What this skill does
A README that's wrong is worse than no README, and the most common way READMEs go wrong is that someone wrote the setup steps from memory instead of from what the project actually requires. This skill reads the actual project — its dependency manifests, config files, entry points, and existing docs — and generates a README grounded in what's really there, not assumed conventions.
How it works
- Project scan — reads the dependency manifest (
package.json,composer.json,requirements.txt, etc.), configuration and environment examples, and entry-point files to determine the real runtime requirements and how the project is actually started. - Structure summary — writes an architecture overview based on the actual folder structure and key files, not a generic template.
- README assembly — assembles a complete README: description, requirements, setup/install steps (verified against the actual manifest, not guessed), usage examples, and a project structure section — flagging anything it couldn't verify (e.g. deployment steps not evidenced in the repo) rather than inventing it.
Requirements
- Works on Claude AI (Claude Code, with filesystem access to the project) and on OpenClaw (reading from a connected repo).
- No external API required.
Setup
- Unzip the package into your Skills directory, keeping the folder structure intact.
- Point it at your project root to generate a README.
- Review the generated README for the flagged "couldn't verify" notes before publishing — it's deliberately conservative about anything it can't confirm from the code itself.
What you get
- A README grounded in your project's actual manifest, config, and entry points — not assumed conventions
- Setup and install steps verified against what the project really requires, not guessed
- An architecture overview based on your real folder structure
- Anything unverifiable clearly flagged instead of invented, so you know exactly what to double-check
- A worked example against a sample Node.js project