reclaude_skill

Overview

This skill refactors CLAUDE.md files to follow progressive disclosure principles so documentation is concise, discoverable, and task-focused. It converts long or disorganized CLAUDE.md content into a minimal root file plus linked topic files and a suggested docs/ folder structure. The goal is faster onboarding for agents and humans by exposing only essentials up front.

How this skill works

The skill scans the CLAUDE.md for contradictions, essential root-level items, and other instructions. It extracts a one-line project description, package manager, non-standard build/typecheck commands, and universally relevant rules into a minimal root file. Remaining content is grouped into logical topical files, flagged for deletion if redundant, vague, or obvious, and a docs/ folder layout is suggested.

When to use it

• CLAUDE.md is long, repetitive, or hard to scan
• Onboarding new agents or contributors quickly
• When CLAUDE.md mixes global rules with task-specific guidance
• Preparing a repository for machine-reading or agent automation

Best practices

• Keep the root CLAUDE.md to essentials only: one-sentence description, package manager, and core commands
• Group related guidance into focused topic files linked from the root
• Resolve contradictions before reorganization and ask the author when needed
• Flag and remove vague or obvious rules; keep actionable specifics
• Use clear filenames and a predictable docs/ folder structure for machine access

Example use cases

• Refactor a monolithic CLAUDE.md into root + docs/type-system.md + docs/testing.md
• Clean up contradictory build instructions and ask which command to keep
• Strip out generic advice like 'write clean code' and move style rules to docs/style.md
• Generate a minimal root file for an automated agent to quickly find build commands