Which agents can use these?
Any agent that can read a URL or a pasted file: Claude (projects, or paste the file),
ChatGPT with browsing, GitHub Copilot Chat, Cursor, Windsurf, Codex, Gemini, or your own agent loop.
If the agent cannot browse, download the .md and attach it — the content is identical.
Is it really just a prompt?
Yes. The “prompt” is an instruction to fetch a specific URL and follow it. The skill itself lives
in the markdown file, so the agent gets 120–150 lines of procedure instead of a one-line request.
That is the whole difference in output quality.
What is inside a skill file?
YAML frontmatter (name, description, version, tags), when to use and when not to,
hard constraints to confirm, a numbered pipeline, tables of exact settings, code templates,
a quality checklist, a failure-mode table and an output contract.
Can I edit or fork them?
They are MIT licensed. Download, edit, add your own house rules, host them anywhere.
Keeping the URL stable is all that matters for the one-prompt install.
Do these work for non-coding work?
Six of the ten are non-coding: 3D animation, video editing, landing page copy,
SEO, dashboards and spreadsheets. The other four are engineering.
How do I host my own copy?
Copy the folder to any static host — Cloudflare Pages, Netlify, GitHub Pages, S3.
This site is a single HTML file plus a folder of markdown, with no build step and no dependencies.