MCP HubMCP Hub
スキル一覧に戻る

context-review

majiayu000
更新日 Today
24 閲覧
58
9
58
GitHubで表示
ドキュメントwordai

について

コンテキストレビュースキルは、設定可能な標準に基づいてCLAUDE.mdファイルの構造的品質と効率性を監査します。行数の分析、適切な設計ドキュメントポインターの確認、トークン浪費を削減する最適化機会の特定を行います。コンテキスト最適化の準備時や、LLM指示が効率的で整理されていることを確認する際に、このスキルをご利用ください。

クイックインストール

Claude Code

推奨
プラグインコマンド推奨
/plugin add https://github.com/majiayu000/claude-skill-registry
Git クローン代替
git clone https://github.com/majiayu000/claude-skill-registry.git ~/.claude/skills/context-review

このコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします

ドキュメント

LLM Context Review

Reviews CLAUDE.md and child CLAUDE.md files to ensure they provide efficient, focused context for AI assistants without token waste.

Overview

This skill analyzes CLAUDE.md files against quality standards defined in .claude/design/design.config.json and the LLM Context Agent design document. It checks line counts, content structure, design doc pointers, and identifies opportunities for improvement.

Instructions

1. Locate Configuration

Read .claude/design/design.config.json to understand quality standards:

  • Root CLAUDE.md max lines (default: 500)
  • Child CLAUDE.md max lines (default: 300)
  • Required design doc pointers setting
  • Module structure for monorepos

2. Find CLAUDE.md Files

Search for CLAUDE.md files in the repository:

  • Root: CLAUDE.md or CLAUDE.local.md
  • Children: {module}/CLAUDE.md for each module
  • Monorepo packages: Check each package directory

3. Analyze Each File

For each CLAUDE.md file found:

Line Count Check:

  • Count total lines (excluding blank lines)
  • Compare against limits (root: 500, child: 300)
  • Flag files exceeding limits

Content Structure Check:

  • Verify high-level imperative instructions (not details)
  • Check for design doc pointers using @ syntax or links
  • Identify overly detailed sections (candidate for design docs)
  • Look for duplicated information between files

Design Doc Pointer Check:

  • Find references to design documentation
  • Verify @ syntax is used correctly: @./.claude/design/module/doc.md
  • Ensure pointers include guidance on when to load context
  • Check that docs being pointed to actually exist

Hierarchy Check:

  • Verify child CLAUDE.md files exist where appropriate
  • Check for modules with complex docs that should have children
  • Identify opportunities to split large root files

4. Generate Report

Create a comprehensive review report with:

Summary:

  • Total CLAUDE.md files found
  • Average line count
  • Files over limit count
  • Missing design doc pointers count

Issues by File:

Group by severity (high, medium, low):

  • High: Files exceeding line limits by >20%
  • High: Missing design doc pointers when design docs exist
  • High: Overly detailed sections that should be design docs
  • Medium: Files approaching line limits (>80%)
  • Medium: No child CLAUDE.md for complex modules
  • Low: Minor structure improvements

Recommendations:

Actionable suggestions:

  1. Which sections to move to design docs
  2. Where to create child CLAUDE.md files
  3. How to improve @ syntax pointer clarity
  4. What duplicated content to consolidate

Example Issue:

### High Priority

CLAUDE.md: 687 lines (limit: 500, 37% over)
- "Testing" section is very detailed (120 lines) → move to design doc
- "Architecture" duplicates content in pkgs/my-package/CLAUDE.md
- Missing pointer to `.claude/design/my-package/architecture.md`

Recommendation:
1. Create `.claude/design/project/testing-strategy.md` for testing details
2. Add pointer: "For testing details → @./.claude/design/project/testing-strategy.md"
3. Remove architecture duplication, keep only pointer to package CLAUDE.md

5. Quality Metrics

Calculate and report:

  • Efficiency Score: Percentage of files within limits
  • Pointer Coverage: Percentage of design docs with CLAUDE.md pointers
  • Hierarchy Health: Appropriate child files for module complexity
  • Duplication Level: Amount of repeated content across files

Output Format

Generate markdown report with structure:

# CLAUDE.md Review Report

**Date:** YYYY-MM-DD
**Repository:** {name}
**Type:** {monorepo|single-package}

## Summary

- CLAUDE.md files: X
- Average line count: Y
- Files over limit: Z
- Overall efficiency: N%

## Issues Found

### High Priority
[List of high-priority issues with file locations and recommendations]

### Medium Priority
[List of medium-priority issues]

### Low Priority
[List of low-priority issues]

## Recommendations

1. [Specific actionable recommendation]
2. [Specific actionable recommendation]
...

## Quality Metrics

- Efficiency Score: X%
- Pointer Coverage: Y%
- Hierarchy Health: Z%
- Duplication Level: N%

## Next Steps

[Suggested order of operations to address issues]

Special Cases

Monorepo:

Review both root CLAUDE.md and package-level files. Check for appropriate delegation between root and children.

No design docs yet:

If .claude/design/ doesn't exist, note that design documentation system should be set up first.

CLAUDE.local.md:

These override CLAUDE.md and should follow same standards. Review both if present, noting which takes precedence.

Success Criteria

A good review report:

  • Identifies specific line numbers for problematic sections
  • Provides concrete recommendations, not vague suggestions
  • Prioritizes issues by impact on context efficiency
  • Includes actionable next steps in recommended order

GitHub リポジトリ

majiayu000/claude-skill-registry
パス: skills/context-review

関連スキル

content-collections

メタ

This skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.

スキルを見る

evaluating-llms-harness

テスト

This Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.

スキルを見る

sglang

メタ

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

スキルを見る

langchain

メタ

LangChain is a framework for building LLM applications using agents, chains, and RAG pipelines. It supports multiple LLM providers, offers 500+ integrations, and includes features like tool calling and memory management. Use it for rapid prototyping and deploying production systems like chatbots, autonomous agents, and question-answering services.

スキルを見る