semantic-router_skill

Overview

This skill is an auto-routing semantic router that selects model pools based on task type and conversational continuity. It provides an interactive setup wizard to detect available models and customize pools. It includes two-step detection for semantic continuity and task-to-pool mapping, plus a fallback chain for model failures.

How this skill works

The skill inspects each incoming message for semantic continuity (is this a continuation of the current conversation?) and then classifies task type using keyword matrices and configurable task definitions. Based on classification it selects a pool (Highspeed, Intelligence, Humanities) and returns a primary model plus an ordered fallback chain. It can run as a message injector to force checks and has automated fallback and periodic rollback to the primary model.

When to use it

• Route requests across heterogeneous models to optimize cost/latency/quality tradeoffs
• Automatically pick appropriate model for a given user intent or task type
• Maintain conversational context and avoid unnecessary model switches
• Enforce a structured fallback chain for robust agent operation
• Run an initial setup to detect available models and generate suggested pools

Best practices

• Run the interactive setup wizard to auto-detect your models and accept AI suggestions before customizing pools
• Keep task keyword lists concise and representative of common intents for reliable routing
• Test fallback chains by simulating model failures (rate limits, timeouts) and review logs
• Enable message injector for consistent enforcement if you need every message inspected
• Monitor memory/model-fallback.log and configure rollback intervals to fit your reliability requirements

Example use cases

• A user asks to search the web — router selects the Highspeed pool and a low-latency retrieval model
• A developer requests code help — router routes to the Intelligence pool and a code-specialized model
• A content creator asks for a long-form article — router uses the Humanities pool and a content model
• A chat session continues with 'continue' or pronouns — router detects semantic continuity and preserves the current pool
• Simulate a primary model outage to confirm fallback to Fallback 1 and Fallback 2 automatically