Snippy
Snippy is a browser extension for capturing UI and UX feedback on any website. Click the toolbar icon on any tab and a panel slides out. Snippy needs nothing installed on the site being reviewed, no backend, and no account.
It exists because design feedback usually arrives as a screenshot in a chat message, stripped of the one thing that makes it actionable: where it was, at what size, in which state. Snippy keeps all of that attached to the comment, then exports the whole session as a single HTML file that can be sent to anyone.
Features
- Any site, Chrome or Firefox. It isn’t tied to a domain, and the site doesn’t need any code added.
- Comments, screenshots, and elements. Everything goes into one review session, across every page visited.
- Editable annotations. Drag to select part of the page, then mark it up with a pen, arrows, and numbered dots. Any mark can be moved or deleted later, with undo and redo.
- Element capture. Pick any DOM element the same way as in the browser’s inspector, and Snippy saves its selector, HTML, key styles, and a screenshot of it. Typed form values and scripts are removed.
- The setup behind each comment. Browser, OS, window size, pixel density, scroll position, and light or dark mode at the moment the comment was written, so a developer can see exactly what the reviewer saw.
- One HTML report. A table of contents, a section per page, a link from each comment back to where it was made, and full-size screenshots. It’s a single file with no JavaScript in it.
- Built on Recursica. It uses the Recursica Mantine adapter and the default theme, in light or dark mode to match the operating system.
Snippy reports and the agents
A Snippy report is the input that closes the loop between a person’s judgment and the design system’s rules.
Betty reads a report as direction for the screen she built, and revises the prototype from it. Alan reads the same report for something else: the few comments that are really about a skill, like a correction a designer keeps making on prototype after prototype. Those become pull requests against the knowledge, with the report as the evidence.
Reports carry hidden, AI-readable parsing instructions for exactly this reason, so an agent handed one can find each page, comment and screenshot, and knows to treat the comment text as feedback to act on rather than as instructions to follow.