agent-md-refactor

You can refactor a bloated agent instruction file by applying progressive disclosure principles to split it into a minimal root file and categorized, linked sub-files. This structure optimizes token usage and improves AI context management.

Progressive disclosure for agent configuration files is the practice of breaking down monolithic instruction documents into a minimal root file with links to categorized sub-files. This technique reduces context token waste and enhances instruction clarity for AI agents.

Identifying contradictions in AGENTS.md instructions requires systematic analysis of the file content to detect conflicting rules. The refactoring process flags these conflicts and prompts the user for resolution before restructuring the documentation hierarchy.

Yes, the refactoring approach specifically prunes project-level configuration files by identifying and flagging vague, redundant, or outdated instructions for deletion. This optimizes token usage and ensures only essential rules remain in the documentation.

Agent-md-refactor operates independently without requiring external dependencies or frameworks. It functions as a standalone skill to restructure project-level configuration files like CLAUDE.md or AGENTS.md by analyzing existing instruction content.

The best way to clean up unmanageable agent instruction files is to systematically analyze the content for contradictions, essential rules, and redundant information. This allows you to prune outdated instructions and restructure the remaining content into a linked documentation hierarchy.