deep-interview
The clarification skill that removes ambiguity before any planning or code change — the real bug or feature, affected behavior, non-goals, and acceptance checks.
deep-interview is the first step of the working loop. It removes ambiguity before any planning or code changes, so the rest of the loop builds on a clear, agreed understanding rather than a guess.
What it clarifies
Before a plan is drawn or a file is touched, deep-interview pins down:
- The real bug or feature — what is actually being asked for, separated from how it was first phrased.
- Affected behavior — what user-visible or system behavior is in scope.
- Non-goals — what is explicitly out of scope, so the work does not sprawl.
- Acceptance checks — how you will know the work is done and correct.
Usage
Invoke it inside a GJC session with /skill:<name>:
/skill:deep-interview clarify the bug, affected behavior, non-goals, and acceptance checksDescribe the problem in your own words; the skill asks the questions needed to turn a vague request into something concrete enough to plan against.
You can tune interview depth with --quick, --standard, or --deep:
/skill:deep-interview --deep <vague idea>When to use it
Use it when
The request is vague, the blast radius is unclear, the bug report is incomplete, or the feature has implicit constraints that were never stated.
Skip it when
The task is already fully specified and small — there is nothing left to disambiguate.
Where it fits in the loop
deep-interview -> ralplan -> ultragoaldeep-interview feeds a clarified problem statement into ralplan, which turns it into a reviewed plan before any mutation. Skipping clarification tends to surface as rework later in ultragoal, once findings contradict an assumption that was never checked.