A browser debugging tool for the bugs you can’t reproduce

Classic debugging advice says reproduce first. That works for deterministic bugs on your own machine. It falls apart for a race condition that shows up on a slow connection, a feature flag enabled for one account, a stale cache, or a 500 from an upstream service that recovered ten minutes later. “Cannot reproduce” is often the honest answer, and the ticket is closed with the bug still live.

A web debugging tool that runs where the bug happens changes the order of operations. Instead of trying to recreate the state, you read what the browser logged when the state was wrong. Bugmark does that as a Chrome extension: it keeps a rolling record of console output, network traffic and user actions in the tab, and freezes it into a report the instant someone captures the problem.

• JavaScript errors and unhandled promise rejections, alongside the last 150 console messages. An await that threw with no catch often produces no visible UI change at all; here it appears in the log.
• The last 150 network requests, with status and timing. Filter to failures to see the 4xx and 5xx responses first, then open a fetch or XHR request to read its headers, payload and response body (bodies are kept for the 60 most recent, up to 16 KB request and 32 KB response).
• Steps to reproduce: the last 30 actions, including clicks with element labels, form submits and SPA route changes, which tells you the sequence that led into the broken state.
• Environment: browser, OS, viewport, pixel ratio and scroll position, the variables that most often explain why a bug is intermittent.

A support agent reports that “Confirm and pay” does nothing for a customer who applied a promo code. Here is how the report narrows it down:

1. The steps show: open checkout, apply promo, type in the name field, click Confirm and pay. The promo step is the unusual one.
2. Filtering the network log to failures shows POST /api/payments/intents returned a 4xx.
3. The request payload shows the discounted amount; the response body explains the rejection reason.
4. The console shows an unhandled promise rejection from the checkout handler, which is why the button appeared to do nothing rather than showing an error.
5. The developer copies the request as cURL, replays it against staging with the same payload, and has a failing case in minutes.

Two bugs come out of one report: the pricing logic that produced an invalid amount, and the missing error handling that hid it. Neither needed a live reproduction session with the customer.

Notice what the agent did not need to know. They never opened DevTools, never identified which request mattered, and never wrote steps. They reproduced the problem once on the customer's account in their own browser, pressed a shortcut and wrote one sentence. The debugging skill stayed with the developer; the evidence collection moved to wherever the bug was seen.

For bugs that happen over several screens, start a recording with ⌥⇧R before walking the flow. Recordings run up to 5 minutes, continue across page loads, and can include microphone narration so the tester can say what they expected at each step. The console and network logs are attached to the recording in the same way as a screenshot. The screen recorder for bug reports page covers this workflow in more detail.

Bugmark is a capture tool, not an interactive JavaScript debugging tool. It does not set breakpoints, step through code, watch variables, map minified stack traces back to source, or profile CPU and memory. It does not record every session in the background either: logs are a rolling window of the most recent activity in the tab, and a video only exists if someone pressed record.

The intended hand-off is simple. Use the report to find which request failed and which error was thrown, then download the HAR into DevTools or replay the request with cURL and set your breakpoints there. If you want to collect logs manually for comparison, see how to capture console logs.

Front-end and full-stack developers debugging reports from staging or production. Support engineers who need to escalate with proof. QA testers on exploratory sessions, where the interesting failures are rarely scripted; see the QA testing tool page. And anyone who has closed a ticket as “cannot reproduce” and suspected the bug was still there.

Bugmark is free, needs no account and keeps data on your device. It works in Chrome, Edge, Brave and Arc. Add Bugmark to Chrome — free.