The Lab · Feature assessment

One real run of the Opportunity Engine

I gave a 15-agent pipeline a two-paragraph feature idea and two screenshots of the Claude mobile app. This walkthrough shows exactly how I used the tool and what it produced at every step, with the real screens, files, and output.

This is a real run, published as it happened. The engine ran once on this idea (July 13, 2026) and everything it produced was saved to disk. These pages show that saved output so you can step through it without me paying for inference every time someone clicks. Nothing is generated live, and nothing has been cleaned up, including the findings that argue against my own feature idea.

How the run happened

the exact procedure, four moves

1. I filled in the engine's intake form and dropped in two screenshots.  2. "Save run folder" wrote the input to disk.  3. I ran one command in Claude Code.  4. Five stages executed and wrote 16 documents and a working prototype. Each step below shows the real thing, not a description of it.

$ /propose inputs/cluade-mobile-app-menu-reorder

Or skip the walkthrough

The end result is a complete assessment package: a one-page verdict, a 21-row risk register, PRD, design spec, tech spec, and the embedded prototype, formatted the way a stakeholder would receive it.