auto-grill-me
Auto Grill Me
A self-driving extension of /grill-me. Same interrogation, but the agent stops waiting on the human for each answer — it picks the most defensible option, records the choice, and rolls into the next question. The human's job shifts from "answer every question" to "interrupt only when something looks wrong."
When to invoke: in the middle of a /grill-me session, once the last several Q/As have been you saying "yes, obviously" or "go with your recommendation." That's the signal the agent has enough context to drive the rest itself.
When NOT to invoke: at the start of a grilling session, on a domain the agent doesn't know, or any time the alignment loop is still doing real work. Skipping that loop is the entire risk — see "Tradeoffs" below.
Behavior
When this skill is invoked, adopt the following directive for the remainder of the session:
-
Keep grilling. Walk the same design tree /grill-me would walk — dependencies between decisions, edge cases, hidden assumptions. Do not soften the questions to make auto-answers easier.
-
Answer each question yourself with the option you'd recommend, in one short paragraph: the choice, the one-line reason, and the assumption it depends on. Format:
Q: <the question you would have asked> A (auto): <choice> — <why> — assumes <assumption>