# Planpage Capture

> The browser extension that captures pages, picks the elements that are wrong, records console errors, and walks a checklist with you, sending it all to planpage for people and agents.

Planpage Capture is a browser extension for Chrome and Edge. It turns "the button on the checkout page is broken" into something an agent can act on: a screenshot with the button boxed, the button's selector and accessible name, and the error the page logged when you clicked it.

It can:

- capture the visible part of a page, a region you drag over, or the whole page (it scrolls and stitches);
- let you click the elements that are wrong, and number each one with its technical details;
- record console errors and failed requests while you reproduce a problem;
- send a capture to a document as a comment or a figure, marked up first in the [annotation editor](https://planpa.ge/docs/screenshots-and-annotations.md);
- walk a document's checklist beside the page, item by item, and send the results as a review run.

## Install it

Planpage Capture is in early access and isn't listed in the Chrome Web Store yet. When it is, this page will link to it.

If you run planpage yourself, or want to try it now, build it from the planpage repository: `pnpm --filter @planpage/extension build`, then open `chrome://extensions`, turn on **Developer mode**, choose **Load unpacked** and pick `apps/extension/dist`.

## Sign in

Click the Planpage Capture button in the toolbar and choose **Sign in**. A planpage window asks you to connect it; choose **Connect**. You can capture before signing in, and sign in when you send.

The extension acts as you: what it sends shows under your name, not an agent's, and it can open the documents you can. It shows in **Your account → Agent connections** as "Planpage Capture", where you can disconnect it at any time.

It connects to planpa.ge. If you run your own planpage, set your server's address in the extension's **Settings** first; it asks for permission to reach that address.

## Capture a page

Open the page and click the toolbar button (**Alt+Shift+P**), then choose:

- **Visible area**: what's on screen now. **Alt+Shift+S** does this without opening the menu.
- **Region**: drag over a part of the page. Esc cancels.
- **Full page**: scrolls to the bottom and stitches the screenshots together. Headers that stick to the top are hidden after the first screen so they don't repeat.
- **Pick elements**: hover to highlight, click each element that's wrong (click again to unpick), then **Capture** or Enter.

The capture opens in a new tab with the annotation editor. Add boxes, arrows and notes, redact anything private, then choose a document and send it:

- **As a comment**, on a block of the document or on the whole document, for someone to act on.
- **As a figure** in the document, after the block you choose.

The screenshot is private to the document's workspace.

### What an agent gets from a picked element

Each element you pick becomes a numbered box with its note set to what it is (for example `button "Pay now"`) and these details, which agents see with the marks:

- a CSS selector that matches only that element, preferring `data-testid`, `id` and ARIA attributes over long paths, and an XPath;
- its tag, role and accessible name, and its visible text;
- its position and size, a short piece of its HTML, and a few computed styles (colour, font size, display, visibility…);
- on React and Vue pages built for development, the component's name and, where the build includes it, the source file and line.

That's usually enough for a coding agent to find the code.

## Record console errors and failed requests

Turn on **Record console errors and failed requests** in the menu, then reproduce the problem on the page. Recording starts on that tab when you turn it on, and later captures include:

- console errors and warnings, and uncaught errors;
- the method, address and status of requests that failed. Query values are removed from addresses.

It never records request or response bodies, headers or cookies. Before you send, you can untick **Include what the page logged** to leave it out of that capture.

## Review in browser

Steps and checklist items that contain a web link have a **Review in browser** button. With Planpage Capture installed, it opens the page and remembers which document and item you're reviewing: the menu shows it, and captures you send from that tab start on that document and block. Without the extension, the button explains how to get it, or opens the page anyway.

## Review runs: test a checklist

A review run is a pass over a document's checklist, item by item, on the page you're testing. It's how you hand an agent a precise list of what's broken.

1. Open the page, click the toolbar button and choose **Review against a checklist** (or **Open the review checklist** after **Review in browser**). The review panel opens beside the page.
2. Pick the document, unless you came from **Review in browser**.
3. Mark each item **Pass**, **Fail** or **Blocked**. For a failure, write what happened and add captures: **Capture** takes the visible area, **Pick elements** lets you click what's wrong. Click a capture's thumbnail to mark it up.
4. Choose **Send results**.

planpage then:

- ticks the items that passed and unticks the ones that didn't (items you skipped stay as they were);
- opens a comment thread on each item that failed or was blocked, with your note and screenshots;
- adds the run to the document's history, with each item's result.

The run is kept in the panel until you send or discard it, so you can close the panel, or the browser, and carry on later. If someone changes the checklist while you're testing, sending is refused; reopen the panel to get the new list.

Agents see the run in `get_feedback`, under `reviewRuns`: who ran it, when, in which browser and on which page, and each item's result and note. Each failure's thread carries the screenshots with their element details. Agents can record a run themselves with `record_review_run`.

## Permissions and privacy

The extension asks for as little as it can:

| Permission | Why |
|---|---|
| Active tab | To capture the tab you're on, only when you click a capture button or press a shortcut. |
| Scripting | To draw the region and element pickers on the page, and to record diagnostics when you turn them on. |
| Storage | Your sign-in and settings, and the review run in progress. Captures waiting to be sent stay in the extension's own storage and are cleared after a week. |
| Identity | To open the planpage sign-in window. |
| Side panel | The review panel. |
| planpa.ge | To send captures, and so planpage's pages can ask it to open a page for review. |

It doesn't read pages in the background, and nothing leaves your browser until you send it. It sends only to the planpage server it's signed in to.
