The Idea Pipeline

I have more ideas than I have weekends, and most of them are bad. The trouble is that I can't tell which ones are bad while I'm still excited about them, so I built a skill that does the telling for me. I paste in one rambling paragraph, walk away, and Claude runs the whole thing through three councils without stopping to ask me anything.

The screenshots on this page all came out of one run. I typed a paragraph about helping a crew of Scouts get to Eagle, walked away, and came back to three prototypes I could click on. This is the dreamy one, where the whole crew climbs a mountain together:

First, it tries to kill your idea

Claude goes off and researches who already tried the idea and what happened to them, dead products included, and then three voices argue it out. The Money Voice only cares about who pays and whether you can reach three of them this week. The Skeptic explains why the last people failed. The Operator checks the idea against your actual skills and your actual hours. There are eight kill criteria and a single hit ends it. A dead idea gets an obituary, one paragraph, with a note on what would have survived.

If the idea lives, the second council isn't allowed to say no. A Fan, a Wildcard and a Craftsman go three rounds, and what they come up with gets sorted into three directions: dreamy, confident and grounded. An editor cuts each of those down to a one-page brief, and then three build agents head off in parallel and make the things. Real interaction, fake data, one smoke test each.

It has to look good, or it doesn't count

This is the part I care about most. A prototype that works but looks like a tax form tells you nothing about the idea, because nobody wants to touch it, including you. You end up judging the gray boxes and not the thing you actually thought of.

So the design rules are baked into the skill. You don't need to install anything else. Before a single line of code, each build has to write down a mood, a palette and a pair of typefaces. Every screen gets one thing you can read from across the room. The fake data has to feel true, with real names and believable numbers. And the three prototypes aren't allowed to look like siblings, which is why the mountain up top, the phone app and this coach's board look like they came from three different shops:

Same idea, three completely different answers, and I didn't pick a single color. If you already have design skills installed Claude will load those too, but the bar is set by the skill itself.

It works for you, and for the good stuff too

The first time you run it, Claude asks a handful of questions about you: who you can already reach, how many hours you have, what you flatly won't do, and what number would make an idea worth your while. It saves that as a profile and never asks again. And if the idea is something you want to do out of the goodness of your heart, say so in the paragraph! The first council eases off on the money and asks who it helps and whether it can keep itself going.