entropy: when a visual scene admits only one valid task, language becomes redundant and a policy can score highly while barely using it. We introduce RoboFollow, a diagnostic benchmark with three principles: (1) High Scene Entropy: each training scene supports multiple kinematically distinct task branches, making vision alone insufficient and forcing reliance on language. (2) Hierar…