modelsense_skill

Overview

This skill recommends the best LLM model and effort level for any task, using benchmark data, task analysis, and the user's configured providers. It provides clear, actionable recommendations and alternatives, plus cost and capability notes so you can pick or apply the suggestion immediately.

How this skill works

I analyze the user's task across domain, complexity, output type, context length, and any special requirements. I match that analysis to benchmark leaders and an effort×model matrix, filter recommendations by the user's available providers, and present a concise recommendation with rationale, cost estimate, and alternatives. If requested, I can also apply the recommendation to the active session or spawn a task run on the chosen model.

When to use it

• When you ask “which model should I use?” or “what’s best for X?”
• When choosing trade-offs between speed, cost, and quality for a task
• When interpreting benchmark results and mapping them to your real task
• Before switching a session model or delegating a task to another model
• When comparing models across providers you have configured

Best practices

• Provide a concise task description and expected output format to improve accuracy
• Specify effort preference (quick / balanced / deep / research) or let it be inferred
• Share which providers/models you already have access to for realistic recommendations
• Ask for alternatives explicitly if you want lower cost or higher quality options
• Confirm before auto-switching the session or delegating execution to another model

Example use cases

• Choosing a model for a security-focused Solidity audit report (long-form, code + reasoning)
• Selecting a fast, low-cost model for summarizing Slack threads (short dialogue summaries)
• Picking a research-grade model for mathematical proofs or graduate-level analysis
• Deciding between Sonnet and Opus for complex multi-step reasoning with function calling
• Filtering recommendations to only models available from your configured providers