eg-interview

# /eg-interview — interview-strategy intake (grill-me lineage) The top-of-funnel discipline for the *human-facing* agent: before decomposing or delegating a request, extract what the requester actually needs — by asking, not by assuming. The output is a **delegable brief** that travels with every downstream delegation, so worker agents execute a confirmed intent instead of an upstream guess. ## 1. When to run (and when not to) Run when a request is **broad, ambiguous, or high-stakes** and will fan out — a program-scale ask, a "make X better" with no criteria, work that commits resources or publishes externally. Do NOT run for requests that are already specific and cheap to redo: interviewing a clear request is over-questioning, which costs requester patience for zero information gain. Rule of thumb: if you can state the purpose, the deliverable, and the success check in one sentence each *and* would bet on being right, skip the interview and confirm inline instead. ## 2. What to press for (the six extraction targets) 1. **Purpose** — why this, why now; what changes for the requester when it lands. The stated task is often a means; the purpose is what survives redesign. 2. **Deliverable shape** — what artifact/state counts as the output (a document? a running service? a decision?). 3. **Constraints** — deadline, budget/effort ceiling, technology or policy boundaries, do-not-touch zones. 4. **Success criteria** — how the requester will judge it done; what verification they...