vc-intent-clarify

Score the request across four signals: ambiguous scope, missing explicit path, multiple possible intents, and first interaction. Scores of 0-1 auto-route silently, score 2 shows a routing summary, and scores of 3 or higher trigger a full structured clarification suite with multi-choice questions.

SIMPLE mode runs in the main thread with a 3-5 file light research pass to fill question options. DEEP mode spawns a research subagent first that returns discovered file paths, decision dimensions, and risk surfaces, producing questions grounded in actual codebase findings.

Auto-skip conditions force Tier 0 routing: continuation phrases like "go" or "proceed", mid-phase-program execution, trivial single-file fixes under 15 lines, explicit mode commands, resuming an active plan, or pure information questions with one obvious routing target.

Each of four signals adds one point: ambiguous scope, no explicit path, multiple intents, and first interaction. The sum determines the tier: 0-1 is Tier 0 silent auto-route, 2 is Tier 1 routing summary with confirmation, and 3-4 is Tier 2 full structured clarification.

State what remains unclear in one sentence, then ask one final direct plain question. If still unresolvable, default to the research agent with the narrowest reasonable scope. Clarification never loops more than twice before routing.

Yes. The included validation script fails output missing a restatement line, an ambiguity score from 0-7, or an unambiguous mode choice of exactly simple or deep followed by a reason or because justification.