Review and refine AI-generated Markdown with tracked changes
Markleft is a tool that brings familiar 'suggestion mode' behavior—similar to Google Docs Suggesting mode or tracked changes—to Markdown files edited with AI assistance. It addresses the broken workflow of providing feedback on AI-generated content, where users typically write prose feedback in a follow-up prompt, receive a completely rewritten document, and then have to spot the differences and re-map their original intent to the changes.
Provides an editor that lets humans create and edit comments, propose changes, and apply suggestions without needing to think about the underlying Markdown format.
Uses a Markdown-compatible specification that AI can parse, allowing comments on text, code, tables, Mermaid diagrams, images, and SVGs.
Instead of prompting a full rewrite, it instructs the AI to append suggestions directly into the Markleft file, enabling review and approval of changes in context.
Runs simply via a bookmarklet in the Chrome browser, operating directly on local Markdown files without complex installation.
Keeps Markdown as the document format throughout the entire editing and review cycle, ensuring portability and compatibility.
It adds a 'suggestion mode' to Markdown files, similar to Google Docs Suggesting mode, where users can leave comments directly on the document and an AI can propose changes for review instead of rewriting the entire file.
Markleft uses a Markdown-compatible annotation specification that the AI can consume. A prompt is provided to the AI that tells it how to address these comments and to append suggestions only in the Markleft format.
You can annotate text, code, tables, Mermaid diagrams, images, and SVGs within the Markdown document.
It runs as a bookmarklet in the Chrome browser and operates on local Markdown files, so you just need to load your file in the editor via the bookmarklet.
The blog post demonstrates the workflow with Claude, but it provides a prompt for instructing the AI on how to handle the comments, making it applicable to other capable AI models that can follow Markdown-based instructions.