What problem does it solve? Running large-context analysis tasks like log audits or content moderation entirely on expensive senior models is costly, while cheap models alone make poor judgments on ambiguous cases. This Skill implements the OpenRouter Advisor pattern so a fast primary model processes bulk context and escalates only borderline decisions to a senior model mid-inference. ## Core Features & Use Cases - Mid-Inference Delegation: Configure the openrouter:advisor tool so a primary model (e.g. deepseek-v4-flash) consults a senior advisor (e.g. claude-opus) for targeted judgments within a single API call. - Robust Response Parsing: Handle NDJSON and chunked streamed responses with fallback line-by-line JSON parsing, retry logic, and debug dumps for truncated bodies. - Cost Tracking & Routing: Log advisor call costs to a spend JSONL file and control provider routing with order, sort, latency, throughput, and price parameters. - Use Case: Audit 80K characters of application logs at Flash-tier prices, escalating only ambiguous error classifications to Opus, cutting costs by over 95% while keeping senior-model accuracy on hard cases. ## Quick Start Use the OpenRouter advisor pattern to analyze these logs with a cheap primary model and escalate ambiguous error classifications to a senior advisor model.
