bounded-brainstorming

name	bounded-brainstorming
description	Use when the task is still ambiguous, needs scope decomposition or option comparison, or is inherently visual enough to justify a visual companion before implementation. Converts upstream brainstorming into a repo-owned, non-mandatory ideation step.

Bounded Brainstorming

Use this skill when the next move is not implementation yet, but decision-shaping.

When to Use

Use this skill when:

• the user explicitly asks for brainstorming, idea exploration, architecture options, or plan-only output
• the task is still too large or fuzzy to implement safely
• the work is UI-heavy, structure-heavy, or comparison-heavy enough that 2-3 options are useful
• a visual companion may help decide IA, wireframe, layout, or state presentation before implementation

Do not use this skill when:

• the user already gave a concrete implementation task and the scope is clear enough to start
• the task already has chosen direction in a PRD/design, handoff, or GitHub-backed task truth
• you are using brainstorming only as a way to delay implementation

Required Rules

1. This is an optional pre-implementation layer, not a universal gate.
2. Do not force per-section approval, mandatory spec drafting, or transition into a second planning system.
3. Keep the output anchored to repo truth: durable professional direction flows into PRD/design; mutable scope and execution planning flow into GitHub task issue evidence.
4. Prefer 2-3 concrete approaches with tradeoffs and one clear recommendation.
5. Only use a visual companion when the problem is inherently visual; do not turn browser mockups into default ceremony.
6. If scope is too large, split it into smaller, executable slices before implementation starts.
7. When the decision concerns gameplay loops, progression, balance, encounter/resource pressure, or player verb semantics, route the downstream brief or review to gameplay_designer.
8. When the decision concerns game visual direction, interaction feel, or player-facing screen flow, route the downstream brief or review to game_visual_interaction_designer.

Core Workflow

1. Restate the decision to make.
2. Decide whether the task is:
   • ready to implement now
   • too large and needs decomposition
   • ambiguous enough to need option comparison
   • visual enough to justify a visual companion
3. If decomposition is needed, split the work into smaller slices with clear boundaries.
4. Produce 2-3 approaches:
   • short description
   • main tradeoff
   • why it fits or does not fit oasis7 workflow truth
5. Recommend one approach.
6. If the problem is visual, decide whether to use agent-browser, gpt-image-2 when explicitly requested/authorized, or another repo-owned visual step to compare layouts, IA, states, or Image2 target images against real native/browser screenshots.
7. Write the chosen direction back into repo truth before implementation:
   • prd.md if scope/behavior/boundary changed
   • GitHub task issue evidence if execution path or affected surfaces changed
   • handoff if another role/subagent will execute it
   • GitHub task issue evidence comments if the decision matters to current task truth

Expected Output

BOUNDED BRAINSTORMING COMPLETE

## Decision
- Problem:
- Why this needs brainstorming now:

## Scope
- Ready now / needs split:
- Proposed slices:

## Options
1. [Option name] - [tradeoff summary]
2. [Option name] - [tradeoff summary]
3. [Option name] - [tradeoff summary, if needed]

## Recommendation
- Chosen direction:
- Why:

## Visual Companion
- Needed: yes/no
- If yes: artifact type and target question
- Target image path:
- Native/browser screenshot path:
- Comparison/gap notes path:
- Non-goal:

## Repo Truth Writeback
- `prd.md`:
- GitHub task issue evidence:
- handoff / GitHub task issue evidence:

Guardrails

• Do not leave the result as chat-only guidance if it changes task truth.
• Do not turn this into mandatory upfront ceremony for every request.
• Do not present option lists without a recommendation unless the user explicitly wants open-ended exploration.
• Do not claim completion from an Image2 target image alone; real UI screenshot, interaction smoke, or QA evidence is still required for implementation convergence.