Brainstormingv2

Brainstorming v2: Ideas Into Designs

A structured process for turning rough ideas into validated, actionable designs through collaborative dialogue — before any code is written.

Why This Matters

Most wasted engineering effort comes from building the wrong thing, not building it wrong. This skill ensures alignment on what to build and why before anyone touches implementation.

<HARD-GATE> Do NOT write code, scaffold projects, invoke implementation skills, or take any implementation action until you have presented a design and the user has explicitly approved it. This applies to EVERY project regardless of perceived simplicity. </HARD-GATE>

Anti-Pattern: "This Is Too Simple To Need A Design"

Every project goes through this process. A todo list, a single-function utility, a config change — all of them. "Simple" projects are where unexamined assumptions cause the most wasted work. The design can be short (a few sentences for truly simple projects), but you MUST present it and get approval.

Process

You MUST complete these phases in order:

Phase 1 — Discover Context

Understand the project landscape before asking the user anything.

• Read project files, README, docs, recent git history
• Identify existing patterns, conventions, constraints
• Note reusable components and where new work would fit

See instructions/discovery.md for detailed guidance.

Phase 2 — Ask Clarifying Questions

Sharpen the idea through focused, sequential questions.

• One question per message — never dump multiple questions at once
• Start with "what problem does this solve?" not "what should we build?"
• Prefer multiple-choice questions when you can offer informed options
• Focus on: purpose, users, constraints, success criteria, scope boundaries
• Stop when you can explain the idea back and the user confirms

See instructions/questioning.md for question frameworks and anti-patterns.

Phase 3 — Propose Approaches

Present 2–3 distinct approaches with honest trade-offs.

• Lead with your recommendation and explain why
• Cover: complexity, maintainability, time-to-ship, risk
• Apply YAGNI ruthlessly — strip unnecessary scope from every option
• Let the user pick, combine, or propose something different

See instructions/approaches.md for how to frame options effectively.

Phase 4 — Present the Design

Walk through the chosen design section by section, validating incrementally.

• Scale each section to its actual complexity
• Cover: architecture, components, data model, user behavior, error handling, testing
• After each section: "Does this look right so far?"
• Revise before continuing — don't accumulate disagreements

See instructions/design-presentation.md for section templates and the checkpoint pattern.

Phase 5 — Document and Transition

• Write the approved design to docs/plans/YYYY-MM-DD-<topic>-design.md
• Commit the design document to git
• Invoke the writing-plans skill to create an implementation plan
• Do NOT invoke any other skill — writing-plans is the only next step

Key Principles

• One question at a time — don't overwhelm
• Problem before solution — understand the "why" before the "what"
• YAGNI ruthlessly — cut scope early and often
• Explore alternatives — always propose 2–3 approaches before settling
• Incremental validation — present design section by section, get approval at each step
• Every project gets a design — no exceptions, scale the depth to the complexity
• Be flexible — go back and re-clarify whenever something feels off

Approaches — Framing Options with Honest Trade-offs

Always propose 2–3 distinct approaches before settling on a direction. This prevents anchoring on the first idea and gives the user meaningful choices.

Structure Each Option

For each approach, cover:

• What it is — one-sentence summary
• How it works — brief description of the mechanism
• Pros — concrete benefits (not vague "more flexible")
• Cons — honest downsides (not hand-waved "slightly more complex")
• Best when — the scenario where this option wins

Example

Option A: Server-side pagination (recommended)

Fetch one page at a time from the API. Simple to implement, scales to any dataset size.

- Pros: Low memory use, fast initial load, works with millions of rows

- Cons: Each page turn requires a network request

- Best when: Dataset is large or unbounded

>

Option B: Client-side pagination

Fetch all data upfront, paginate in the browser.

- Pros: Instant page turns after initial load, simpler API

- Cons: Slow initial load for large datasets, high memory use

- Best when: Dataset is small and bounded (< 500 rows)

>

Option C: Hybrid with virtual scrolling

Fetch data in chunks, render only visible rows with virtual scrolling.

- Pros: Smooth UX, handles large datasets, reduces DOM nodes

- Cons: More implementation effort, requires a virtualization library

- Best when: UX is a priority and the team has capacity for extra complexity

Presenting Your Recommendation

Lead with your recommended option and say why:

I'd recommend Option A because your dataset will grow over time and you mentioned performance matters. Option B would be simpler but won't hold up past a few hundred rows.

Let the user pick, combine, or propose something different. Don't be attached to your recommendation.

YAGNI Check

Before presenting, review each option and strip out:

• Features that solve hypothetical future problems
• Abstractions that only matter if requirements change
• Configurability that nobody asked for
• "Nice to haves" disguised as requirements

The right design is the simplest one that solves the stated problem.

When Options Aren't Useful

Sometimes there's only one reasonable approach. In that case, say so:

There's really one clear way to do this: [approach]. Here's why the alternatives don't make sense: [reasons]. Does this direction work for you?

Don't fabricate artificial alternatives just to hit the "2–3 options" target.

