grill-me

The interview focuses on decisions that could change your proposal: ownership, data flow, failure behavior, rollout, and proof. Each question includes a recommended answer and explains why the decision matters.

Install the skill with the command above. Invoke it explicitly with /grill-me in Claude or $grill-me in Codex. Point it at the proposal, for example /grill-me docs/sync-redesign.md.

Use it to challenge an architecture or implementation approach, uncover vague scope or hidden assumptions, and settle decisions before building. If you need a specification, plan, or tickets as the result, use planning.

The agent reads the proposal and its sources first. It looks up repository facts in code, tests, and documentation instead of asking you to find them. Then it asks one consequential question at a time. For example:

Question: Should exports run on a schedule or only on request?
Recommended answer: Only on request; daily data is not needed (assumption).
Why it matters: A schedule adds a job runner and alerts to the proposal.

An answer without evidence is labeled as an assumption. The agent settles a decision before asking one that depends on it. Questions follow the proposal's open decisions; they are not a fixed checklist. The interview stops when the proposal is coherent, a required fact is unavailable, or you ask it to stop.

SKILL.md has the interview and log contract.

• Resolved decisions and their reasons.
• Changed assumptions, remaining risks, and unknowns.
• Recommended changes to the proposal.

Decisions stay in conversation unless you request a record. A saved Markdown log records material decisions and changed assumptions, not a transcript. Apart from that log, grill me changes no files. It does not write code, specifications, or tickets.