compare

To compare trigger phrase generation across LLM model sizes, you run a controlled experiment on a source file to study agreement, divergence, and shared terms across haiku, sonnet, and opus models. The Skill spawns study agents and logs cross-model results in a structured JSON file for manual analysis.

Cross-model divergence analysis identifies how different model sizes generate varying trigger phrases from the same source file. It works by spawning multiple study agents to process the file, applying confidence scoring, and logging comparative results into structured JSON output for review.

Yes, you can use a single file or a small batch of repository files for controlled LLM comparison experiments. This allows you to study model-specific wording and agreement across haiku, sonnet, and opus while logging relative-path results into a structured JSON file.

You log LLM agent workflow results to JSON for manual analysis by running a cross-model comparison task on your source files. The process requires agent spawning and automatically stores trigger phrase generation outputs, confidence scores, and relative-path results in a structured JSON file.

The limitation of running cross-model comparison is that it is designed for controlled experiments on a single file or a small set of repository files. It requires agent spawning to study haiku, sonnet, and opus divergence, making it unsuitable for processing large-scale codebases or unstructured data batches.