Walkthrough Recording: Talk Through the Bug and Let the Agent Fix It

I built Toado because I was tired of being the clipboard between Chrome and Claude Code. The extension fixed most of that. Click, capture, ticket, agent. But there was a second problem I did not fix on the first pass, and it was the bigger one.

I do not file most of the bugs I find.

Here is what a review session looks like for me. I open the app, I click around, and inside five minutes I have seen six things that are wrong. The dropdown clips on the settings page. The empty state has the old copy. Saving a project flashes the wrong toast. And so on. Then I have to stop, go back to the first one, capture it, write a sentence about what I expected, capture the second one, write another sentence, and by the fourth one I have talked myself into “I’ll remember the rest.” I never remember the rest.

The screenshot was never the hard part. The writing was.

What walkthrough recording does

So the newest thing in Toado is a walkthrough recorder, and the idea is that you look at the app and talk. Toado does the writing.

You open the pin control on any page and press Record. The extension starts recording the tab and your microphone. You use the app the way you would if a teammate were sitting next to you. When you see something wrong, you drop a pin on it (Alt+C, then click the element) and say what you see. “This dropdown is getting cut off on the right, it should be sitting inside the card.” Then you keep going.

When you stop, Toado transcribes what you said and splits it by pin. Each pin becomes its own ticket. The ticket gets:

• A screenshot of the page at the moment you pinned
• The exact DOM element you pointed at, anchored so it can be found again later
• The transcript of what you said around that pin, as the description
• A clip of the screen for that moment, if you recorded the screen

Five minutes of clicking around and complaining turns into six tickets on the board. I did not type any of them.

The part that matters for the agent

I want to be specific here, because “AI reads your bug report” is a phrase that has been stretched to cover a lot of things that do not work.

A ticket that says “dropdown broken on settings page” sends an agent on a scavenger hunt. It has to guess which dropdown, guess what broken means, and guess what you wanted instead. It will produce something. It might even be right. But you are going to spend the time you saved on the report reviewing a fix that missed.

A walkthrough ticket sends the agent to the line. The transcript says what you expected and what happened instead, in your words. The element says exactly where. The screenshot shows it. If there is a clip, the agent can pull that too. Over MCP, Claude Code calls get_ticket and gets the description and transcript, then get_capture_asset for the screenshot and clip. That is the whole handoff. No copy-paste, no re-explaining, no “let me find that screenshot.”

The difference in the fixes is not subtle. When the agent knows which element and what you meant, it goes and does the thing. When it has to guess, it guesses.

Two modes, because video is expensive

Video is the most expensive thing Toado stores. On real walkthroughs it is about 95 percent of the bytes. So there are two modes, and you pick one in the extension settings.

Screen and narration records the tab and your voice. Every ticket gets a clip. This is the one you want for motion bugs: flicker, layout shift, an animation that stutters, anything you cannot explain in a still. It runs around 6 MB a minute.

Narration only records your voice and nothing on screen. Each ticket still gets its screenshot, the element you pointed at, and the transcript, because the pins are captured through the browser separately from the recording. What you give up is the clip. What you get is roughly forty times less to upload, around half a megabyte a minute, and nothing on your screen leaves the machine. This is the mode for copy, spacing, and logic bugs, which is most of them.

A project can also set the mode for everyone who records in it. If your project documents a visual product, set it to screen. If most of your tickets are copy and layout, set it to narration only and stop thinking about it.

What we keep, and for how long

I want to be upfront about this because it is the kind of thing that gets buried in a pricing page.

Video clips are kept for 7 days on the free tier and 45 days on Personal and Team. After that the clip is deleted. The ticket, the screenshot, the pinned element, and the transcript are kept as long as the ticket exists, on every plan. The transcript is what your agent reads, so it is never the thing that expires.

The audio is transcribed by OpenAI. Under their API terms it is not used to train models. The full disclosure is on the privacy page, and the retention detail is on the security page.

How I use it now

My review sessions changed shape. I record for three or four minutes, I talk, I drop pins, I stop. A minute later the tickets are on the board. I skim them, fix a word or two if the transcription got a product name wrong, and then Claude Code picks them up one at a time through MCP. By the time I am back from getting coffee, half of them are in review.

The thing I like most is that I file the bugs I would have forgotten. The sixth bug in a session used to be the one that never got written down. Now it costs me one sentence out loud.

If you want to try it, walkthrough recording is included on every plan, including the free tier. Install the extension, open the thing that is bugging you, press Record, and complain.