[GitHub Trend Summary] AI Agent 'Research Capability' and 'Token Saving' Tools Surge - 2nd Week of June '26

GitHub trends (rankings of OSS, or open-source software available to everyone) are an indicator that reveals in real-time the tools that developers around the world have judged to be useful.
This week, 'skills and tools that enhance AI agents' were concentrated at the top, making it a week where the next stage of AI utilization became clearly visible.

mvanhorn/last30days-skill — A skill that gives AI agents real-time research capabilities

This is a skill for AI agents that researches any topic across Reddit, X, YouTube, Hacker News, Polymarket, and the web to generate summaries backed by evidence.
It supports over 50 agent harnesses (execution environments that run AI agents) such as Claude Code, Cursor, Codex, and Goose.

The reason it is attracting attention now is the ease of use: Reddit, Hacker News, GitHub, and Polymarket can be used by AI as-is with zero configuration, while X, YouTube, and TikTok can be unlocked with just 30 seconds of setup.
From Claude Code, it can be introduced immediately with two commands: '/plugin marketplace add mvanhorn/last30days-skill' followed by '/plugin install last30days'.
It is the #1 GitHub trend this week, highly rated for allowing AI agents to directly extract real-world public opinion, market predictions, and video information.

A word for beginners: A plugin that allows your AI assistant to research social media and news on your behalf.

chopratejas/headroom — A token-saving library that compresses inputs to LLMs by 60–95%

A Python library that compresses AI agent tool outputs, logs, RAG chunks (fragments of text retrieved via search), and files before passing them to an LLM.
It reportedly achieves a 60–95% token reduction while maintaining the same level of response quality.

It can be used in three formats: as a compression library, a transparent proxy, or an MCP server (an intermediary server for when AI agents call tools), and its ease of integration into existing configurations is why it is supported.
It also supports integration with LangChain, LiteLLM, and MCP, and can be installed via 'pip install headroom-ai'.
It can be used as-is in Japan, making it a particularly compelling configuration for developers thinking about reducing API costs.

A word for beginners: A tool that tightly compresses inputs to AI to significantly reduce API fees.

Leonxlnx/taste-skill — A design skill to prevent AI from generating generic UI

A framework that provides design heuristics (practical criteria for judgment) as a SKILL.md file to prevent the monotonous and mediocre frontend UI that AI tends to create (mass-produced designs known as 'AI slop').
It offers 12 types of skill variants, including taste-skill, minimalist-skill, and brutalist-skill.

It works with SKILL.md-compatible agents like Cursor, Claude Code, Codex, and Gemini CLI, and can be installed instantly with 'npx taste-skill'.
A feature of this week is that it is gaining popularity among designers and side-hustle developers who want to break away from the 'AI-like mass-produced feel' when creating landing pages and portfolios.

A word for beginners: A tool that prevents AI from creating copy-paste-like designs and helps generate unique UI.

microsoft/markitdown — A Microsoft tool that converts any file into Markdown

A Python tool that converts over 15 file formats, including PDF, Word, Excel, PowerPoint, and HTML, into Markdown, and can be installed via 'pip install markitdown'.
It is designed specifically for RAG (Retrieval-Augmented Generation: a technology that incorporates searched documents into AI responses) pipelines and document input preprocessing for LLMs.

The cumulative number of stars has reached approximately 139,000, and it has trended again this week.
This is a sign that the demand for solving the challenge of 'how to get AI to read documents' in AI development has not waned.
It supports the conversion of Unicode documents, including Japanese, and is used for a wide range of purposes both domestically and internationally.

A word for beginners: A tool that instantly converts PDFs and document files into a format that AI can read.

Panniantong/Agent-Reach — Adds access to over 16 platforms to AI agents without needing API keys

A Python-based CLI tool that adds access to over 16 web and social media platforms, including Twitter/X, Reddit, YouTube, GitHub, Bilibili, and XiaoHongShu, to AI agents.
It operates without using API keys by combining cookie authentication, public scraping (a method of automatically acquiring data from web pages), and free MCP services.

It supports major agents such as Claude Code, Cursor, Windsurf, and Codex, and the reason for its popularity is its simplicity, with setup completed in two commands: 'pip install agent-reach' followed by 'agent-reach install'.

Caution: Since these mechanisms use cookie authentication or public scraping, please be aware of potential conflicts with the terms of service of some platforms, such as Twitter/X.
We recommend checking the terms of each platform before use.

A word for beginners: A setup tool that allows AI assistants to read social media for free.

Was there any tool among these that you felt like trying out?

This week's GitHub trend list

• lfnovo/open-notebook
  TypeScript ⭐4,796
  An open-source implementation of Google NotebookLM. A NotebookLM clone with expanded flexibility and features

• santifer/career-ops
  JavaScript ⭐4,111
  An AI job-hunting system based on Claude Code, featuring 14 skill modes, a Go dashboard, and PDF generation support

• apple/container
  Swift ⭐4,081
  An official tool written in Swift for running Linux containers on Apple Silicon Macs, announced at WWDC26

• phuryn/pm-skills
  (No language) ⭐4,005
  A collection of over 100 agent skills, commands, and plugins for product managers

• refactoringhq/tolaria
  TypeScript ⭐3,514
  A desktop application for managing Markdown knowledge bases

• CopilotKit/CopilotKit
  TypeScript ⭐2,751
  A frontend framework for building AI agents and generative UI for React, Angular, and mobile

• aaif-goose/goose
  Rust ⭐2,509
  An open-source AI agent written in Rust that can perform installation, execution, editing, and testing using any LLM

• Open-LLM-VTuber/Open-LLM-VTuber
  Python ⭐1,676
  A locally running VTuber x LLM tool that combines hands-free voice interaction with Live2D characters

• openai/plugins
  JavaScript ⭐1,424
  Official OpenAI plugin repository

• opencv/opencv
  C++ ⭐1,176
  A standard OSS library in the field of image and video analysis (computer vision)

• NVIDIA/cosmos
  Jupyter Notebook ⭐1,099
  A world model and toolset for Physical AI aimed at robotics, autonomous driving, and smart infrastructure

• aquasecurity/trivy
  　Go ⭐792
  　A security tool that detects vulnerabilities, misconfigurations, and secrets in containers, Kubernetes, and code repositories.

• sveltejs/svelte
  　JavaScript ⭐523
  　The official repository for 'Svelte', a lightweight and fast JavaScript front-end framework.

Summary

Looking at this week's GitHub trends, the direction of 'giving something to AI agents' was clearly established.
last30days-skill is designed to supplement agent weaknesses, with 'research capability' for last30days-skill, 'cost efficiency' for headroom, 'design sense' for taste-skill, and 'internet access' for Agent-Reach.
The fact that markitdown has trended again shows that the challenge of 'how to pre-process data' in AI development is still very much active.
It was a week where the infrastructure development phase for AI agents progressed steadily.

If you found this article helpful, a 'Like' or 'Follow' would be greatly appreciated.
Also, please let me know in the comments if there are any topics you would like to know more about.
※ Feel free to just copy and paste the topic 'heading' into the comments.

Past news can be found here

#GitHub #Programming #Engineer #OSS #AI #Python #AITools #last30days-skill #headroom #taste-skill