Record the bug happening instead of describing it. Every reproduction step is preserved as a key moment, so a teammate — or a coding agent — can follow the sequence without watching the video.
Begin the recording before the bug happens, from a state someone else can get to. A recording that opens mid-failure proves the bug exists but not how to reach it, and reaching it is the expensive part.
Say what should happen right before each action. Expected-versus-actual is the single most useful thing in a bug report and the thing most often left out, because it is obvious to the person who found it.
After the failure, show the console, the network tab, or the error state. That is usually what turns a report into a fix without a round trip of questions.
Paste the link into the issue. If a coding agent is doing the triage, give it the recording's agent-readable file and it can work from the steps directly.
If you are explaining a change rather than reporting a failure, the code walkthrough recorder is the better fit. For how agents read recordings, see the agents page.
Written reproduction steps fail in a predictable way: the reporter leaves out the step that was obvious to them. The state they were already in, the account they were logged into, the width of their window. The person picking it up cannot reproduce it, asks a question, and the report sits for a day. A recording does not have that failure mode, because the sequence is whatever actually happened. It also carries the things nobody writes down — how long a thing took before it broke, what else was on screen, what the error actually said rather than a paraphrase. The trade is that video is slow to consume and impossible to search, which is what the text version of the recording is for.
Alongside the video, every recording produces a text document: the transcript, a summary, and a list of key moments, each with a timestamp and what was on screen when it happened. It is designed to be read by a coding agent rather than a person. Because this page declares the recording as a bug report before you start, the key-moment planner treats it as procedural. It keeps up to 100 moments rather than the usual 30, and is explicitly instructed to preserve every reproduction step, state change, error, and before-and-after comparison rather than summarising them. For a repro, losing three steps in the middle is the difference between a fix and a round trip. Agents can pull that document through Clipy's CLI or MCP server, so triage does not have to start with a human watching the clip.
A bug report is the recording type where losing detail costs the most, so it is the one where telling Clipy what it is up front changes the result the most. When a recording is known to be a bug report, the key-moment planner treats it as procedural: it keeps up to 100 moments rather than 30, and is instructed to preserve every reproduction step, state change, error, and before-and-after comparison instead of compressing them into a tidy summary. Normally that classification happens after the fact, from the transcript, and a quiet repro narrated in half-sentences is exactly the kind that gets read as something else. You can record from the general screen recorder and re-label it afterwards. Starting here just means the first pass is already right.
A bug is often reproduced against real data in a real account, and a screen recording captures all of it. Nothing is redacted automatically. Where the bug allows it, reproduce against test data. Share a single tab or window rather than your whole display so notifications and other applications stay out of the frame, and close anything holding credentials or customer records before you start. The recording is private until you share the link. If the repro has to touch real data, you can put a password on the link or give it an expiry date from the recording's settings rather than relying on the link staying unshared.
It is the same recorder, told in advance that this is a bug report. That changes how the recording is processed: procedural recordings keep up to 100 key moments instead of 30 and are processed to preserve every step rather than compress them.
Yes. Paste the share link into the ticket. It opens for anyone, including people outside your team, without an account.
Yes. Every recording has an agent-readable version with the transcript, summary, and key moments, and Clipy ships a CLI and MCP server so an agent can fetch it directly.
Pause the recording through the waiting and resume into the same file. One recording with the dead air removed is much easier to follow than several clips.
No watermark on any recording, on any plan. The free tier includes an allowance for how much you can record before upgrading.
Tell us what broke. We resolve every report within 24 hours.