grilllisted

# Grill Grill surfaces unknowns by questioning only the current decision frontier; it is not a questionnaire. By default its structured close routes to `to-spec`, `wayfinder`, or a local brief and executable plan. Say `quick`, `just talk`, or `don't write files` to keep only the chat log. ## Use / skip Use for fuzzy scope, hidden constraints, or an explicit request to stress-test an idea. Skip bugs, documented scope, clear acceptance criteria, and P0 work. ## Protocol Run **`@lib/interview.md`**. It owns cadence, facts-first lookup, delete-first drilling, pushback, search space, stopping rule, and escape hatch; do not restate or fork it here. **If the interview already ran this session, do not run it again.** Carry its answers into the close. **If an interview is still unfinished when the user switches skills, say so before the next question.** The shared protocol owns the exact announcement and the receiving caller owns the new close. ## Chat-only output When the user requested no artifact, emit only: ```markdown ## Grill log — <topic> — <date> ### Confirmed ### Open / deferred ### Surfaced contradictions ### Recommended next step ``` Otherwise this log is the running record and the structured close follows. ## Structured close After the interview stopping rule, run Stage A, Stage B, then exactly one route. Do not skip or reorder these stages. - **Stage A — Alternatives.** Use `@lib/stop-rule.md`: produce 2–3 named approaches with effort, pros, cons, a recomme