Tool to compare your Claude.md and Agents.md against 40k public repos
- Source
- concepts.laxmena.com
- Author
- laxmena
- Date
Context
Developer Lax Meiyappan has measured 39,535 CLAUDE.md and AGENTS.md files from public GitHub repositories and built a page where you can paste your own file and see where it sits among them. These are the instruction files a coding such as Anthropic's Claude Code reads at the start of every session, so everything in them costs on every run. The median CLAUDE.md in the census is about 1,379 tokens, the largest tenth run past 4,469, and the median AGENTS.md is about 1,009. The check runs in your browser, and the census itself kept only counts and shares, not file text or repository names.
The sharpest finding concerns files that still begin with the stock sentence written by Claude Code's /init command, which drafts a starter CLAUDE.md for a repository. According to Meiyappan's write-up, 26% of CLAUDE.md files open with that line. Those files contain an overview section, a description of what the project is, 87% of the time against 28% for the rest, and overviews take up 41% of their tokens against 7%. An overview, like a pasted directory tree, describes things an agent can find by reading the repository, so his advice is to cut what the agent can infer and keep conventions, pitfalls and the reasons behind unusual choices. The match is on text alone, so a file that kept the line may have been rewritten since.
The census also shows how teams handle having two file names for the same job. Among repositories with both files, 58% link one to the other, 12% keep near copies by hand, which Meiyappan calls the ones to watch, and 30% write two separate files. Only about 4% of repositories commit a , a command the agent runs automatically at a fixed moment such as before a tool call, and Meiyappan argues that this is where important rules belong: "Prose tells the agent what to do. Hooks make sure it happens." The page is explicit that it counts what files contain, not whether they help an agent, that about 90% of the repositories are small or personal ones with fewer than 10 stars, and that token counts are estimates from a tokenizer other than Claude's. An October census by Pawel Zarzycki took the other route, reading what the instructions in popular repositories actually say, such as test and commit rules, and how often they give a reason.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- agent hook — A script the harness runs automatically at a fixed point in an agent's loop — before a tool call, after an edit, on session start.
Checking sign-in…
Loading comments…
