clarify
Skill: clarify
Understand what the user needs before committing to work.
LLMs default to assuming rather than asking — models overwhelmingly proceed without asking even when information is missing (70% of cases in controlled studies). This skill serves as the plan pipeline's clarifier and counteracts that bias by classifying what's unclear and generating targeted clarifying questions.
adaptive-questioning.md at ${CLAUDE_PLUGIN_ROOT}/references/adaptive-questioning.md is a shared reference: the plan pipeline's clarify gates also link to it (Phase 3 planning checkpoints use it for mode-switching).
Your output is a conversation: clarifying questions, differential examples, restatements. Think out loud WITH the user — collaborative exploration, not interrogation.
The fault-type and hypothesis machinery below is your analysis for choosing the right question — not something to narrate. The user sees the question itself, and where useful a differential example; never "this is a Parameter fault" or a listed set of interpretations.
Classifying What's Unclear
Not all gaps are the same. Classifying the type determines what kind of question to ask.