What problem does it solve? Teams routinely overspend on LLM calls by defaulting to frontier models for mechanical work, or underspend by forcing cheap models onto judgment-heavy tasks and paying twice for failures. This Skill provides a structured decision framework to pick the cheapest model tier that still clears the task's capability requirements. ## Core Features & Use Cases - Five-axis task classification: Scores tasks on specification clarity, reasoning depth, stakes, open-endedness, and context size to determine the right tier. - Tier-to-model mapping: Maps cheap/mid/frontier tiers to concrete Claude models (haiku, sonnet, opus) and generalizes to other provider families. - Application guidance: Shows how to pass the chosen tier to Claude Code subagents via the Agent tool's model parameter, or route runtime calls through a Manifest router at localhost:2099. - Use Case: Before fanning out 10 subagents for mechanical renames across a repo, classify each task and dispatch all 10 on haiku instead of opus, cutting cost without sacrificing correctness. ## Quick Start Ask the AI which model tier to use for your task, for example: which model should I use to rename a function across the repo and update all call sites?
