What's Inside a GitHub Repository? An Empirical Study on the Contents of 10K Projects
This paper presents an empirical analysis of 10,000 GitHub repositories over a decade, revealing key trends such as the standardization of core artifacts, the dominance of GitHub Actions, shifts in configuration formats, and the emergence of generative AI-related content, while highlighting how GitHub's ecosystem increasingly guides open-source evolution.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the internet as a giant, bustling city where millions of people build houses, factories, and parks. In the world of computer software, this city is called GitHub. It's not a place for bricks and mortar, but for code—the instructions that tell computers what to do. Think of a "repository" (or "repo") as a single house in this city. Inside each house, there are rooms (directories), furniture (files), and specific rules about how the house is built (extensions). For years, experts have wondered: what do these houses actually look like? Do they all have a front door sign? Do they have a kitchen? Do they use the same kind of bricks? Understanding this is like taking a census of the entire city to see how people are really living, rather than just guessing based on a few fancy mansions. This helps teachers know what to teach, builders know what tools to use, and researchers understand how the city is changing over time.
Now, let's peek inside 10,000 of these digital houses to see what's really going on. A team of researchers took a massive snapshot of GitHub, looking at 10,000 popular projects that were still being worked on in 2026. They didn't just look at the top 10,000 most famous houses; they picked a random mix to get a true picture of the neighborhood. They also traveled back in time, comparing these houses to how they looked in 2021 and 2016, to see how the city has evolved over a decade.
Here is what they found: The digital houses have become much more standardized, almost like a cookie-cutter neighborhood where everyone follows the same blueprint. The most common items found in almost every single house are a README.md (a big sign on the front door explaining what the house is), a .gitignore (a list of things you don't want to show to guests, like messy tools), and a LICENSE (a rulebook saying who owns the house and how others can use it). In fact, nearly 95% of the houses have a README and a .gitignore, and about 73% have a license. It seems that GitHub itself, the platform hosting these houses, has become the architect, gently guiding everyone to build their homes in the same organized way.
The city has also seen a massive shift in how people build and maintain their structures. Ten years ago, many houses used Travis CI (a specific type of construction crew) to check their work, but that crew has largely packed up and left, dropping from 46% of houses to just 5%. In their place, GitHub Actions has taken over as the dominant construction crew, now found in 77% of houses. It's as if the entire city switched to a new, more efficient tool for building and testing.
The "furniture" inside the houses has changed, too. The old, heavy, wooden furniture made of XML is being replaced by lighter, more flexible materials like YAML, JSON, and TOML. These new formats are used for configuration, like setting up the lights or the thermostat. At the same time, the types of "bricks" used for the walls are shifting. Houses built with Python and JavaScript bricks are growing rapidly, while those built with older C, C++, Java, and PHP bricks are shrinking. It's a clear sign of which tools are becoming the new standard for the next generation of builders.
Perhaps the most exciting new trend is the arrival of "AI helpers." Just as a house might have a smart thermostat, these digital homes are starting to include special files for Artificial Intelligence and Large Language Models. The researchers found new files like CLAUDE.md, AGENTS.md, and SKILL.md. These are like instruction manuals for robot assistants, teaching them how to help write code or manage the house. There are also new folders named .claude and skills, showing that these AI helpers are becoming a permanent part of the household.
The study also noticed that the houses are getting bigger and more complex. In 2016, the average house had about 111 files; by 2026, that number grew to 211. The variety of file types (extensions) also increased, suggesting that the tools and technologies inside these houses are becoming more diverse. However, the researchers are careful to note that this picture is based on houses that are actively maintained and have at least 100 "stars" (a measure of popularity). So, while this tells us a lot about the popular, well-kept parts of the city, it might not perfectly represent the quiet, less active corners or the closed-off private estates.
In short, this paper suggests that the world of open-source software is not just growing organically; it is being shaped by the platform itself. The "best practices" are becoming the "only practices," with a clear rise in AI integration, a shift to modern configuration tools, and a steady decline in older technologies. It's a city that is constantly renovating, guided by a new set of blueprints that prioritize documentation, automation, and intelligent assistance.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.