Skip to main content

Workflow skill

Brainstorm Skill

Generate, challenge, and narrow ideas, from a simple baseline to three strong picks.

On this page

Install command

AI DevKit installs its built-in skills as one curated set. Run this once to install brainstorm with the rest of the built-in workflow skills.

bash
npx ai-devkit@latest skill add --built-in

What it does

Use brainstorm when you want options before committing: product ideas, technical approaches, names, content angles, or decisions. It frames the goal, starts from the simplest honest baseline, explores distinct options, challenges assumptions and over-engineering, and converges on three picks with a next step.

When to use it

  • Compare technical approaches before writing a design
  • Pressure-test an idea for weak assumptions and failure modes
  • Generate and narrow names, content angles, or experiment ideas

Supported agents

  • Claude Code
  • Codex
  • Cursor
  • opencode
  • Gemini CLI

The SKILL.md file

This is what your agent reads when the skill runs. View it on GitHub (opens in a new tab)

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.

esc