A small research site that looks at the markdown files developers write to instruct
AI coding agents — CLAUDE.md, AGENTS.md, and SKILL.md — and surfaces the
patterns that show up across hundreds of public GitHub repositories.
👉 analyze-claude-md.onrender.com
Every day the site scrapes public examples of each file type from GitHub, runs topic modeling (NMF) over the text, and renders the results as:
- Topic cards showing the dominant themes and the words that define them
- Term trends tracking how individual words rise and fall across daily runs
- Topic evolution measuring which topics stay stable and which churn over time
A tab nav at the top of the homepage switches between the three file types so you can compare the conventions developers use for each.
These instruction files are a young genre. There aren't established conventions
yet — every project invents its own structure. This site is an attempt to look
across the corpus and see what's actually catching on: what people put in their
CLAUDE.md, how AGENTS.md differs in practice, and how the new SKILL.md
format is shaping up as it spreads.
It's a curiosity project, not a product. Have a look around the live site — that's the whole thing.