You paste a screenshot into a chat and the agent stares back with nothing useful. It cannot read the image well, and the words around it are scattered across three messages. A markdown screenshot fixes that by pairing the picture with text the receiver can actually parse. Here are the terms, defined plainly, with the moment each one matters.
Markdown screenshot
A markdown screenshot is a captured still image bundled with a comment, written in plain-text markdown so both people and machines can read it in order. The image shows the state of the screen; the markdown carries the words, the file path, the expected behavior, and any pin references.
This matters the moment you hand feedback to something that reads text better than pixels. A person skims the picture and the note together. An AI coding agent reads the markdown line by line and knows exactly what you meant. If you want the full breakdown of the format, the guide on markdown screenshots and how agents read them lays it out end to end.
Markdown
Markdown is a lightweight plain-text format that marks up structure with simple characters: # for a heading, - for a list item, **bold** for emphasis. It renders as formatted text but stores as readable characters.
The practical point is that markdown has no hidden layout, no proprietary container, and no binary blob to decode. When you export a review at /r/<slug>/markdown, an agent gets a document it can read straight through. That is why teams building with Cursor or Claude Code paste markdown rather than screenshots alone. The screen capture to markdown path exists for exactly this reason.
Review item
A review item is a single unit of feedback: one screenshot with its comment, or a free-floating comment with no image at all. Each item stands on its own so the receiver can act on one thing without untangling it from the rest.
Items keep a long review scannable. Ten items means ten discrete things to fix, each with its own picture and note, each markable as resolved by the review owner. Compare that to a five-minute video where the receiver has to scrub back to find the one comment that mattered. If you are weighing the two, the guide on screen capture versus screen recording covers when each fits.
Numbered pin
A numbered pin is a marker placed on a screenshot that points at a specific spot, tied to a number the comment can reference. Instead of writing "the button near the top," you drop pin 2 on the button and write "Pin 2: label truncates at 320px."
Pins remove the guessing. On a dense screen with three forms and a nav bar, a pin turns a vague location into a precise coordinate the receiver can trust. This is one of the biggest differences between feedback that gets fixed and feedback that generates a reply asking which button you meant.
Crop
A crop is the one image edit available: drag a rectangle over the captured still to keep only the relevant part, or use the full frame. There is no drawing of arrows, boxes, or text on top of the picture.
Cropping is deliberate. Annotation on the image is the part a machine cannot read, so the meaning goes into the markdown comment and the pins instead. You crop to focus attention, then you write. That split is what keeps the output readable to both a designer and an agent. The common mistakes that make screen feedback hard to act on mostly come from cramming meaning into an image no one can parse later.
Public review link and export formats
When you publish, the review is saved to a short public URL of the form /r/<slug>. Anyone with the link reads it in the browser, no account required. The same review is also a markdown file, a PDF, and a Word document.
One review, several receivers. A client opens the link and reads it like a web page. A stakeholder wants the PDF for the record. An agent takes the markdown and gets to work. You do not rebuild the feedback three times; you capture once and each receiver takes the format that suits them. This is the core of an agent-readable feedback workflow, and it is the same output a person acts on.
Dictation
Dictation is typing by voice, using the browser's built-in speech recognition to turn spoken comments into text on an item. It works in Chrome and Edge. Firefox does not support it, so you type there.
Talking through a screen is faster than typing when you are walking a flow and noticing things in real time. You capture the frame, hit record on the comment, and describe what is wrong while it is fresh. The words land as text, which means they land as markdown, which means the receiver still gets something readable rather than an audio file to transcribe.
Where these terms come together
A finished review is a set of items, each a cropped screenshot with numbered pins and a typed or dictated comment, published to a link and exportable as markdown, PDF, or Word. No install, no extension, no signup. It runs in the browser and it is free.
The fastest way to understand the format is to build one. Start a new review, capture a screen you are working on, drop a pin, write one comment, and publish. Open the /markdown version of your own review and you will see exactly what an agent reads. For a step-by-step run, follow capturing a screen to markdown.