brainstorm
Brainstorm
Use this skill to guide a user from a rough idea to a clear design/spec by asking targeted questions, one at a time, after first inspecting the project state.
Workflow
-
Scan the project context
- Review the working directory to understand the current state, constraints, and any existing docs or code.
- Summarize what you saw in 2-4 sentences only if it helps frame the next question.
-
Ask refining questions (one per message)
- Ask exactly one question per response.
- Prefer multiple-choice questions with 3-5 options; include an "Other: ____" option when helpful.
- For multiple-choice questions, mark exactly one option as recommended so the user has a clear default.
- In Codex
request_user_input, mark the recommended choice by suffixing the label with(Recommended). In ClaudeAskUserQuestion, use an explicitrecommendedflag when available or clearly label one option as recommended. - Use open-ended questions only when options would be misleading.
- Sequence questions from highest-uncertainty to lowest-uncertainty.
- Do not ask questions that can be trivially answered by inspecting the repo or using available tools; look it up first.
- Keep each question tightly scoped and decision-oriented.
More from btraut/skills
beads-implement
Implement Beads or beads_rust work items from an issue id. Use when the user runs $beads-implement <bead-id> or asks to implement a bead/epic; if it is an epic, implement all sub-tasks, committing after each one and only interrupting when blocked or a decision is needed.
28beads-create
File Beads or beads_rust epics/issues from a finalized plan/spec AND do the polish pass (clarity, acceptance criteria, sizing, deps). Use when asked to create Beads from a plan/spec (OpenSpec, PRD, design doc), convert an external plan into issue tracker structure, or review/refine an existing Beads set.
27export-chatgpt
Export a ChatGPT shared conversation to Markdown or JSON via a share URL.
18review-team
Run multi-agent code review by profiling risk, selecting reviewer personas, spawning sub-agents, and merging structured findings into one severity-ordered report.
13