Human Review

Human Review (github.com/petergyang/human-review) fixes a problem every agent user knows but few have named: chat is a terrible place to review visual work. Spotting a clumsy sentence or an oversized image is instant; describing the fix in prose, then waiting a full agent round-trip to see whether it was understood, is not. Peter Yang — who previously built the /no-ai-slop skill — built this so you can just make the change yourself.

Run /human-review (your file) on an HTML or Markdown file, or point it at a localhost URL, and the page opens in a browser-based editor: preview on the left, feedback panel on the right. You edit text directly, apply bold/italic, build lists, resize images by dragging a corner, delete elements outright, and — the best part — select a phrase or click an element to leave a comment anchored to it, exactly like Google Docs. Command-clicking links lets you walk a multi-page site in one pass without losing feedback. Then a single Send to agent delivers every edit and comment as one batch; the agent applies them to the source and refreshes for the next round.

Installation is a paste-the-link instruction to your agent (or npx -y human-review setup --global), and everything runs locally — no account, cloud service, database, or API key. One rule worth internalising: HTML edits save automatically, but Markdown and localhost pages only reach the source when you click Send. As a way to stop small fixes costing a conversation turn — and to keep an agent from quietly slop-ifying a project you never really read — it earns its place in the toolkit. See our companion guide, Edit Your AI's Work Like a Google Doc, for the full walkthrough.