brainstorm

name	brainstorm
description	Freeform ideation and structured shortlisting. Generates 8-10 approaches including wild cards, then converges on 2-4 viable options with tradeoffs. Use before /forge or standalone for any creative exploration.
metadata	{"workbench.argument-hint":"<topic or problem statement>"}

Path resolution: This skill may run from any repo. All context/ and config.yaml paths are relative to the workbench root, not the current working directory. Read ~/.codex/workbench-root or ~/.claude/workbench-root to get the absolute workbench path, then prepend it to all context/ and config.yaml references. See PATHS.md.

/brainstorm — Divergent Ideation + Convergent Shortlisting

Generate approaches to a problem before committing to a direction. Two modes: divergent (quantity, wild ideas) then convergent (evaluate, shortlist).

Also invoked as optional Phase 0 inside /forge.

Usage

/brainstorm <topic>                    # Standalone brainstorm
/brainstorm "observability for MCP"    # With quoted description

Process

Step 1: Understand the Problem

If the user's prompt is vague, use AskUserQuestion to clarify:

• What problem are you trying to solve?
• Any constraints or non-negotiables?
• What have you already ruled out?

If the prompt is clear, proceed directly.

Step 2: Freeform Ideation (Divergent)

Generate 8-10 approaches to the problem:

• Include the obvious solutions (the user expects these)
• Include at least 2 "wild card" ideas that challenge assumptions about the problem itself
• No filtering, no judgment — quantity over quality
• Each approach: 1-line description + 1-line "why this might work"

Present as a numbered list. Do NOT evaluate yet.

Step 3: Structured Shortlist (Convergent)

Use AskUserQuestion to understand evaluation criteria if not already known:

• "What matters most — speed to ship, long-term maintainability, simplicity, or something else?"

Then evaluate all approaches against known constraints and criteria. Produce a shortlist of 2-4 viable approaches:

For each shortlisted approach:

• Relative complexity: low / medium / high (NOT effort estimates — those come later with full context)
• Risk level: low / medium / high
• Key assumption: what must be true for this approach to work
• Tradeoff: what you give up by choosing this

Step 4: User Selection

Use AskUserQuestion:

• "Which approach(es) do you want to explore further?"
• Options: each shortlisted approach + "Combine elements from multiple" + "None — brainstorm more"

If "brainstorm more" → loop back to Step 2 with refined constraints (max 2 loops).

Step 5: Output

Standalone mode: If saving the brainstorm, read context/standards/html-plan-standard.md and save the primary artifact to context/notes/brainstorm/YYYY-MM-DD-slug.html with:

• The full ideation list
• The shortlist with evaluations
• The selected approach(es)
• Suggest: "Run /forge <name> to plan this."

Create a .md file only as a short compatibility synopsis if another workflow needs Markdown.

Inside /forge: Write to state.md:

• Selected approach description
• What's known (from brainstorm discussion)
• What's NOT yet known (for INTAKE to focus on)
• Rejected approaches and why (prevents re-research)

Key Principles

1. Diverge before converging. Don't skip the wild ideas phase. The best solutions often come from combining unexpected approaches.
2. No effort estimates. Use relative complexity. Effort estimation requires constraints, team size, and timeline — brainstorm doesn't have those yet.
3. Challenge the problem, not just the solution. At least 2 wild cards should question whether the problem as stated is the right problem to solve.
4. User picks, system doesn't. Present options with tradeoffs. Don't recommend. The user's intuition about what fits their context is more valuable than algorithmic ranking at this stage.
5. Short and fast. Brainstorm should take 5-10 minutes, not 30. If it's taking longer, the problem needs research, not brainstorming.