Skip to main content

brainstorm

AI DevKit · Use when the user asks to brainstorm, ideate, generate ideas, expand options, challenge ideas, pressure-test ideas, compare concepts, narrow choices, name something, plan content angles, explore strategy, evaluate product ideas, technical approaches, experiments, or decisions.

来源信息

仓库
codeaholicguy/ai-devkit
最近来源活动
2026年8月22日 15:05
检测到的 SKILL.md 语言
英语
星标
1,644
分支
256

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
brainstorm
description
AI DevKit · Use when the user asks to brainstorm, ideate, generate ideas, expand options, challenge ideas, pressure-test ideas, compare concepts, narrow choices, name something, plan content angles, explore strategy, evaluate product ideas, technical approaches, experiments, or decisions.
# Brainstorm Run a compact diverge-to-converge loop. ## Workflow 1. Frame: goal, constraints, audience, success criteria, assumptions. 2. Baseline: state the simplest honest solution first. Every option must beat it; the baseline stays a candidate. 3. Diverge: distinct options across practical, high-leverage, and unusual angles. 4. Challenge: weak assumptions, failure modes, tradeoffs, rejections. Always: reject speculative generality (each abstraction, layer, flag, and data copy needs a current caller) and verify load-bearing "needed" claims in the consuming code, not doc assertions. 5. Compare: only relevant criteria, plus deletion cost (how easily the option can be removed later). Ties go to the smaller option. 6. Converge: strongest 3 picks with rationale and a next step. Self-review first: for each new constant, object, layer, and flag ask "why do we need this?" Cut answers that are only messaging or future-proofing. ## Formats Use Quick by default. Use Deep only for ambiguous or high-stakes decisions. Ask at most one clarifying question only if missing context would materially change the brainstorm; otherwise state assumptions and proceed. - Quick: clusters, challenges, 3 picks. - Deep: assumptions, clusters, evaluation table, recommendation. - Naming: tone groups, fit, ambiguity, pronunciation. - Technical: approaches, tradeoffs, risks, validation. - Content: angles, hooks, audiences, outlines. Avoid filler, near-duplicates, generic best practices, premature recommendations, and scope without a current caller. If asked for more ideas, explore a new axis before listing variants.
在 GitHub 查看