| name | interview |
| description | Clarify the problem, values, constraints, non-goals, failure modes, and intended direction with minimal high-value questions. Writes .context/interview.md and hands naturally to /architect when enough is known. |
/interview
Read the shared contracts before output: ../_shared/operator-output.md, ../_shared/attention-budget.md, ../_shared/evidence-record.md, ../_shared/state.md, ../_shared/cognitive-load.md, ../_shared/collaboration.md.
Purpose
Turn uncertainty into a small set of locked decisions and open questions.
Inputs
User prompt, current conversation, repo evidence, .context/scout.md when present, issue/PR context, or a review comment that exposed uncertainty.
Reads
Repo evidence when a question is answerable by inspection. Do not ask the user for facts the repo can answer.
Writes
.context/interview.md.context/session-state.md
Process
- Identify the missing decision that blocks better work.
- Explore repo facts first when applicable.
- Ask 1-3 questions that reduce the most uncertainty.
- Provide a recommended answer for each question when a default is reasonable.
- Record locked decisions, rejected paths, values, constraints, non-goals, incentives, and remaining unknowns in
.context/interview.md.
- Update session state.
Operator Output
If more input is needed, output the current phase, what is already locked, 1-3 questions, and why each matters. If enough is known, output Core shape locked, three evidence bullets, risk, and suggest /architect.
Stop Conditions
Stop when the next architectural decision can be made responsibly, or when the user must choose between real tradeoffs.
