Window and control detection
Dragging a box around a nested dialog or button by hand often catches background pixels or the window shadow. Snapdown reads the element under your cursor from Windows itself and offers its exact bounds:
| Mechanism | Source | Behavior |
|---|---|---|
| Window bounds | DWM extended frame bounds | Uses DWMWA_EXTENDED_FRAME_BOUNDS so the invisible drop shadow around a window is left out and only the visible window is captured. |
| Control hierarchy | UI Automation | Reads the accessibility tree under the pointer: the control, its container panel, then the application window. |
| Preview | Overlay outline | Outlines the current target before you click. |
| Widening the scope | Mouse wheel | When more than one target is stacked under the pointer, the wheel moves up or down that chain, and a badge shows Scroll to widen · N / count. |
| Commit | Left click | Locks the outlined rectangle without a click-and-drag gesture. |
If an application exposes no usable UI Automation tree, Snapdown falls back to the top-level window bounds, so capture still works.
Region selection and the pixel loupe
Start with a region, not the full desktop. A region gives a human reviewer or an agent the local context it needs while leaving out unrelated panels, notifications, and surrounding noise. It also makes the exported image smaller before any compression happens.
During selection, the pixel loupe enlarges the pixels around the cursor. Use it to put an edge on the boundary that matters: the clipped bottom of a button, a missing one-pixel separator, the end of an input field, or a baseline that does not align. It is for choosing evidence accurately, not for decorating a capture afterwards.
- Include the control and enough adjacent context to identify its state.
- Leave out unrelated sidebars, browser chrome, and other windows unless they are part of the reproduction condition.
- For a layout issue, capture both items whose relationship is wrong.
- For a responsive issue, put the viewport width or scaling factor in the note.
Numbered markers
A numbered marker is not a generic pin. It is the visual half of a numbered note in the Markdown. Place marker 1 on the image and write its note; in the copied Markdown it is line 1. under Marker Notes, with its position as a percentage of the image. The same number is readable in both places, with no need for a description such as “the thing near the lower right.”
# Settings review 2026-09-24 10:15 ## Finding 1  ### Notes Settings dialog at 125% display scaling ### Marker Notes 1. [loc:48%,91%] The Save button is clipped at the bottom edge. Expected: the whole button stays visible at 125% scaling. 2. [loc:22%,40%] The label baseline sits 6 px below its input field. Expected: label and input share a baseline.
Place a marker beside the target, not on top of the text or icon that must be inspected. The marker should direct attention and leave the evidence beneath it visible. Numbering starts at 1 on every finding, so refer to a marker as “Finding 2, marker 3”.
Ribbon tools
The toolbar has seven tools: six annotation tools and one reframing tool. Click an icon to pick a tool; the active tool is highlighted.
| Tool | Category | What it does |
|---|---|---|
| Marker | Annotation | Places numbered markers that count up automatically, starting at 1 on each finding. Each becomes one numbered note in the Markdown. |
| Shape (focus box) | Annotation | Draws a rectangle around a layout container, an overflow area, or an error. |
| Callout | Annotation | Adds a text bubble with a pointer tail. |
| Blur | Annotation | Blurs what is underneath with a three-pass box blur, burned into the bundle's image so tokens and personal data are unreadable in what you share. The original capture stays in your vault until you delete it. |
| Arrow | Annotation | Points at a small alignment flaw or a missing separator line. |
| Text | Annotation | Writes free text directly on the canvas. |
| Crop | Reframing | Shows a crop rectangle you can still cancel. Committing the crop replaces the source image and deletes the pre-crop file. |
Annotations are burned into the exported image in a fixed red (#DC2626), so they look the same in light and dark readers.
Blur for sensitive material
Before you export or paste a capture anywhere, check it for personal information, credentials, API keys, customer content, email addresses, access tokens, and internal URLs. The blur tool hides those regions while leaving the surrounding layout visible enough to review.
Capture the smallest region
Take only the part of the screen that establishes the issue.
Blur sensitive values
Blur every credential or identity that has to stay in the frame to show the layout or state.
Read the note too
Secrets get pasted into Markdown as easily as they appear in screenshots.
Check what you hand over
Read the copied Markdown and look at the images once before sending the bundle to an agent, an issue tracker, or a colleague.
Image size presets
Vision models count image input by its pixel size, so a raw 4K capture uses far more context than the detail it contains needs. Snapdown downscales and re-encodes each capture when it is saved, using one of five fixed presets or an Auto or Custom mode:
| Preset | Longest edge | Quality | Suggested use |
|---|---|---|---|
| Sharp | 2560 px | 90% | High-DPI design reviews, typography checks, and pixel-level asset checks. |
| Balanced | 1600 px | 75% | Engineering hand-offs and everyday desktop defect reports. |
| Small | 1280 px | 50% | Chat attachments (Slack, Discord) and slow connections. |
| Agent Normal (default) | 1024 px | 80% | Default size and quality for a coding agent to inspect. |
| Agent Compact | 800 px | 80% | Smaller images for vision models when context is tight. |
Auto picks the scale from the captured resolution. Custom lets you set the longest edge and the quality yourself. Resizing and re-encoding run on your computer.
Output: Markdown, Images, and PDF
A bundle leaves Snapdown in one of three ways. Because the numbered markers are burned into the images and written into the notes, an agent and an engineer read the same references.
- Copy Markdown: from the bundle's row menu in the Library. Paste it into an issue, a pull request, or an agent chat. Its image links are absolute paths into your vault, so they include your Windows user name, and only an agent running on the same computer can open them.
- Copy image: copy one finding's image to the clipboard, with or without the annotations burned in.
- Export PDF: a printable document built with Typst, which ships inside Snapdown, so there is nothing else to install.
Two quicker ways out: Full Screen in the capture bar captures the whole screen instead of a region, and Copy Only in the capture panel copies the capture to the clipboard without saving a finding. Settings > Agent Hand-off holds the instruction Snapdown adds to copied Markdown; you can edit it, restore the default, or leave it out.