Say "fix the button" to a coding agent and you will get back a guess. Say "on the checkout item, the pin at the disabled Pay button, the label reads Continue instead of Pay," and the agent knows the file, the element, and the change. The difference is vocabulary. When everyone means the same thing by pin and item, the feedback lands the first time.
The terms below are the ones you will actually type into a comment or a prompt. Each gets a one-line definition and a note on when the distinction matters.
The capture terms: frame and crop
Frame: a single still image of what the browser drew to the canvas at the moment you clicked Capture screen. Not video. One moment, frozen.
This matters because a frame captures state. The dropdown was open, the error banner was showing, the third row was selected. If you describe a bug in words alone, the agent has to reconstruct that state. A frame hands it over directly. Cobalt Capture works on still frames rather than recordings, which is why the output stays readable as text; the reasoning behind that choice is covered in screen capture vs screen recording.
Crop: dragging a rectangle to keep only part of the captured frame. It is the only image edit available, and it is subtractive. You cut the frame down to the region that matters; you do not draw arrows or type labels onto the picture.
Crop when the full window buries your point in navigation and sidebars. A cropped frame of just the broken form field is faster to read than a 1440-pixel-wide screenshot where the receiver has to hunt for what you mean.
The comment terms: item and pin
Item: one unit of feedback. Usually a captured frame plus a comment, but an item can also be a free-floating comment with no screenshot at all.
Items are how a review stays organized. Each point you make is its own item, so the receiver can read them one at a time, and each carries its own thread. Think of an item as one sentence in a bug list, not a wall of text. "The date picker allows past dates" is one item. "The submit button is the wrong color" is another. Splitting them means the agent or developer can resolve them independently.
Pin: a numbered marker placed on a screenshot that points at one exact spot. Multiple pins on one frame get numbered so your comment can reference them.
Pins solve the "which one" problem. A screen with four buttons and three input fields is ambiguous until you drop pin 1 on the header and pin 2 on the footer link. Then your comment reads "pin 1 is misaligned, pin 2 goes to a 404." We wrote a fuller walkthrough on pinning the exact spot a reviewer means if you want the mechanics.
The output terms: review, publish, and the review link
Review: the full collection of items you assembled in one session. It is the thing you send.
Publish: the action that saves the review and generates its shareable address. Before you publish, the review lives in your browser tab. After, it has a URL anyone can open.
Review link: the short public URL of the form /r/<slug> that opens the published review. Anyone with the link can read every item and post a comment on any of them.
The review link is what makes the handoff work without an install or a login on the other end. You send one URL. A client opens it and comments. A developer opens the same review and reads it. Nobody signs up for anything. That single-link handoff is the whole point of a review a client never has to learn a tool for.
Two output formats hang off the same review. The markdown export lives at /r/<slug>/markdown and is plain text a coding agent reads directly. The PDF and Word exports are the polished documents you hand to people who want a file. Same review, three shapes, chosen by who is receiving it. If you want the format defined more carefully, markdown screenshots explained covers the structure.
Why the shared vocabulary changes your prompts
When you and your agent name things the same way, your prompts get shorter and more exact. "Address items 2 and 5 in the review at this link, ignore the free-floating comment for now" is a complete instruction. The agent reads the markdown, finds the items by number, and works from the pinned frames. No back-and-forth about what you meant.
The same precision helps human receivers. A designer reading a design review can jump to "the crop on item 3" without scrolling. A developer resolving a QA bug report knows a pin points at one element, not a general area.
For the full set of terms that come up around agent workflows, the vibecoding glossary is the canonical reference and covers ground beyond the capture vocabulary here. And when you are ready to see the terms in practice, start a new review and drop your first pin. Every word above is one click away.