| name | grill |
| description | Fire BEFORE building anything non-trivial, or whenever a request is vague, broad, or likely to grow. A structured interview that surfaces the real scope — deploy target, distribution, proof bar, docs/media, design bar, positioning — and gets explicit sign-off before any brief is written. The front-end mirror of the close-out report. Model- or user-invocable. Inspired by mattpocock/skills `/grill`. |
Grill — align before you build
"No-one knows exactly what they want." The first stated ask is rarely the
whole ask. Grilling extracts the real one before you commit to a plan, so
you don't discover the scope mid-build (every late discovery forces a re-plan).
When to fire
- Before decomposing any multi-step or multi-round build (it's the front end
of
cto-orchestration Role 1).
- Any time the request is vague ("make it better", "build me an X"), broad,
or smells like it will grow.
- When you catch yourself about to assume a scope decision the human never made.
When NOT to fire
- A single, fully-specified, reversible task. Just do it.
- Mid-flight, for one new requirement — that's a normal supplement, not a
reason to stop and re-grill. (But 3+ mid-flight additions = you skipped
this skill; grill next time.)
The loop
- Restate the ask in one sentence, then ask "is that right?" Cheapest
possible misalignment catch.
- Surface the load-bearing scope questions — ask only the ones whose
answer changes the plan:
- Deploy — where does this run in production? (cloud / container / host / nowhere yet?)
- Distribution — does anything get published? (a package, a registry, a tag, a link?)
- Proof bar — what counts as done/tested? (it compiles / unit / e2e / recorded demo / a reviewer pass?)
- Docs & media — README only, or tutorial + screenshots + video?
- Design bar — functional-first, or branded/polished?
- Positioning — does it need to beat or compare against alternatives?
- Constraints — deadline, stack lock-ins, things you must not touch.
- Name what you're assuming for anything they didn't answer, and let them
correct it. Defaults are fine; silent defaults are not.
- Play back the resolved scope as a short list and get a yes before fanning out.
Output
A 5–10 line scope contract: the one-sentence goal, the answered scope
questions, the explicit assumptions, and the out-of-scope list. This becomes
the top of the first brief (and, in a CTO round, anchors the frozen contract).
Anti-patterns
- Interrogating with 20 questions. Ask the 4–6 that move the plan; infer the rest.
- Grilling a trivial task into paralysis. Match grill depth to build size.
- Treating the asked ask as the whole ask. The job is the real one underneath.
Pairs with
cto-orchestration (Role 1), domain-modeling (the answers seed CONTEXT.md),
and the close-out report (grill opens the round; the report closes it).