What problem does it solve? Teams default to one expensive frontier model for every task, overpaying for routine work while under-specifying genuinely hard jobs. This Skill replaces habit-based routing with a repeatable method: classify the work first, then assign it to a daily-driver, cheap-workhorse, or frontier tier with explicit escalation rules. ## Core Features & Use Cases - Three-tier routing with specialists: Classify jobs using seven questions (ambiguity, definable good, inspectability, sensitivity, modality, action, context location) and assign them to daily driver, cheap workhorse, or frontier tiers, attaching specialist capabilities for vision, live data, or tool use. - Eval protocols and benchmark discipline: Run 30-minute and one-week tests on a personal eval set of 3-5 real tasks, and interpret benchmark launches by per-task results and failure modes rather than leaderboard averages. - Frontier job briefing and launch triage: Spec expensive runs with a nine-field brief (outcome, source pack, tool access, boundaries, review standard, proof trail, human gate) and filter new model launches with a five-question adoption test. - Use Case: Your team pays frontier prices for weekly summary decks. Classify the job, run a one-week challenger test against a cheap model with a four-column log (time, rework, quality, would-send), then set a routing table with escalation triggers and present the evidence to management. ## Quick Start Ask the assistant to classify your recurring AI jobs with the seven questions and build a routing table assigning each job to a daily driver, cheap workhorse, or frontier tier.
