soda-propose

Investigate and propose multiple approaches for the given problem or goal.

Use English for internal reasoning (thinking). All user-facing output — problem restatement, proposals, comparisons, AskUserQuestion options, and Proposal Summary — must be in Japanese.

If $ARGUMENTS is empty, ask the user what they want to explore before proceeding.

## Phase 1: Problem Understanding

Restate the problem described in $ARGUMENTS in your own words. Include your best assessment of:

- **Scope**: What appears to be in scope and out of scope
- **Priority axis**: What seems to matter most (e.g., speed, maintainability, compatibility, UX quality)
- **Constraints**: Any apparent technical or business constraints

Present this understanding to the user, then use AskUserQuestion to confirm:
- "This understanding is correct"
- "Scope needs adjustment"
- "Priority axis is different"
- "There are additional constraints"

If the user selects anything other than confirmation, incorporate their feedback (via free-text input), revise your understanding, and re-present. Do NOT proceed to Phase 2 until the user confirms.

If consensus cannot be reached after 2 confirmation attempts, use AskUserQuestion: "Run exploratory investigation first, then redefine the problem" / "Try defining the problem one more time" / "End this exploration".

## Phase 2: Investigation

### Step 1: Common Context

Launch a sub-agent (Task, subagent_type: Explore) with a prompt that includes:
- The problem statement and scope confirmed in Phase 1
- Instruction to report: project structure, key dependencies, existing patterns, and relevant conventions

Summarize the agent's findings into a Common Context block for use in Step 2.

### Step 2: Focused Investigation

For each focused agent, construct the prompt to include:
1. The Common Context block from Step 1 (summarized, not raw output)
2. The specific investigation question for this agent
3. The scope and priority axis confirmed in Phase 1

Launch 1-3 agents in para