classify-stats

You can classify agent performance statistics by running a classifier over YAML files in the .workflows/stats/ directory, which proposes complexity and output_type labels based on agent actions and artifact content before requiring user confirmation to update files in place.

Labeling agent task complexity and output type involves assigning categories like rote, mechanical, analytical, or judgment for complexity, and code-edit, research, or review for output type, generating a structured dataset for performance analysis.

Yes, you can batch process multiple unclassified stats entries found in .workflows/stats/*.yaml. The Skill handles entries efficiently and presents classification proposals together for user confirmation before applying updates.

Yes, classify-stats requires manual confirmation before updating files in place. It dispatches a classifier subagent to propose complexity and output_type labels, then waits for user approval to write changes to the YAML files.

The classifier determines labels by analyzing agent actions and artifact content from the statistics entries. It uses this data to propose a complexity level, such as rote or analytical, and an output type, such as code-edit or research.

You need to classify agent stats entries when you have unclassified performance data in .workflows/stats/ YAML files and want to structure it by task complexity and output type for easier workflow analysis and performance evaluation.