What problem does it solve? Running every step of a task on an expensive frontier model wastes cost, but running everything on a cheap model risks flawed architecture decisions, missed constraints, and premature "done" verdicts. This Skill defines when and how to insert a stronger-model review at the few load-bearing moments of a task while a faster model handles the mechanical work. ## Core Features & Use Cases - Checkpoint Timing Guidance: Identifies the three moments worth a consult — before committing to a non-obvious approach, when stuck, and before declaring a non-trivial task done — and when to skip it. - Interactive Agent Dispatch: Instructs dispatching an Agent call with a stronger model tier (opus or fable), a concrete plan or diff, and a specific question rather than a generic review request. - Workflow Script Pattern: Provides a propose → advise → refine JavaScript pattern using per-call model overrides on agent() so bulk stages run cheap and only decision points use the strong model. - Use Case: A workflow drafts an implementation plan on a fast model, passes the draft to an opus-tier agent to flag flaws or missed constraints, then resumes implementation on the fast model informed by the review. ## Quick Start Ask the AI to review the current approach or diff with a stronger model before proceeding with implementation.
