Superpowers’ brainstorming skill now asks whether you want to pick the platform, or leave it to the agent
1 month ago
Superpowers’ brainstorming skill now finds out early which choices about how a project gets made (platform, language, libraries, medium) you want to make yourself, and states any defaults it picks out loud. Merged to the dev branch; measured in Claude Code, inconclusive on Codex. https://github.com/obra/superpowers/pull/2482
Ask about this presentation
Chapters
0:00Superpowers’ brainstorming skill now asks whether you want to pick the platform, or leave it to the agent
0:19Asked to design an iOS app, brainstorming built a web app
0:34A clean Codex session on dev made the same mistake
0:46A new “How it gets made” step asks early who owns each choice
0:57What it does next depends on who it’s working with
1:09A default the agent picks goes in the design as its own call
1:21The same step covers a painting or a cake
1:33In Claude Code, every hobbyist run now offered a say before choosing
1:47On Codex, the results are inconclusive
1:59It still guesses the kitchen equipment before planning a cake
2:12The change is on the dev branch, after release 6.4.2
Show transcriptHide transcript
Superpowers changed its brainstorming skill on October 9. Before, someone asked for an iOS app and got a web app, because the agent picked the platform without asking. Now it asks first: do you want to pick the platform, language and libraries yourself, or should the agent pick them and tell you what it chose?
Drew asked Codex, running Superpowers 6.4.2, to design an iOS app for managing OCD. Brainstorming never asked about platform. The design was a web app with a home-screen icon. Drew wanted a native app.
In a fresh Codex session, brainstorming talked through the experience for three turns. Then it announced a phone-friendly prototype, and built a web app without ever bringing up platform.
Every project rests on choices like platform, language and libraries. Early on, the agent now asks which of those you want to make, which you're handing over, and what you already have.
Some users have firm views on their stack, and novices want the agent to decide. It talks it through with the first, picks defaults out loud for the second, and builds on an existing project's stack.
A choice the agent makes goes in the design as its call, never as something you agreed to. Filling in defaults used to be a red flag. Now it's fine, if the agent says them out loud.
Brainstorming gets used outside software too. The first draft only covered technology. The merged version says medium, tools, materials and resources, so it applies to a painting or a cake.
In Claude Code with Claude Opus 5.5, tested against simulated people, every hobbyist run asked whether they wanted a say, then stated defaults. On an existing Flask app, all three runs named the stack.
On Codex, where Drew's failure happened, the authors don't claim an improvement. Most runs stalled on approval prompts. They report no regression, and hand testing on Codex comes next.
One gap remains. Planning a scaled-up cake, the agent still assumed which pans the baker owned and let the baker correct it. The assumption was said out loud, and silent choices were Drew's failure.
It's on the dev branch, so the latest release, 6.4.2, still has the old wording. The full results are in pull request 2482 in the Superpowers repository.
Up next
2:54Anthropic’s Excel skill stops LibreOffice from rewriting your workbook
@skill-repos
4 hours ago
2:01Agent Skills’ frontend skill stops installing test runners just to check its work
@skill-repos
15 hours ago
2:23Trail of Bits’ taint-trace skill adds the source and sink models your taint analyzer is missing
@skill-repos
2 days ago
1:30Anthropic’s Word, PDF, PowerPoint and Excel skills are now Apache 2.0
@skill-repos
1 week ago
2:32Trail of Bits' cvss-scoring skill turns a bug description into a CVSS v4.0 vector
@skill-repos
1 week ago
3:02Superpowers 7.0.2 fixes skill loading on Pi, OpenCode and Copilot for Windows
@skill-repos
1 week ago
2:12Superpowers’ brainstorming skill now asks why first and stops asking filler questions
@skill-repos
1 week ago
1:37Agent Skills' code review skill now drops findings your code already rules out
@skill-repos
1 week ago
2:10Superpowers 7.0.1 keeps its five-round fix limit after Claude Code compacts
@skill-repos
3 weeks ago
2:11Trail of Bits' regression-hunt skill only reports a bug a fresh agent can reproduce
@skill-repos
3 weeks ago
2:30Superpowers’ brainstorming skill now reads your project before it asks you anything
@skill-repos
3 weeks ago
2:07Trail of Bits' gh-cli plugin stops long Claude Code sessions from breaking every Bash call
@skill-repos
1 month ago