← ClaudeAtlas

readme-writerlisted

Write, generate, overhaul, or fix a project's README.md (or any "readme", repo front page, or top-level project documentation). Use this skill whenever the user wants to create or improve a README, document a repo, or figure out how to present a codebase to readers — covering libraries and packages, CLI tools, research/ML repos, web apps, services, and developer tooling. Trigger even when the user only says things like "document this project", "write docs for my repo", "make a readme", or points at a codebase and asks how to present it, even if the word "README" never appears.
Punpun1643/claude-skills · ★ 0 · Data & Documents · score 62
Install: claude install-skill Punpun1643/claude-skills
# README Writer A skill for writing a README that does its real job: getting the right reader to the right action as fast as possible. A README is the front door to a project — most people decide whether to keep going based on the first screen. Treat it as a designed object, not a form to fill in. ## The rule that matters most: never fabricate The dominant failure mode when an AI writes a README is confident, plausible, wrong content — an `npm install` for a package that isn't published, a badge pointing at CI that doesn't exist, a usage snippet calling functions that aren't exported, a config key that was never read from anywhere. A wrong install command is worse than no install section, because it spends the reader's trust on the very first thing they try. So: every concrete claim in the README must trace back to something real in the project. - **Install / run commands**: derive from the actual package manifest and entry points (package.json scripts, pyproject.toml / setup.py, requirements.txt, Cargo.toml, go.mod, Makefile, Dockerfile). Don't assume a package is published to a registry unless you can confirm it. - **Dependencies / requirements**: read them from the manifest; don't guess versions. - **Usage examples**: use real, exported, importable names. If you show example output, it should be the output the code would actually produce — not an idealized version. - **Badges**: include a badge only if the thing it reports on exists (a real CI workflow, a real publish