Denoise

The focus here is on formalizing and sharing agent instructions as 'skills' rather than just prompts. Repositories based on insights from figures like Andrej Karpathy and Matt Pocock show a move toward expert-driven, structured knowledge for controlling LLMs.

• forrestchang/andrej-karpathy-skills

  A single Markdown file that distills Andrej Karpathy's observations on LLM coding into a structured skill for Claude. Its high rank for a simple text file shows the value placed on expert-curated prompting techniques over generic prompt libraries.

  repeated⭐ 99k
• mattpocock/skills

  A collection of Claude skills from a well-known TypeScript expert. Like the Karpathy-skills repo, this signals a trend where domain experts are codifying their prompting and interaction patterns into shareable, reusable assets for the community.

  rising⭐ 42k
• ComposioHQ/awesome-codex-skills

  A curated list of skills for the Codex agent framework. This is another example of the 'skill' abstraction, providing a repository of pre-built automations and demonstrating the growing ecosystem around agent skill-sharing.

  new⭐ 4.6k

This category is a diverse collection of tools for building and controlling AI agents, reflecting an ecosystem in early-stage fragmentation. Projects range from desktop control infrastructure (trycua/cua) to API middleware (ds2api) and harness builders (Archon).

• NousResearch/hermes-agent

  An AI agent designed to evolve and improve over time by acquiring new skills. While not new, its continued presence indicates sustained interest in self-improving agent architectures that can learn from interaction and user feedback.

  cooling⭐ 124k
• 1jehuang/jcode

  A harness for testing and evaluating coding agents. This project represents the emerging need for standardized tooling to benchmark and ensure the reliability of AI code generation, moving the field towards more rigorous engineering practices.

  new⭐ 1.1k
• CJackHwang/ds2api

  A middleware tool that converts proprietary client protocols from services like Deepseek into a standard API format. It addresses the practical problem of integrating closed systems, reflecting a common need in the fragmented AI service landscape.

  rising⭐ 2.6k
• obra/superpowers

  Presents an agentic skills framework coupled with a software development methodology. This project attempts to bring structure and process to the often chaotic world of AI agent development, targeting repeatability and reliability.

  new⭐ 173k
• coleam00/Archon

  An open-source harness builder for AI coding agents. Similar to jcode, it focuses on making agent behavior deterministic and repeatable, which is a critical step for deploying agents in production environments.

  cooling⭐ 20k
• lukilabs/craft-agents-oss

  An open-source framework for building craft agents. The description is sparse, but it appears to be part of the broader trend of creating structured systems for developing and managing AI agents.

  new⭐ 5.2k
• trycua/cua

  Open-source infrastructure for building Computer-Use Agents that can control entire desktops. This is a significant infrastructure play, providing the sandboxing and SDKs necessary to train and evaluate agents on complex, real-world user interfaces.

  new⭐ 15k
• lsdefine/GenericAgent

  A self-evolving agent that grows a skill tree from a small seed of code. It claims full system control with high token efficiency, representing research into autonomous agents that can expand their own capabilities without constant human intervention.

  repeated⭐ 8.2k

This category is shifting from standalone agent tools to fully integrated, agentic development environments. Warp is a prime example, aiming to replace the terminal, while Hugging Face's ml-intern showcases a specialized agent for a complex domain.

• warpdotdev/warp

  An agentic terminal that aims to be a complete development environment. Its trend indicates a significant push to integrate AI agents directly into core developer workflows, potentially redefining the role of the command line.

  new⭐ 40k
• Alishahryar1/free-claude-code

  A command-line tool and VSCode extension providing free access to a specific Claude model. Its popularity highlights the intense demand for powerful proprietary coding models and the community's effort to create unofficial wrappers for wider access.

  repeated⭐ 18k
• huggingface/ml-intern

  An open-source AI agent from Hugging Face designed to automate ML engineering tasks like reading papers and training models. Its appearance signifies a move towards specialized, domain-expert agents capable of handling complex, multi-step workflows.

  new⭐ 7.4k

This category is dominated by low-signal, all-in-one script collections aimed at beginners, such as GhostTrack and hackingtool. While perennially popular, they represent noise rather than innovation. Maigret stands out as a more focused OSINT tool.

• HunxByts/GhostTrack

  A collection of scripts for location and mobile number tracking. This is a low-signal repository typical of the 'hacking tool' genre, attracting attention but offering little in terms of novel technology or security research.

  new⭐ 11k
• Z4nzu/hackingtool

  An 'all-in-one' collection of hacking scripts. Like GhostTrack, this repo's popularity is driven by interest in simple, pre-packaged tools rather than any underlying technical innovation. It's a perennial but noisy trend.

  repeated⭐ 68k
• soxoj/maigret

  An OSINT tool for collecting information on individuals by username across thousands of sites. Unlike other trending security tools today, this is a focused, legitimate utility for digital investigation and reconnaissance.

  new⭐ 20k

A highly varied category with no central theme. It includes long-standing, essential utilities like uBlock Origin and PowerToys alongside niche, newer projects like a stock analysis tool. Their simultaneous trending suggests a lack of a single, dominant new tool.

