Ryker Chrome extension turns visual HTML and Markdown edits into agent-ready change requests

A visual editor built for AI-generated documents

A Chrome extension called Ryker brings visual editing of HTML and Markdown into the browser and, in the background, converts every edit into a structured change request that can be handed to an AI agent. According to a post on dev.to, the export is designed to be pasted into a prompt so that the agent, not the human, applies the corrections to the source.

The developers describe a workflow gap that many people working with large language models will recognise. Models can produce complete HTML documents quickly, but reviewing that output still usually happens in a source view. That works well for development, the post argues, but it is clumsy for proofreading, reordering content, or making small editorial fixes.

Edit first, prompt the agent later

Ryker's design separates review from implementation. Rather than patching a document by hand and leaving the source out of sync with what the model produced, the user edits the rendered document directly. Ryker tracks each change and compiles everything into a plain-text blob of structured change requests that a machine can parse. When the review session ends, that blob goes back to the agent, which performs the actual edits.

The developers are explicit that Ryker is not meant to replace the agent or quietly rewrite the source underneath it. The extension is positioned as a review surface, a more natural way to read and mark up a document, while the agent remains responsible for the final implementation.

Live pages and local files

Two workflows are supported:

• Editing any live HTML page directly in the browser.
• Loading HTML and Markdown files from the local machine for inline editing.

In both modes, edits are recorded as they happen and can be exported as a change request at the end. Ryker is distributed as a Chrome extension through the developers' gbti.network site, which leaves Firefox, Safari and other browsers out of scope for now.

Why it matters

HTML has quietly become a standard delivery format for LLM output: generated reports, documentation, prototypes and full pages. The tooling around that output, however, is still largely source-editor-first. Ryker bets that the review step belongs in a WYSIWYG-style surface and, more importantly, that the handoff back to the agent should be structured data rather than a paragraph of prose describing what to fix.

That second point is the notable one. Treating human corrections as data keeps the agent as the single writer of the source file, which avoids the drift that occurs when someone hand-patches generated output and the model no longer knows what changed. For teams iterating on LLM-produced documents, that clean split between reviewing and implementing could remove a recurring source of errors.

Questions remain. The dev.to post does not specify the export format or how expressive the recorded requests are for complex structural edits, and Chrome-only availability limits the audience. But as a pattern, visual review paired with an agent-readable change log, Ryker is a sign of tooling catching up to how generated documents are actually worked on.