An AI That Operates Your Apps, Plus a Tool That Polishes AI-Made Screens
An AI assistant that operates Android apps on your behalf and a tool that polishes AI-generated screens were among this week's standouts. Reference material for image creation and helpers for coding AI also made the list.
This week’s standouts were projects that either hand off phone chores to AI or clean up what AI produces. The use cases range from operating apps and switching off ads to screen design and image creation. We look at 7 new projects first, then 3 development helpers that are drawing attention again.
deepseek-ai/dsh-superagent-prototype: research code for autonomous AI agents
A prototype built to study AI that handles tasks on its own (agents). It has 7,028 GitHub stars (bookmark-style markers). The team explains that the code came out of routine development work and that a coding AI wrote a substantial portion of it. This is research code, not an official DeepSeek product. It suits developers who want to inspect the architecture and adapt it rather than put it straight to work.
Image: GitHub deepseek-ai/dsh-superagent-prototype (MIT)
rddmwmddr/android-ad-cleaner: switching off phone ads
A tool that turns off the system ads that appear on Chinese-made Android phones. It can also switch off trending search terms, news recommendations and marketing notifications. It has 5,623 stars. The developer says users can revert the changed settings without rooting their phones. The desktop app connects to the phone and is operated with buttons, which means you can tidy up ad settings without knowing anything about programming or AI tools. Version v0.3.0, released on November 1, fixed errors reported on several devices and expanded the list of supported models.
Image: GitHub rddmwmddr/android-ad-cleaner README (MIT)
anhao/ai-pose-gallery-zh: reference material for image generation
A Chinese-language creative tool that collects images of poses, outfits and faces. It has 5,124 stars. The project says it holds 6,162 reference entries and 12,022 images. It fits anyone looking for visual references to follow when generating images of people. Adult reference images are hidden by default in safe mode. The open-source license (MIT) applies only to the code, so the image material held in the repository is excluded. Collecting the images to resell or distribute for a fee is prohibited, and other uses require separately checking the original creators’ rights.
heydamir/unslop: polishing AI-made screens
A collection of instructions that teach Claude Code and Codex how to approach screen design work. It has 4,775 stars. The developer explains that it spots excessive glow effects, unnecessary emoji and uneven spacing, then reworks the screen to suit the product. It is a tool for people who want to tidy up the design after building a web interface with AI.
Image: GitHub heydamir/unslop README (CC-BY-4.0)
wangzan101/Mika: an AI that operates apps for you
An AI assistant that handles requests on an Android phone by operating apps and the web directly. It has 3,769 stars. The developer explains that it works on a separate, invisible screen, so you can keep using your phone. It targets users who want to delegate tasks that span several apps by voice. A Korean-language README is also available.
Image: GitHub wangzan101/Mika README (Apache-2.0)
l0ng-ai/openidea: development work in one place
A development tool that brings code writing, error tracking and change history management into a single app. It has 3,354 stars. It aims at developers who are used to IntelliJ IDEA’s shortcuts, tool windows and unified search. It supports more than 60 languages and ships installers for macOS, Windows and Linux.
Image: GitHub l0ng-ai/openidea (Apache-2.0)
paipiange/nvzhuang: outfit references and virtual try-on
A site collecting 1,407 AI-generated clothing images with no people in them. It has 1,861 stars. The material can be browsed in Chinese and English. If you want, there is also a virtual try-on feature that uses AI to put the clothes on. It fits users who are sketching out outfit ideas or want to apply different clothing to a photo.
Image: GitHub paipiange/nvzhuang README (MIT)
Among existing projects, tools that lay out how coding AI should work drew renewed attention.
obra/superpowers: building workflows for coding AI
A tool that strings together multiple instruction sets for coding AI to form a software development process. It has 310,595 stars. It targets developers who use Claude Code or Codex and want to structure the AI’s workflow systematically.
Image: GitHub obra/superpowers (MIT)
mattpocock/skills: pick-and-choose development instructions
A repository gathering small instruction sets needed for development work. It has 309,190 stars. The developer explains that the instructions were made easy to modify and mix together. It suits developers who want to tune coding AI to their own way of working.
Image: GitHub mattpocock/skills README (MIT)
affaan-m/ECC: memory and security management for coding AI
A tool bundling work instructions, memory and security features for the likes of Claude Code and Codex. It has 294,747 stars. It targets developers who want to set up a coding AI’s working environment themselves. The README also includes a Korean-language guide section.
Image: GitHub affaan-m/ECC README (MIT)
New releases this week
- vllm-project/vllm: v0.33.0 released on November 1.
- pytorch/pytorch: a calculation error was fixed in the development build on November 1.
- huggingface/transformers: v5.22.0 released on October 29. The release title explicitly names Mistral Large 4 (Le Chonk).
- openai/codex: 0.166.0 released on October 29.
- ggml-org/llama.cpp: v0.8.0 released on October 28.
- modelcontextprotocol/servers: v2.2.0 released on October 28.
Image: GitHub vllm-project/vllm (Apache-2.0)
This week’s list saw interest in both tools that delegate work to AI and tools that clean up the results. If you are in Korea, it is worth first checking whether your phone is on the supported model list and whether the tool works in Korean as well.