bf-brainstorm

Brainstorm

Converge a fuzzy idea into core stances, concept model, and constraints. Teamlead facilitates, PM + Architect drive (Designer or Security pulled in as needed). Usually 5-15 rounds; each locks 1-2 concepts.

Direct Invocation Guard

If bf-workflow is not active, STOP here. Load bf-workflow with the user's input; do nothing else in this skill until it routes back.

When to use / not use

Use	Don't use
New product starting up (→ bf-blueprint-write)	Implementation choices (SQLite vs Postgres → spec brief)
New module added to existing product	Milestone with stance already locked (→ bf-milestone-fourpiece)
Existing stances in conflict	Mechanical PRs, small patches, hotfixes
Major blueprint rewrite	—

Multi-round structure

Round	Focus	Output
1: Scope	Q1: first-class concepts (≤3)? Q2: relation to existing concepts? Q3: what's NOT in this module?	PM + Architect each answer ≤200 words
2-N: Stance debate	Pick 1-2 stances per round. Can you write "X is, Y isn't"? Conflicts? v0/v1 boundaries?	≤5-line stance draft per round
Final: Freeze	Teamlead summarizes 5-7 stances + constraints	PM + Architect sign off → bf-blueprint-write

Teamlead facilitation

Principle	Detail
Don't answer for others	Assign to PM/Architect, only arbitrate (conflict? v0/v1 clear? constraint writable?)
Push toward closure	Every round produces ≤5-line stance draft. "Wait for more info" = not allowed
Converge to 5-7	Too many stances → unmemorable, execution drifts. 10-15 = product total; 5-7 per module

Stance format (hard rule)

Each stance must contain:

Element	Requirement
Claim	≤30 words, something a user could repeat
Constraint	"X is, Y isn't" — prevents drift
Key scenario	A demo example you can actually run
v0/v1 boundary	What's done now, what's left for later

Worked example — "silence beats fake loading":

• Claim: don't show spinner / progress bar / "thinking..."
• Constraint: UI stays still while agent processes; only show result when done
• Scenario: agent edits artifact, user sees no intermediate state until commit
• v0: fully silent; v1 may add "thinking" hint (still no fake progress)

Anti-patterns

• ❌ Locking all stances in round 1 (no convergence, drags)
• ❌ Teamlead answering PM's user-side stance (no real perspective)
• ❌ Round produces no stance draft (just talk)
• ❌ Abstract platitudes (can't write constraint → redo round)
• ❌ Implementation details hijacking discussion (Teamlead cuts it off)

Output checklist

• All 5-7 stances have claim + constraint + scenario + v0/v1
• Constraints machine-checkable (reverse grep / assertion)
• PM + Architect signed off
• Handed off to bf-blueprint-write

How to invoke

follow skill bf-brainstorm