What problem does it solve? AI agents default to whichever LLM was best at their training cutoff, silently missing newer, cheaper, or better models released since. This Skill forces a 5-10 minute fresh research pass before any code change that adds or modifies an external-LLM API call. ## Core Features & Use Cases - Live registry refresh: Pulls current models and pricing from the OpenRouter models endpoint and mechanically selects the latest releases per vendor family instead of relying on memory. - Task-type benchmark mapping: Classifies the task (agentic coding, long context, multilingual, etc.) and looks up the benchmarks most predictive for it, with per-locale slicing rules for multi-language deployments. - Channel and tier cost analysis: Enumerates direct API, OpenRouter, BYOK credit pools, and batch/flex service tiers, computing gross-cost-equivalents so "free" channels don't silently exhaust credit pools. - Use Case: Before writing a new script that calls the Gemini API, run the pick flow to discover a model released last week offers equivalent quality at half the price, then log the verdict to a JSONL pick log. ## Quick Start Before adding any external LLM call to my code, run the model-pick flow to refresh the model registry, compare 3-5 candidates across cost tiers, and log the chosen model with rationale.