• Fincept-Corporation/FinceptTerminal

  A terminal-style application for financial market analysis and research. Its continued popularity suggests a durable interest in open-source tools that provide alternatives to expensive financial data platforms like Bloomberg terminals.

  repeated⭐ 18k
• HKUDS/DeepTutor

  An agent-native personalized learning assistant. This project applies the agentic paradigm to education, aiming to create AI tutors that can adapt to individual student needs. It's an example of agents moving into vertical applications.

  cooling⭐ 23k
• siddharthvaddem/openscreen

  An open-source alternative to screen recording and demo creation tools like Screen Studio. It offers a free, no-watermark solution, which is a common driver for popularity in developer-focused productivity tools.

  cooling⭐ 34k
• ZhuLinsen/daily_stock_analysis

  An LLM-driven stock analysis system for Chinese and US markets. This is a practical application of LLMs for a specific, high-value task, integrating data from multiple sources to provide automated financial insights.

  new⭐ 32k
• EbookFoundation/free-programming-books

  A massive, long-standing collection of links to free programming books. Its reappearance on the trending list is likely due to external promotion or social media attention rather than any new development in the repository itself.

  new⭐ 387k
• iv-org/invidious

  An alternative, privacy-focused front-end for YouTube. Its trending status reflects ongoing user demand for services that offer more control, privacy, and fewer ads than the official platforms they wrap.

  new⭐ 19k
• gorhill/uBlock

  The repository for the popular uBlock Origin ad blocker. As a foundational internet utility, its trending is likely a result of renewed interest in ad-blocking and privacy, not new features. A classic tool being rediscovered.

  new⭐ 64k
• microsoft/PowerToys

  Microsoft's suite of utilities for Windows power users. This is a mature, well-maintained project. Its presence on the trending list indicates its enduring utility and a steady influx of new users discovering its capabilities.

  new⭐ 132k

Projects in this category are tackling the context limitation of AI agents. GitNexus creates a local, client-side knowledge graph from code, while zilliztech/claude-context aims to make entire codebases searchable for agents, highlighting different approaches to the same problem.

• abhigyanpatwari/GitNexus

  A tool that creates a local, in-browser knowledge graph from a GitHub repository or ZIP file. It addresses the challenge of code comprehension by providing an interactive, client-side way to explore codebases without server-side processing.

  new⭐ 33k
• zilliztech/claude-context

  A code search tool designed to give Claude Code context over an entire codebase. This project directly targets the context window limitation of LLMs, using search and retrieval to augment the information available to the coding agent.

  new⭐ 10k
• mksglu/context-mode

  A tool for optimizing the context window for AI coding agents by sandboxing tool output. It claims a significant reduction in context size, which is a critical optimization for reducing API costs and improving agent performance.

  new⭐ 11k

Major players like Microsoft (VibeVoice) and Google (gallery) are pushing open-source, on-device, and multimodal AI. This signals a strategic move to make advanced capabilities like voice synthesis and local generative AI more accessible to developers.

• microsoft/VibeVoice

  An open-source text-to-speech model from Microsoft described as 'frontier' quality. This is a significant release, making high-fidelity voice synthesis technology more accessible and signaling a push by major players to open up multimodal AI.

  new⭐ 45k
• google-ai-edge/gallery

  A showcase of on-device machine learning and generative AI use cases from Google. It provides examples and models for local execution, reflecting the industry-wide effort to move AI processing from the cloud to edge devices.

  cooling⭐ 22k
• AIDC-AI/Pixelle-Video

  An engine for fully automated short video generation using AI. This project packages various AI capabilities (text, image, video, audio) into a single workflow, demonstrating the trend of creating compound AI applications for content creation.

  new⭐ 7.6k
• Anil-matcha/Open-Generative-AI

  A collection of over 200 generative AI models for image and video, presented as an open-source studio. It appears to be an aggregator or a user interface for existing models rather than a new model itself.

  new⭐ 10.0k
• microsoft/markitdown

  A Python tool from Microsoft for converting various file formats, including Office documents, into Markdown. A simple but useful utility that addresses a common developer need for document conversion and standardization.

  cooling⭐ 119k
• google-ai-edge/LiteRT-LM

  A lightweight, real-time language model inference runtime from Google for on-device applications. This is low-level infrastructure work, critical for enabling the efficient execution of LLMs on resource-constrained devices like mobile phones.

  cooling⭐ 4.5k
• hugohe3/ppt-master

  An AI tool that generates editable PowerPoint (PPTX) files from documents. Its key feature is creating native PowerPoint shapes instead of images, making the output genuinely useful for business presentations. It represents a practical application of generative AI.

  cooling⭐ 9.4k

With only one repository, Kronos, this category today points to the development of highly specialized, domain-specific foundation models. This suggests a move beyond general-purpose models towards fine-tuned systems for specific industries like finance.

• shiyu-coder/Kronos

  A foundation model specifically designed for financial markets. Its presence suggests a trend toward creating specialized, domain-specific models trained on proprietary data, moving beyond general-purpose LLMs for high-stakes applications.

  cooling⭐ 22k