What problem does it solve? Hardcoding a specific LLM provider into every text-generating step makes pipelines brittle and expensive to maintain. This Skill centralizes provider selection so any task (caption, script, compliance, humanization) is routed to the right model based on a configuration matrix, with automatic fallback when a provider fails. ## Core Features & Use Cases - Task-based routing: Maps task types like caption, script, compliance, and humanization to default and fallback providers using a PROVIDERS.md matrix and .env keys. - Override support: Honors provider_override arguments and per-piece overrides declared in .specs/pieces/*.md files. - Fallback and logging: Retries with a fallback provider on network or quota errors and appends usage records (tokens, cost, outcome) to data/llm-usage.jsonl. - Use Case: A caption-generation skill calls the router with task: caption; the router resolves the configured provider (e.g., deepseek), verifies its API key exists, executes the call, and logs token usage and estimated cost. ## Quick Start Ask the agent to route a caption task with the prompt "Hook for IG reel about color analysis" through the LLM router and return the generated text with the provider used.
