PluginBench
Skill
Pass
Audit score 90

reclaude

brianlovin/agent-config

Refactor CLAUDE.md files to follow progressive disclosure and best practices.

What is reclaude?

Reclaude analyzes and restructures CLAUDE.md configuration files to improve clarity and maintainability. Use it when your CLAUDE.md is too long, disorganized, or lacks critical sections like verification commands.

  • Audits CLAUDE.md length and flags files exceeding 100 lines for refactoring
  • Integrates workflow orchestration principles including plan mode, subagent strategy, and self-improvement loops
  • Ensures verification sections exist with actionable test/lint/build commands
  • Identifies contradictions and suggests resolutions between conflicting instructions
  • Extracts reusable content into global skills and organized `.claude/rules/` files
  • Removes redundant, vague, or inferrable content that clutters the configuration

How to install reclaude

npx skills add https://github.com/brianlovin/agent-config --skill reclaude
Claude Code
Cursor
Windsurf
Cline

How to use reclaude

  1. 1.Run the skill and provide your current CLAUDE.md file
  2. 2.Review the audit report showing line count and identified issues
  3. 3.Answer questions about contradictions and which instructions to keep
  4. 4.Accept suggestions for workflow integration and global skill extraction
  5. 5.Implement the refactored structure with root CLAUDE.md and organized `.claude/rules/` files

Use cases

Good for
  • Consolidating a sprawling CLAUDE.md that has grown to 200+ lines across multiple projects
  • Adding missing verification sections so Claude can validate its own work after changes
  • Extracting tool-specific instructions into reusable global skills shared across projects
  • Reorganizing project-specific rules into categorized `.claude/rules/` files for better navigation
  • Resolving conflicting instructions that confuse Claude's behavior
Who it's for
  • Developers maintaining multiple Claude projects with inconsistent configurations
  • Teams standardizing CLAUDE.md structure across codebases
  • Anyone using Claude Code or Cursor with complex project-specific instructions

reclaude FAQ

What's the ideal length for CLAUDE.md?

Under 50 lines is ideal; 50-100 lines is acceptable. Files over 100 lines should move detailed content to `.claude/rules/` subdirectory files.

Why is a Verification section critical?

Claude performs dramatically better when it can run commands to verify its own work. The skill helps identify test, lint, typecheck, and build scripts to include.

What should stay in the root CLAUDE.md?

Only essentials: one-line project description, non-obvious commands, links to `.claude/rules/` files, and the verification section. Everything else moves to organized rule files.

Can this create reusable global skills?

Yes. The skill identifies content that's tool/framework-specific and substantial enough (>20 lines) to become a reusable global skill in `~/.claude/skills/`.

What content gets removed entirely?

API documentation (link instead), code examples, type definitions, generic advice, obvious instructions, redundant info, and vague non-actionable guidance.

Full instructions (SKILL.md)

Source of truth, from brianlovin/agent-config.


name: reclaude description: Refactor CLAUDE.md files to follow progressive disclosure principles. Use when CLAUDE.md is too long or disorganized.

reclaude

Refactor CLAUDE.md files to follow progressive disclosure principles.

Prompt

I want you to refactor my CLAUDE.md file to follow progressive disclosure principles.

Follow these steps:

1. Check length

Report the current line count. Flag issues:

  • Ideal: <50 lines
  • Acceptable: 50-100 lines
  • Needs refactoring: >100 lines (move content to .claude/rules/ files)

2. Integrate workflow orchestration

Read the workflow skill at ~/.claude/skills/workflow/SKILL.md and incorporate its principles into the CLAUDE.md or a .claude/rules/workflow.md file. Adapt the content to fit the project — don't copy verbatim, but ensure the key behaviors are represented:

  • Plan mode for non-trivial tasks
  • Subagent strategy
  • Self-improvement loop with tasks/lessons.md
  • Verification before marking tasks done
  • Elegance checks for non-trivial changes
  • Autonomous bug fixing

For short CLAUDE.md files, add a concise workflow section. For longer ones, create .claude/rules/workflow.md and link to it.

3. Ensure verification section exists

Check for a ## Verification section with commands Claude can run after making changes. If missing:

  • Look in package.json for test/lint/typecheck/build scripts
  • Look for Makefile, justfile, or other task runners
  • Add a ## Verification section with discovered commands

This is critical—Claude performs dramatically better when it can verify its work.

4. Find contradictions

Identify any instructions that conflict with each other. For each contradiction, ask me which version I want to keep.

5. Check for global skill extraction candidates

Look for content that could become a reusable global skill in ~/.claude/skills/:

  • Is about a tool/framework (not project-specific)
  • Same instructions appear (or would apply) in 2+ projects
  • Is substantial (>20 lines)

If found, suggest creating a global skill with name and description.

6. Identify essentials for root CLAUDE.md

Extract only what belongs in the root CLAUDE.md:

  • One-line project description
  • Package manager (if not npm)
  • Non-obvious commands only (skip npm test, npm run build if standard)
  • Links to .claude/rules/ files with brief descriptions
  • Verification section (always required)

7. Group remaining content

Organize remaining instructions into .claude/rules/ files by category (e.g., TypeScript conventions, testing patterns, API design, Git workflow).

8. Flag for deletion

Identify content that should be removed entirely:

  • API documentation — link to external docs instead
  • Code examples — Claude can infer from reading source files
  • Interface/type definitions — these exist in the code
  • Generic advice — "write clean code", "follow best practices"
  • Obvious instructions — "use TypeScript for .ts files"
  • Redundant info — things Claude already knows
  • Too vague — instructions that aren't actionable

Target Template

# Project Name

One-line description.

## Commands
- `command` - what it does (only non-obvious ones)

## Rules
- [Topic](/.claude/rules/topic.md) — brief description

## Verification
After making changes:
- `npm test` - Run tests
- `npm run lint` - Check linting

What to Keep vs Remove

Keep in CLAUDE.md:

  • Commands Claude can't guess from package.json
  • Non-standard patterns specific to this project
  • Project gotchas and footguns
  • Links to detailed rules files

Move to .claude/rules/:

  • Detailed conventions (>10 lines on a topic)
  • Style guides
  • Architecture decisions
  • Workflow documentation

Remove entirely:

  • Anything Claude can infer from reading the codebase
  • Standard practices for the language/framework
  • Documentation that exists elsewhere (link instead)