Rail Grill

Interview the user relentlessly about every aspect of the plan until shared understanding. Walk each branch of the decision tree, resolving dependencies one by one. For each question, provide your recommended answer.

Rules

• Ask one question at a time; wait for the answer before continuing
• Prefer multiple-choice when options are clear
• If a fact is findable in the environment, look it up — do not ask
• Decisions belong to the user — put each one to them
• Do not write any docs/monorail/ files (including CONTEXT.md / ADRs) — no codebase / intentionally stateless
• Do not implement code
• Every question uses the Question format below — without exception

Question format

A hard template. Every question must have exactly this structure:

**问题：** <question text>

**推荐：A**

- **A.** <conclusion>.<one line of reason>
- **B.** …

Constraints:

• The **问题：** line carries the question; the **推荐：A** line is separate and follows it
• A is always the recommended option. Write the recommended answer first and label it A — never list it in a later slot
• **推荐：A** is fixed text — the recommendation line never names a letter other than A
• Options are prefixed - **A.** , - **B.** , and so on, starting from the recommended one
• Option length is unconstrained — expand as the problem warrants
• No other phrasing of a question is permitted

Completion

When every decision-tree branch is resolved:

1. Output the plan — a concise synthesis of decisions (goals, scope, key choices, out-of-scope). Do not ask the user to confirm "is this the full scope?" or re-gate on shared understanding; the interview already did that work.
2. Ask only whether to start implementation, or if anything still needs adjusting. Wait for that answer.
3. Do not suggest /rail-setup, /rail-align, /rail-spec, or any other skill handoff in this closing message.