Tool for Humans · browser-only · no upload
Feedback Markup: a free screenshot annotation tool for AI feedback.
Feedback Markup is a free screenshot annotation tool for giving an AI agent precise feedback. Paste a screenshot or an image your agent produced. Put numbered marks on it, say what you want at each one, and export an annotated image with instructions the agent can follow precisely: where each mark is, and what to do there. Everything happens in your browser. Nothing is uploaded, there is no account, and no language model is involved.
Numbered marks with a comment and a type
- Paste a screenshot with Ctrl+V (⌘V on a Mac), or drop or open PNG, JPEG, WebP and GIF images
- Several images become separate pages of one bundle
- Six marks: pin, box, circle, arrow, freehand line and a text note written on the image
- Every mark gets the next number, a comment (the instruction to the agent) and a type
- Select, move, resize and delete marks; reorder the list to renumber them
- Undo and redo, zoom and pan
- Mouse, touch and pen; every mark is also reachable and movable from the keyboard
An annotated image plus instructions
page-N-annotated.png: each page at its own resolution with the numbered marks drawn on itfeedback.md: prompt-ready instructions, one line per markfeedback.json: every mark with pixel and normalised coordinates- One ZIP with all of the above and the original images under
original/ - Copy the instructions or the annotated image straight to the clipboard
Mark types
Each type has its own badge shape, line pattern and written label, so it never depends on colour alone.
| Type | What it asks the agent to do | How it is drawn |
|---|---|---|
fix | Change this as described. | Solid line, round badge |
question | Answer this before changing anything. | Dashed line, diamond badge |
remove | Delete this. | Dotted line, square badge |
keep | Leave this exactly as it is. | Dash-dot line, hexagon badge |
Export format, for agents
feedback.md: prompt-ready instructions
One line per mark in number order, under a short header that explains the conventions. Percentages are measured from the top-left corner of the page: “across” is left to right, “down” is top to bottom. Each number matches the badge with the same number on the annotated image.
# Feedback on homepage.png
2 numbered marks on 1 page. Each number matches the badge
with the same number drawn on the annotated image
(page-1-annotated.png).
## Marks
1. [fix] Upper left, box at 12%-34% across, 8%-15% down (page 1): The title is misspelled.
2. [keep] Lower right, point at 81% across, 77% down (page 1): This chart is right, leave it.
page-N-annotated.png
Each page at its own resolution with the numbered marks drawn on it. Marks are sized to the image, so the numbers stay readable when a large screenshot is scaled down. Images above 16 megapixels are scaled to fit that limit.
The ZIP
feedback.md, feedback.json, one page-N-annotated.png per page, and the untouched originals under original/.
feedback.json: schema version 1
{
"version": 1,
"source": {
"name": "homepage.png",
"kind": "image",
"pages": [{
"index": 1,
"width": 1440, "height": 900,
"name": "homepage.png",
"annotated": "page-1-annotated.png"
}]
},
"generalComment": "Overall the layout works.",
"marks": [{
"n": 1,
"page": 1,
"kind": "rect",
"type": "fix",
"comment": "The title is misspelled.",
"anchor": "upper left",
"box": { "x": 173, "y": 72, "width": 317, "height": 63 },
"boxNormalized": { "x": 0.12, "y": 0.08, "width": 0.22, "height": 0.07 }
}]
}
| Field | Meaning |
|---|---|
version | Schema version, currently 1. Fields may be added; none is renamed or removed without a new version. |
source.name, source.kind | What the feedback is about. kind is "image" from this tool and "pdf" from the PDF Tool. |
source.pages[] | index (page number, from 1), width and height in pixels, and optionally name (the file the page came from), annotated (file name of its annotated PNG) and unit ("pt" when a PDF page is measured in points instead of pixels). |
generalComment | A comment about the whole thing. May be empty. |
marks[].n | The number on the mark's badge. Marks are listed in this order. |
marks[].page | The index of the page the mark is on. |
marks[].kind | pin, rect, arrow, stroke (freehand) or text (a note written on the image). |
marks[].type | fix, question, remove or keep. |
marks[].comment | The instruction to the agent. For a text mark it is also the text written on the image. |
marks[].anchor | The position in words: a cell of a 3 × 3 grid (“upper left”, “centre”, “lower right”…), a band or side when the mark spans most of the page (“upper band, full width”), or “most of the page”. An arrow is described by its head. |
marks[].box | x, y, width, height in whole pixels of the page. The origin is the top-left corner; x runs right, y runs down. A pin is a zero-size box at its point. |
marks[].boxNormalized | The same box as 0–1 fractions of the page's width and height, so it still holds if the image is resized. |
marks[].points, marks[].pointsNormalized | Only for arrow (two points, tail then head: it points at the last one) and stroke (the freehand line, simplified to at most 48 points). |
Questions people ask
How do I give an AI agent precise feedback on a screenshot?
Open Feedback Markup at https://agenticweb.tools/feedback, paste the screenshot with Ctrl+V, and put a numbered mark on each thing you want changed: a pin, a box, an arrow, a freehand line or a text note. Give each mark a comment and a type (fix, question, remove or keep). Then copy the instructions and the annotated image into the agent's chat, or download one ZIP. The agent gets a numbered list that says where each mark is and what to do there.
Is my screenshot uploaded anywhere?
No. Feedback Markup makes no network requests with your image. It is opened, marked up and exported inside your browser tab, and closing the tab discards it. There is no account and no language model involved.
What does the exported feedback.json contain?
A version number, the source (name, kind and a list of pages with their pixel size), a general comment, and one entry per mark: its number n, page, kind (pin, rect, ellipse, arrow, stroke or text), type (fix, question, remove or keep), comment, an anchor such as “upper left”, its box in pixels, the same box as 0–1 fractions of the page (boxNormalized), and for arrows and freehand lines the points. Coordinates start at the top-left corner; x runs right and y runs down.
What do the coordinates in feedback.md mean?
They are percentages of the page measured from its top-left corner: “across” runs left to right and “down” runs top to bottom. A line such as “1. [fix] Upper left, box at 12%-34% across, 8%-15% down (page 1): The title is misspelled.” means mark 1 is a box spanning 12% to 34% of the width and 8% to 15% of the height on page 1.
Can I mark up a PDF?
Feedback Markup is for screenshots and images. PDFs are commented in the PDF Tool, which exports its comments in the same feedback format.
Does it work without a mouse, or on a phone?
Yes. Marks are drawn with pointer events, so a finger or a stylus works, and the page still scrolls normally when the Select tool is chosen. From the keyboard, every mark is in the list; its number button selects it on the image, the arrow keys move it, Delete removes it and Enter edits its comment.
Is Feedback Markup free?
Yes. Feedback Markup is free, with no account and no sign-up. It runs in your browser, so there is nothing to upload.
Related tools: PDF Tool, a free PDF editor that runs in your browser; Markdown to Word / PDF, a converter that runs in your browser; HTML Presentation Studio, a visual editor for HTML slide decks. All AgenticWeb.Tools.