Agent Skills: Brainstorming Ideas

Brainstorming Ideas

Turn a vague idea or draft plan into a well-formed design. Ask one question at a time. If the answer is discoverable from code, inspect code instead of asking.

Core principles

• One question at a time.
• YAGNI at every step.
• Present design in small sections and confirm before continuing.
• Propose 2-3 concrete options with trade-offs.
• Use project domain language from CONTEXT.md / CONTEXT-MAP.md when present.
• Offer ADRs only for hard-to-reverse, surprising, real trade-off decisions.

Step 0: Load domain context

Before design questions, look for relevant project docs:

• CONTEXT.md
• CONTEXT-MAP.md
• docs/adr/
• nearest */CONTEXT.md or */docs/adr/

Read them when present. Use those terms in questions and designs. If no docs exist, create them lazily only when a real term or decision is resolved.

Step 1: Understand the idea

Ask: "What's the idea or plan you'd like to explore?"

Follow up one question at a time:

• What pain triggered this?
• Who uses it?
• What existing feature does it build on or replace?
• What is explicitly out of scope?

Stop when you can state the problem in one sentence.

Step 2: Grill the plan when requested

If the user passed plan, grill, or asked to challenge/stress-test a plan, walk the decision tree. Keep it focused on design quality and assumptions, not implementation task breakdown; use the spec-plan skill for tasks.

For the full interview discipline (phase order, per-question format, final summary, failure handling), read references/grill-protocol.md.

For each question:

• Provide your recommended answer.
• Explain why the branch matters.
• Inspect code instead of asking when possible.
• Flag vocabulary conflicts with CONTEXT.md.
• Use concrete edge-case scenarios.

When a domain term is resolved, propose a tight CONTEXT.md entry:

**Term**:
One-sentence definition.
_Avoid_: overloaded synonym

Offer an ADR only when all three are true:

1. Hard to reverse.
2. Surprising without context.
3. Result of a real trade-off.

Step 3: Surface requirements and assumptions

Use 5WH, skipping anything already clear:

1. WHO uses it?
2. WHY is it needed?
3. WHAT is the core capability?
4. WHERE does it live?
5. HOW should it work, if there is a strong constraint?

Then state assumptions explicitly and ask which are wrong.

Step 4: Explore the codebase

Find:

• similar modules or flows
• conventions and testing patterns
• integration points
• constraints from ADRs or existing architecture

Summarize in 3-5 bullets. Use project vocabulary.

Step 5: Research external solutions only if requested

Compare patterns, trade-offs, and common failure modes. Summarize before proposing approaches.

Step 6: Propose approaches

Present 2-3 options. For each:

• What: one-sentence approach
• Trade-offs: complexity vs flexibility
• Best when: scenario where it wins

Mark one as recommended. Ask which fits best.

Step 7: Detail the chosen design

Present ~200-word sections and confirm after each:

1. Architecture overview
2. Data flow
3. API or interface
4. Error handling
5. Testing strategy

Apply YAGNI at each section. Cut speculative pieces.

Step 8: Capture outcome

If the outcome is more than a short answer, offer to write a concise design note:

docs/plans/YYYY-MM-DD-<topic>-design.md

Include only: Problem, Chosen approach, Trade-offs, Open questions, Testing strategy.

If domain terms or decisions crystallized, update CONTEXT.md or create a short ADR only with user approval.

Failure handling

• Idea conflicts with CONTEXT.md / CONTEXT-MAP.md: surface the contradiction explicitly ("the glossary defines X as A, this idea assumes B"). Resolve the term with the user before designing — do not silently pick one.
• A constraint blocks every approach: stop generating options. State the blocker, what would unblock it, and ask the user to relax the constraint or change scope.
• Idea is too vague to design (user cannot state the problem in one sentence): stay in Step 1. Ask one narrowing question at a time; do not fabricate requirements or jump to approaches.

Output format

BRAINSTORM COMPLETE
===================
Topic: <topic>
Approach chosen: <name>
Design note: docs/plans/YYYY-MM-DD-<topic>-design.md or none

Key decisions:
- <decision>

Domain docs:
- <CONTEXT/ADR updates or none>

Open questions:
- <unresolved>