Brainstorm Before You Build: A Design-First Skill for Coding Agents

Where this fits

A developer asks their coding agent to “add a dark mode toggle” and the agent immediately starts editing CSS. The result: a half-baked implementation that doesn't account for contrast, theming variables, or user preferences. This skill forces the agent to pause and run an interactive design discussion whenever you explicitly invoke use brainstorming or use awesome-code. It solves the common problem of agents jumping to code before the actual requirements are understood, turning coding sessions into collaborative design dialogues.

Why agents benefit

• Enforcement of the iron law — NO IMPLEMENTATION WITHOUT DESIGN DISCUSSION FIRST: the skill hard-stops the agent from writing a single line until you've agreed on approach, trade-offs, and scope.
• Structured clarification: the agent asks targeted questions to decompose vague requests (e.g., “What does ‘optimize’ mean for you?”) and surfaces hidden constraints you didn't mention.
• Alternative design generation: it proposes 2–3 concrete design options with pros/cons, so you're not limited to the first idea the agent thinks of.
• Design documentation: it persists the discussion and decisions into a task workspace under ./.bensz-api/ — useful for audit trails and future maintenance.
• YAGNI enforcement: it ruthlessly strips out “nice-to-have” features you didn't actually need, preventing scope creep.

Practical scenarios

Designing a new user onboarding flow

You're unsure whether to use a wizard, a single-page form, or a progressive disclosure pattern. The agent launches a brainstorming session, asks about user segments and drop-off points, then presents three flows with interaction costs and implementation complexity.

Refactoring a legacy module

“Improve this payment service” is too vague. With brainstorming, the agent first maps out the current API surface, asks about idempotency requirements and retry policies, then suggests refactoring strategies (strategy pattern vs. adapter) before touching code.

Defining an API contract for a microservice

Instead of the agent auto-generating a REST spec, it uses brainstorming to discuss resource naming, error handling, and versioning with you. The final design is approved before any scaffold is generated.

Add it to your agent workflow

Install the skill with:

npx skills add huangwb8/skills --skill brainstorming

Then trigger it in a conversation:

use brainstorming to design a rate-limiting mechanism for our public API

The agent will reply with a structured set of questions, multiple design proposals, and a recommendation — no code until you explicitly approve a direction.

Compare before adopting

Before committing, evaluate the skill against:

• Quality signals: the GitHub repo has 48 stars and is maintained by huangwb8. Check issue response time and recent commit activity.
• Maintenance freshness: review the last update date and whether it follows current agent-skill conventions.
• Alternatives: consider lighter-weight approaches — e.g., just writing “don't implement yet, ask me questions” in your prompt. The skill adds structure and persistence, but might be overkill for trivial tasks.
• Workflow fit: does your team already have a design-review ritual? If so, this skill can automate the initial brainstorming phase; if not, it might introduce process overhead.

Why it is worth tracking

The brainstorming skill addresses a real pain point in agent-assisted development: premature implementation. Its explicit “iron law” and structured workspace make it a good model for how design-first principles can be encoded into agent behavior. With 48 stars and an active author, it's a niche but growing project worth evaluating when you want your agent to think before it writes. You can also discover similar skills on OpenAgentSkill to see how it compares with other design-focused agent skills. If you regularly build features that require user feedback or complex trade-offs, tracking this skill could save you from countless reworks.