QUICK ANSWER
Show Zed’s agent what you mean by recording your screen with Clipy. Talk through the change, point to the relevant part of the app, and share the link. The recording gives your request context, so you can work from the interface in front of you instead of translating it into a long prompt.
Make the changes I explain in this Clipy recording: [link].
How do you give Zed's agent a recording?
Paste the normal Clipy recording link into Zed's agent and say what you want help with. You do not need to edit the URL or configure a server to try a public recording.
Read this Clipy recording and help me with the change I described: [paste the recording link]. Tell me what you understood before editing.
What if Clipy is not connected yet?
Start with the recording link and the task. If the agent needs the Clipy skill or account access, ask it to handle setup and guide you through anything it needs. You can keep explaining what you want in ordinary language.
How do you ask about your own recording library?
Ask the agent to connect your Clipy library when you want to search private recordings. Follow the sign-in prompt if needed. A public recording link can be read without library access; private material still requires the right account.
Connect my Clipy library, then find the recording where we discussed the webhook retry logic.
The connection reference is available for the agent to use when needed.
What should you ask the agent to do?
Read this Clipy recording. List the changes I asked for, then implement the agreed changes. Verify them in the app and attach Clipy proof to the GitHub PR.
Name anything that must stay unchanged. That keeps a focused request from turning into a redesign. The Clipy skill supplies the recording workflow; your prompt supplies the goal.
When a recording is the right input
Recording is worth it when the bug has a when or a which, and not otherwise.
- Timing: a spinner that never resolves, a double-click race, a toast that disappears too quickly to read.
- Pointing: anything you would explain by gesturing at the screen. "That one, not the other one" is unactionable as prose and trivial as a frame with a coordinate.
- Visual judgement: spacing, contrast, alignment, the thing that looks wrong but you cannot name.
- Multi-step state: flows where step six only breaks because of what happened at step two.
For a stack trace, paste the stack trace.
How do you get proof back in the PR?
Verify this change in the running app. Record the verification with Clipy and attach the Clipy proof link to this GitHub PR: [PR link]. Show the problem before the fix and the result after it.
The agent handles recording and sharing. Open the proof link in the PR to see what it actually checked. Ask it to include the important screens and a reload when the change should persist.
A recording shows the run that was captured, not every possible case. If the agent cannot record in its current environment or cannot update the PR, it should say so. A walkthrough assembled from screenshots should be labeled as such, not presented as continuous screen capture.
What if the agent gets stuck?
Ask it what it could read or record and what is missing. It may need the Clipy skill, permission to open the link, a sign-in, or access to the app being tested. Let it use the troubleshooting guide instead of asking you to debug commands.
If the recording is still processing, let the agent wait for the available context. If a private link is inaccessible, connect the right account. Do not make private work public just to bypass an access problem.