petergyang/human-review

What it solves

It eliminates the friction of providing feedback to AI coding agents by replacing long, descriptive chat messages with a visual editor. Instead of typing instructions like "change the third paragraph," users can edit text, move elements, and leave comments directly on the file, then send all changes to the agent in one batch.

How it works

Human Review acts as a skill for AI agents (such as Claude Code or Codex). When triggered, it opens HTML or Markdown files (or localhost URLs) in a browser-based visual editor. Users make direct edits, rearrange blocks, or anchor comments to specific text and images. Once the user clicks "Send," the agent receives the consolidated feedback and updates the source files accordingly.

Who it’s for

Developers and creators who use AI agents to generate landing pages, Markdown documents, or website plans and want a more intuitive way to refine the output without tedious chat-based prompting.

Highlights

• Visual Editing: Direct text editing, basic formatting, and drag-and-drop rearranging of page blocks.
• Image Management: Resize and move images, or paste images directly from the clipboard.
• Comment Anchoring: Ability to select specific phrases or elements to leave targeted feedback.
• Batch Feedback: Sends all edits and comments to the agent at once to reduce chat noise.
• Local Execution: Runs entirely on the user's machine without requiring accounts, cloud services, or API keys.