# Bug reports developers can act on: what to capture

> The difference between "the checkout is broken" and a report a developer can start on without a single question is four pieces of context, all of which the browser already has.

Published 2026-10-08, about Snag.

Ask a developer what they want from a bug report and the answer is always the same: how do I make it happen?
Most reports do not say.
"The checkout is broken" is a feeling; a report is a path.
The good news is that the browser already holds everything a report needs at the moment the bug appears.
The job is to capture it then, not to reconstruct it later.

## The four things to capture

**The page and the screenshot.**
The URL, the title, and what the screen showed, marked up so the problem is circled rather than described.
A screenshot with a circle around a greyed-out button says more than a paragraph.

**The steps that led there.**
The last navigations, clicks and inputs before the capture.
Not a description of them; the actual trail, with the element each click landed on.
This is what turns a report into a reproduction, and later into a test.

**The console.**
The error the page logged when it broke.
In many front-end bugs the console names the function and the line, so a developer who has it can open the right file instead of first trying to make the bug happen again.

**The network.**
The request that failed and its status.
When "the save button does nothing", the cause is often a request that failed without the page saying so, and the failed request names the endpoint and the status the server returned.

## The same bug, reported twice

A customer's promo code is rejected at checkout.
This is the report a developer usually gets:

```text
Checkout is broken, please fix ASAP
```

No page, no browser, no steps, no error.
The developer's first move is a question back.

This is the report the browser could have filed at the same moment:

```text
Promo code SAVE10 rejected at checkout

Page:      https://shop.example.com/cart
Browser:   Chrome 130 on macOS
Steps:     1. Open the brass pen  2. Add to cart  3. Open the cart
           4. Type SAVE10 in Promo code  5. Click Apply
Expected:  "Code applied" and the total drops by 10%
Actual:    "This code is not valid", total unchanged
Console:   TypeError: Cannot read properties of undefined (reading 'percent')
           at applyPromo (cart.js:212)
Network:   POST /api/cart/promo  500 Internal Server Error
Screenshot: the cart, with the error message circled
```

The second one says where to look before anyone opens the code.

## What not to capture

Anything the reporter typed into a password field, and anything that would embarrass them in a screenshot.
Cover sensitive text with a solid box before you file: a blur can leave large text or a short known value readable, and a solid box leaves nothing to work from.
A capture tool should never record secrets.
A report that leaks a password is worse than no report.

Writing a report by hand?
The [free tools](https://testfinch.com/tools/) include a [bug report template](https://testfinch.com/tools/bug-report-template/) with these sections and a [redact tool](https://testfinch.com/tools/redact-screenshot/) that covers private details with a solid box before you attach a screenshot.

## From report to fix, without a meeting

With those four things attached, the triage conversation changes.
Severity can be set from the evidence.
The developer can start without asking a question.
A coding agent can too: give it the report over MCP and it has the error, the request and the steps to investigate from, which is more than most humans get.

## From fix to proof

The last step is the one most teams skip.
The steps in the report are a test.
Run them again after the fix on a fresh session and assert that the error is gone.
If the bug comes back, that test fails, and the next agent that changes the page finds out before a customer does.
