An editor for hiding private details in screen recordings.

Turning failed attempts into a more useful brief.

I defined the workflow, designed the editing interactions, and directed development with AI coding agents. I tested the results against my own recordings and changed the requirements when the output was not usable.

The central judgment was deciding what to expect from automation. I had tried repeated approaches to finding and covering sensitive text. When those attempts still left visible fragments, improving detection alone was no longer enough. I needed direct control over the correction.

From automatic detection to an editable Mac app.

The work developed through several iterations. In July 2026, I was asking for masks that could move over time and for a project both a person and an AI agent could edit. In August, with recordings waiting to be finished, I narrowed one version to the essential manual-masking workflow.

The native Mac editor was built with Apple’s interface and video frameworks. A companion command-line tool lets an agent prepare work without clicking through the app. Both use the same saved project and redaction engine, so proposed masks can remain editable when I open the visual editor.

That shared foundation connected preparation, human review, and export. It also made the product’s limits clearer: scans can point me toward a problem, but I still need to inspect whether the mask actually covers it.

Keep automatic work correctable.

Text can move, animate, or become partly hidden behind another window. Automatic attempts missed details in those situations. A completed-looking video was not useful if I could still see information that should have been covered.

I shifted the goal toward a useful starting point with manual controls. Scans and alerts could identify places to review; I could place, resize, and correct the masks. That reduced the promise of the automation while giving me a way to finish the job.

Treat timing as part of every mask.

Covering the right rectangle in one frame is not enough. A mask has to stay over the sensitive area for the relevant part of the recording. If the area moves, the cover must move with it or be adjusted.

I explored timeline controls and keyframes: points in time where a mask’s position or size is specified. I also tested a simpler manual version when the larger approach was holding up production. These were different iterations, with a tradeoff between more precise control and a smaller tool I could put to use.

Preserve both the original and my corrections.

There were two things an automated step could accidentally undo: the source recording and a mask I had already checked. I made preserving them part of the workflow. The original remains available, and I explicitly required agent work to retain approved manual masks.

Saving editable masks alongside the project gives me a way to revisit a decision. Writing the result back into Screen Studio keeps the task inside the editing process I already use. If an export or correction needs another pass, I have a route back.