grilllisted
Install: claude install-skill YasMax91/groundwork
# Grill
Interview the user until the thinking holds. This is the clarify protocol run on its own — outside the
task pipeline, with no classification, no impact map, and no spec. Use it on a plan, an architecture
call, a product idea, or an intent that has not become a task yet, including work that never becomes
code.
Run the rounds per `${CLAUDE_SKILL_DIR}/../../guidelines/clarify-protocol.md`: find every fact yourself,
put every genuine decision to the user, ask the frontier in one `AskUserQuestion` call per round, lead
each question with your recommendation and its consequence, and let each answer push the frontier
outward.
This skill is the **manual entry point** — for grilling something that is not a task yet. Inside a task,
the agent offers the same loop and, once you accept, runs it directly (see "Unbounded mode" in
`${CLAUDE_SKILL_DIR}/../../guidelines/clarify-protocol.md`); you never have to type the command to get it.
Two things differ from the in-pipeline loop:
- **No round cap.** The user asked to be grilled and ends it by saying so. Keep going while the frontier
refills.
- **Explore for real.** With no discovery phase behind you, the facts are unmapped — read the code, the
CRD, and the docs as questions demand them, and spawn subagents for anything wide. A fact you skipped
becomes a question you should not have asked.
## Close with the decision summary
When the frontier empties, report:
1. **Settled** — each decision and the option chosen.
2. **Still