Design Presentation — Walking Through the Design

Present the design section by section, getting approval at each step. Never dump the entire design at once.

Section Order

Scale each section to its complexity. A simple feature might need one sentence per section. A complex one might need a few paragraphs.

1. Overview

What are we building and why? One paragraph max. Tie it back to the problem statement from Phase 2.

2. Architecture

Where does it live in the system? What components are involved? How do they connect?

For simple features: "New endpoint in the existing API, new React component in the dashboard."

For complex features: describe the data flow between components. Use a short list or diagram if it helps.

3. Data Model

What data does this feature create, read, update, or delete? What's the shape?

Skip this section entirely if the feature doesn't touch persistent data.

4. User-Facing Behavior

What does the user see and do? Walk through the key interactions step by step.

For UI work, describe the states: empty, loading, populated, error, edge cases.

5. Error Handling

What can go wrong? How does the system respond? What does the user see?

Focus on likely failures. Don't enumerate every theoretical edge case — cover the ones that will actually happen.

6. Testing Strategy

What gives confidence this works? Be specific about what to test.

Good: "Test that expired tokens return 401 and redirect to login" Bad: "Add comprehensive test coverage"

The Checkpoint Pattern

After each section, pause:

"Does this look right so far? Anything you'd change before I continue?"

If the user pushes back: 1. Acknowledge the concern 2. Revise that section 3. Re-present and confirm 4. Only then continue to the next section

Don't accumulate disagreements — resolve them in place.

Design Document

Once all sections are approved, write to docs/plans/YYYY-MM-DD-<topic>-design.md:

# <Feature Name> — Design

## Problem
<What problem this solves, from Phase 2>

## Decision
<Chosen approach from Phase 3, and why>

## Design

### Architecture
<From section 2>

### Data Model
<From section 3, if applicable>

### User-Facing Behavior
<From section 4>

### Error Handling
<From section 5>

### Testing Strategy
<From section 6>

## Out of Scope
<What was explicitly excluded>

## Open Questions
<Anything deferred or unresolved>

Commit this file to git before transitioning to the writing-plans skill.

Discovery — Understanding the Project Landscape

Before asking the user a single question, build context on your own. The better you understand what exists, the sharper your questions will be.

What to Examine

1. README and docs/ — project purpose, architecture, conventions 2. Recent commits (git log --oneline -20) — what's actively being worked on 3. Directory structure — how the project is organized, where new code would live 4. Existing patterns — how similar features were built before (routes, components, tests) 5. Dependencies — what libraries and frameworks are already in use 6. Open issues or TODOs — related work that's planned or in progress

What to Look For

• Constraints you shouldn't violate — established patterns, naming conventions, architectural boundaries
• Reusable pieces — existing utilities, components, or abstractions the new feature can leverage
• Gaps — missing tests, docs, or infrastructure the new work might expose
• Recent direction — what the last 5–10 commits tell you about current priorities

How Deep to Go

Scale your discovery to the project size:

• Small project (< 20 files): Skim the whole thing. Read the main entry point and any config files.
• Medium project: Focus on the area relevant to the idea. Read related modules and their tests.
• Large project: Start with docs and directory structure. Drill into the specific subsystem that will be affected.

Output

After discovery, you should be able to answer:

• What does this project do?
• How is it structured?
• What patterns does it follow?
• Where would the new work fit in?

Don't present discovery findings as a report to the user. Use them to inform the questions you ask in Phase 2.

Questioning — Sharpening the Idea

Good questions are the core of brainstorming. Ask one question per message. Listen to the answer before choosing the next question.

Question Sequence

Start broad, then narrow:

1. Problem — "What problem does this solve?" or "What's the pain point today?" 2. Users — "Who will use this? How do they interact with it?" 3. Success — "How will you know this works? What does 'done' look like?" 4. Scope — "What's explicitly out of scope for the first version?" 5. Constraints — "Are there performance, compatibility, or timeline constraints?" 6. Edge cases — "What happens when [unusual condition]?"

You don't always need all six. Stop when you can explain the idea back and the user agrees.

Question Types

Multiple choice — when you can offer informed options:

Should we store this in (a) the existing user table, (b) a new preferences table, or (c) local storage?

Open-ended — when the design space is wide:

What should happen when a user loses network connectivity mid-upload?

Confirmation — to validate your understanding:

So the goal is to let admins bulk-invite users via CSV, and they should get feedback on which rows failed — is that right?

Prioritization — when scope needs trimming:

If we had to ship with only two of these three features, which would you cut?

Knowing When to Stop

You have enough information when:

• You can describe the feature in your own words and the user agrees
• You understand who uses it, what triggers it, and what success looks like
• You know what's out of scope
• You can name at least two reasonable implementation approaches

Anti-Patterns

• Question dumps — asking 3+ questions in one message. Split them up.
• Leading questions — "Should we use React?" when you haven't explored alternatives.
• Premature detail — asking about database indexes before understanding the feature.
• Ignoring answers — asking something the user already addressed. Track what you know.
• Interrogation mode — asking too many questions when the idea is already clear. Read the room.