adaptive-reasoning

Adaptive reasoning optimizes tokens by evaluating task complexity through weighted scoring, activating deep thinking only for high-difficulty tasks while auto-downgrading to fast responses for simple queries. This prevents wasted processing on straightforward lookups and saves tokens.

To automatically adjust reasoning depth, present your code debugging or system design prompt directly. The Skill evaluates multi-step logic and architecture needs, automatically activating extended thinking with visual indicators like 🧠🔥 to deliver thorough root cause analysis for complex technical problems.

Yes, auto-scaling reasoning adjustment supports simple follow-up questions. The Skill automatically downgrades to fast responses after completing complex work, respecting user requests for quick answers to avoid unnecessary extended thinking and optimize token efficiency.

Weighted task complexity scoring is a mechanism that evaluates user requests on dimensions like multi-step logic, ambiguity, code architecture needs, and math requirements. It generates an accurate complexity score to determine if extended thinking should be activated for deep analysis.

You should not use extended thinking for simple lookup queries, straightforward questions, or follow-up tasks after complex work. The Skill automatically downgrades reasoning depth in these scenarios to provide instant responses and avoid wasting tokens on unnecessary deep analysis.