How do I stop superpowers v7 brainstorming from asking pointless opener questions?
With superpowers 7.0.0 in Claude Code (2.1.302, Opus 5.5, no other plugins), the brainstorming skill opens with "if you'd rather skip the questions, say so" and a broad, unhelpful question — even when the prompt references an existing spec or plan, e.g.
/brainstorming Implement part B of @spec-name. The questions are usually already answered by the repo or earlier context, wasting time and tokens. How can I get the old 6.4.x behavior back?Originally asked on GitHub: https://github.com/obra/superpowers/issues/2530
6 answers
Still open — no official fix was released for the forced brainstorming questions introduced in superpowers 7.0.0. The thread only contains user workarounds.
What worked for others:
- Push back in the chat. One user replied "Why do you ask me this? We already have a plan, you just need to implement it." The skill steered back to the existing plan and the next phase ran without questions (comment 1).
- Downgrade to 6.4.2. One user on Opencode 2 (muse spark 1.3) found that after downgrading from 7.0.0, brainstorming read the code first and asked concrete questions (e.g. whether the app should use AWS CLI or keys) instead of the generic opener (comment 11). Another commenter reports 6.4.1 never produced the forced opener.
- Repeat the pushback each session. One user had to push back every time the skill loaded; the model apologised, agreed it already had the context, and started the actual design — once admitting it had followed the 7.0.0 SKILL.md "Your First Message" template word for word (comment 6). This works per session but must be repeated.
About the built-in skip option ("if you'd rather skip the questions, say so"): the issue reporter explicitly rejects it, worried that telling the model to skip questions will make it pick implementation directions on its own. Another commenter, however, likes the short-circuit option and just builds (comment 5). No one in the thread reports what the skip alone produces.
A community fork, superpowers-remix (opens external site) (based on 7.0.0), restores the 6.4.2 brainstorming flow and moves the 7.0.0 Q&A into a separate
discovering-intentskill that only runs when invoked. The author tested it against simonw/sqlite-utils, but no other user reports trying it.Did this work for you?What @katyatitkova tried on GitHub (Worked)
Replied "Why do you ask me this? We already have a plan, you just need to implement it" to the v7 opener. The skill steered back to the existing plan, and the next phase ran without questions.
Source: https://github.com/obra/superpowers/issues/2530#issuecomment-6213012400 (opens external site)
Did this work for you?What @alexxburg tried on GitHub (Worked)
Downgraded from superpowers 7.0.0 to 6.4.2 (Opencode 2, muse spark 1.3). Brainstorming then read the code first and asked concrete questions, e.g. whether to access AWS via CLI or keys, instead of the generic opener.
Source: https://github.com/obra/superpowers/issues/2530#issuecomment-6276157652 (opens external site)
Did this work for you?What @Biomaj tried on GitHub (Partly worked)
Pushed back every time the 7.0.0 opener loaded; the model apologised and started the actual design. It worked per session but had to be repeated, and the model once admitted it had followed the SKILL.md 'Your First Message' template verbatim.
Source: https://github.com/obra/superpowers/issues/2530#issuecomment-6251561877 (opens external site)
Did this work for you?What @fcavalieri tried on GitHub (Outcome unclear)
Forked 7.0.0 into superpowers-remix, restoring the 6.4.2 brainstorming flow and moving the 7.0.0 Q&A into a separate discovering-intent skill. The author's README documents the changes, but no other user reports using it.
Source: https://github.com/obra/superpowers/issues/2530#issuecomment-6277087385 (opens external site)
Did this work for you?What a GitHub user tried on GitHub (Didn't work)
Ran /brainstorming with v7 on existing work (implementing part B of a three-PR spec, fixing a tech-debt item, reviewing integration test coverage). Each time the skill asked high-level questions — past PR experience, why you're fixing the debt, hand-written vs ORM queries — instead of proceeding with the design work described in the prompt.
Source: https://github.com/obra/superpowers/issues/2530 (opens external site)
Did this work for you?