What problem does it solve? Single-model passes break down on tasks too large for one context window or one generation call, and naive multi-model loops fail from context leaks, silent partial failures, and judgment applied too late. This Skill turns your coding agent into the orchestrator of a three-tier model team that splits big tasks into self-contained briefs, runs them across cheap parallel workers, verifies every result, and consults a stronger advisor model exactly twice: before work starts and before delivery. ## Core Features & Use Cases - Three-tier model team: an orchestrator plans and verifies, stateless workers (default Gemini 3.7 Flash via the agy CLI) execute one subtask each in parallel, and an advisor (default Claude Fable 5) reviews the plan and the final draft without ever executing. - Stateless worker briefs with verification: every dispatch carries full inline inputs and acceptance criteria, and each result is judged PASS, FIX (redispatched naming the failure), or ESCALATE, with no silent partial passes. - Budgets and fallbacks: every run states a budget up front, and missing CLIs fall back to the Gemini or Anthropic APIs, with a clearly labeled degraded mode when a role has no working path. - Use Case: Ask the agent to research 12 competitors in parallel: it frames success criteria, gets the plan reviewed by the advisor, fans out 12 worker briefs in batches, verifies each report, and delivers a synthesized comparison with a verification ledger. ## Quick Start Ask your agent to orchestrate this task across a model team because it is too big for one pass, fanning the subtasks out to parallel workers with advisor review.
