FIX WHAT IS BROKEN
Show your agent a UI bug, get the fix
The slow part of a UI bug is almost never the fix. It is getting another person, or an agent, to understand which thing is wrong out of the forty things on the screen. A recording removes that step: the agent sees the state, hears what you expected instead, and knows which element you were pointing at when you said it.
Quick answer
To fix a UI bug with an AI agent, record the bug with Clipy, point at the broken element and say what should have happened, then paste the link into Claude Code, Codex or Cursor. The agent reads the summary, the timestamped transcript, the key moment frames and the click evidence, so it knows which element is wrong before it touches the code.
What to record
Start on the screen before the bug, so the agent sees what normal looks like.
Do the thing that triggers it, at a normal pace.
Let the broken state sit on screen for a couple of seconds.
Point at the element that is wrong and say what should have happened.
Say this while you record
Say the expected behavior out loud. "This badge should say three, it says zero" is worth more than any amount of pointing.
The prompt to paste
Replace the link with your own recording's share URL, then paste this into Claude Code, Cursor, Codex, or whichever agent you connected.
Read https://clipy.online/video/<id> It is a screen recording of a bug in this repo. Before you change anything, tell me what you understood: what breaks, which screen it is on, what I said should happen instead, and which element I was pointing at. Then find the code responsible and propose a fix. If the recording does not tell you something you need, say what is missing instead of guessing.
Asking it to repeat the request back first is the cheapest check there is. If it misread the recording you find out before it edits anything.
What the agent gets from that link
A summary of what happened
A short written account of the recording, so the agent starts with the shape of the request rather than a wall of raw text.
A timestamped transcript
Your narration with times against it, when audio was captured and processed. This is where your reasoning lives, and it is usually the part that decides whether the agent fixes the right thing.
Key moments with frames
Selected instants from the recording with an image of the screen at that point, so the agent can look at the state you were describing instead of guessing at it.
Pointer and click evidence
Where you clicked and what you were pointing at, with the control's accessibility label when the capture recorded one. It turns "this button" into a specific element.
What changes
| Writing it down | Recording it instead | |
|---|---|---|
| Describing the bug | A paragraph you rewrite three times, plus a screenshot that misses the step before. | You do it once on screen and say what you expected. |
| What the agent knows | The words you managed to choose. | The frames, the click target, your narration with times against it. |
| Wrong-thing-fixed rate | High, because "the button on the right" is ambiguous in most layouts. | Lower, because the click evidence names the element. |
| Handing it to a teammate too | You write the description again, for a human this time. | The same link. They watch it, the agent reads it. |
Tips
Connect your agent once
Run the setup at /agent-setup once. After that every Clipy share link works in the same session you are already in.
Paste the link once the transcript shows
The transcript, summary and key moments appear on the watch page moments after Stop.
Narrate as you go
Narration gives the agent your intent, so it knows what "wrong" means instead of reading it from the pixels. It is the single biggest lift to the quality of the fix.
Record just the reproduction
Twenty minutes of working, with the bug somewhere inside it, buries the moment that matters. Record the reproduction and nothing else. Thirty seconds is usually plenty.
Common questions
Does the agent actually see the screen, or just the transcript?
Both, within limits. The recording context includes frames for the key moments Clipy selected, so the agent can look at the screen state at those points. It is not watching the video frame by frame. If the moment you care about is not one of the selected ones, say when it happens in your narration so the agent knows where to look.
Which agents can read a Clipy link?
Any agent you have connected through the setup at /agent-setup. That covers Claude Code, Cursor, Codex, Windsurf and OpenCode through the Clipy skill, plus anything that can use the MCP server.
Do I have to make the recording public?
No. A public watch link's context follows the same public access rules. For a private recording, the connected MCP server reads it with your authenticated account, so it stays private and your agent can still read it.
How long should the recording be?
As long as the reproduction takes and no longer. Most useful bug recordings are between fifteen seconds and a minute.
Other jobs
The same job, elsewhere on Clipy
This page is the job. These cover the same ground for a specific agent, a specific team, or as a tool you can open right now.
The full list is on the use-case index. If your agent is not connected yet, start at agent setup.