| name | fw-brainstorm-approach |
| description | Question cadence for structured discovery. Load when a brief is ambiguous and a spec needs to be written. |
Brainstorm Approach
Used by the fw-draft agent when the brief is fuzzy and a spec cannot be written without
more context. Defines the question order, the conversation rules, and the threshold for
"enough information to write a spec."
Without this, discovery degrades in predictable ways: too many questions at once, questions
about the solution before the problem is understood, or moving to spec writing while open
questions still exist.
Question order
Ask in this order. Each question is asked only after the previous answer is clear.
- What is the problem being solved? Not the solution. Not the feature. The situation or
pain that motivates the work.
- Who encounters this problem and when? The actor and the trigger. A problem nobody hits
does not need solving.
- What does success look like from their perspective? Observable outcomes from the actor's
point of view. Not implementation milestones.
- What are the constraints? Technical, time, scope. What must be true. What cannot change.
- What are the edge cases or failure modes worth anticipating? Not exhaustive — the ones
the user already knows about or is worried about.
Rules for the conversation
- Ask one question at a time. Batching questions forces the user to context-switch and
produces shallower answers. One question, one answer, then the next.
- Do not suggest solutions during discovery. The goal is to understand the problem fully.
Solutions during discovery anchor the conversation to whatever was suggested first, even
when the suggestion is poor.
- Do not proceed to spec writing until all five areas have a clear answer. A clear answer
means the user has stated the answer themselves — not nodded along to your interpretation.
- If a later answer reveals an earlier one needs revisiting, do so before continuing.
Discovery is not strictly linear. A constraint discovered at step 4 can change what success
means at step 3. Loop back, confirm the revision, then continue.
Threshold for "enough information"
All four conditions must hold before writing the spec:
- The problem is clearly stated and the user has confirmed your statement of it.
- Success is defined in observable terms — not "it should work well" but "the user can
do X without Y."
- At least the primary constraints are known — the ones that would shape the solution
space, not every minor preference.
- No open question remains that would force a major assumption during implementation.
If any condition fails, ask the next question. Do not paper over a gap with an assumption —
record it as an open question in the spec instead, and decide with the user whether to
resolve it now or defer.