Bug report checklist

Run this before you file a bug. Check each box so the report has everything an engineer — or a coding agent — needs to fix it in one pass, not three rounds of questions.

A bug report has one job: get the bug fixed without a round of "how do I reproduce this?" This is the complete list of what to include before you file. Check each box so the report stands on its own — whether the fixer is an engineer or a coding agent reading the markdown.

Identify

  • A clear, specific title — what's broken, not just "it doesn't work"
  • Where it happens: the page, screen, or feature
  • Severity: how badly it blocks the user, and how many users it hits

Reproduce

  • Numbered steps to reproduce, starting from a known state
  • What you expected to happen
  • What actually happened
  • How often it happens — every time, or intermittently Without the frequency, an intermittent bug gets closed as "can't reproduce".

Evidence

  • A screenshot or recording, annotated to point at the actual problem
  • The exact URL or screen where it occurred
  • Console or network errors, if the bug is technical

Context

  • Browser, OS, and device
  • Account or role, if the bug is permission- or data-specific
  • Anything that changed recently — a deploy, a setting, new data

Handoff

  • Everything the fixer needs is in one place, not scattered across a chat thread A complete report is the difference between a one-pass fix and three rounds of back-and-forth.

Frequently asked questions

What makes a bug report good?

Reproducibility. A fixer needs to see the bug happen: exact steps from a known starting point, what you expected, what happened instead, and an annotated screenshot showing where. Everything else — severity, environment, recent changes — helps them prioritise and narrow it down.

How is a bug report for an AI coding agent different?

It isn't, much — an agent needs the same completeness a human engineer does, just in a format it can read. A screenshot with numbered callouts plus the repro steps as text gives an agent both the visual signal and a clean index of what to fix. That's exactly what Cobalt Capture produces.

Capture your first review.

About a minute from open tab to a shareable URL your agent can ingest.

Start capturing