Multi-LLM Tiered Effort Router Skill

Overview

Dispatches agentic tasks across a multi-provider LLM cascade based on task complexity. Ensures trivial tasks (syntax linter, formatting, basic unit tests) do not burn expensive frontier reasoning tokens, while critical tasks (architecture design, security boundary verification) receive maximum reasoning depth.

Router Tier Hierarchy

Usage Pattern

import { resolveEffortRoute, executeCascadeRequest } from './router';

async function dispatchAgentWork(taskName: string, complexity: 'MICRO' | 'STANDARD' | 'COMPLEX' | 'CRITICAL', prompt: string) {
  const route = resolveEffortRoute(complexity);
  return await executeCascadeRequest(route, prompt);
}

Key Invariants

1. Zero-Cost Preference: Always attempt free-tier or locally hosted providers first for Tiers 1 and 2.
2. Quota Isolation: If a provider returns HTTP 429 or quota exhaustion, trigger that specific provider's circuit breaker without blocking alternative pool providers.
3. Audit Trail: Log model used, tokens consumed, and latency profile for every dispatch.