Superpowers’ brainstorming skill now reads your project before it asks you anything

Superpowers fixed how its brainstorming skill opens, in a change merged October 20. In version 7, it offered to skip the questions even after you asked for brainstorming, and asked about your project before reading it. Now it reads your project first, plays back what you want, and asks one design question. If you hand it a spec, its first reply now starts from that spec.

Version 7 opened with two fixed lines: an offer to skip the questions, then a question about the moment you wished this existed. The skip line opened 9 of 12 sessions on version 7, against 0 of 134 on version 6.

It also asked before it looked. With part of a spec already built, it asked how the experience was. On a detailed request to build against AWS, it asked who opens the web app and what they do with the PNG, instead of how to reach AWS.

The fix adds a section called Look Before You Ask. When you hand the agent a spec, a plan or files, it reads them and the code they touch before its first message. It skips any question the project already answers.

If you've already said what you want, the first reply is a playback of about 200 words: what the project showed, and the two or three guesses that matter most. It ends with one line inviting corrections and one design question.

Where there's a real choice in how to build it, the agent recommends one approach, names the alternatives and what each trades, and offers to dig in. On that AWS request, Opus and Sonnet now recommend a Postgres driver, because the AWS CLI can't run queries.

The skip offer moves to the last line, and only when brainstorming started on its own, after a request like make a React to-do list. Across 12 runs each, it appeared every time the user hadn't asked, and in 0 runs where they had.

The fix landed on the dev branch. The version 7 release and the main branch still open with the skip line. It was tested in Claude Code on Opus, Sonnet, Haiku and Fable, and in Codex CLI on GPT-6 Astra and Luna, on earlier wording.

The 200 words is a target, and replies run longer, up to about 360 words on Sonnet. The full change is in Superpowers pull request 2560. Which agent do you run Superpowers brainstorming in?