deep-think

Deep Think v2 — Adaptive multi-persona reasoning with evidence-based depth. Uses tiered complexity (2-5 agents), targeted critique→reflect cycles, convergence shortcuts, and 3-pass PENCIL-inspired synthesis. Detects execution environment to choose between Agent Teams (normal mode) and Task Orchestration (plan mode). Use when the user prefixes with "deep think", "딥씽크", "깊게 생각해", or requests thorough analysis. Best for complex architecture, debugging, algorithmic, or multi-domain problems. NOT for simple lookups. Requires: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 (Track A only)