skills/claude-dir-governance/SKILL.md
.claude 目录配置治理和性能优化规范
npx skillsauth add OliverOuyang/shuhe-work-skills claude-dir-governanceInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
.claude 目录配置治理框架,确保配置规范和性能优化。
此 skill 提供 .claude 目录的治理规范,包括 CLAUDE.md、MCP 服务器、Skills、Hooks 的变更管理流程,以及性能优化指导和自动化验证工具。
参考 skill-governance 的治理模式,适配用户级 Claude Code 配置管理。
当执行以下操作时自动触发:
~/.claude/CLAUDE.md~/.claude/mcp.json~/.claude/skills/ 下新增/修改/删除 skill~/.claude/rules/ 下的规则文件~/.claude/settings.json 中的 hooks 配置# 验证当前配置
/claude-dir-governance --validate
# 查看治理规则
/claude-dir-governance --rules
# 检查性能指标
/claude-dir-governance --performance
~/.claude/CLAUDE.md 内容~/.claude/rules/*.md 文件rules/ 目录(单个主题 > 20 行)rules/report-workflow.md)详见 .claude/rules/<filename>.md# 检查所有交叉引用是否有效
grep -r "详见.*\.md" ~/.claude/CLAUDE.md | while read line; do
file=$(echo "$line" | grep -oP '\.claude/rules/\K[^)]+')
[ -f ~/.claude/rules/"$file" ] || echo "❌ 缺失: $file"
done
mcp.json 中新增/修改/删除 MCP 服务器配置步骤 1:需求确认
步骤 2:配置规范
{
"mcpServers": {
"server-name": {
"command": "command",
"args": ["arg1", "arg2"],
"env": {
"ENV_VAR": "value"
},
"disabled": false,
"_comment": "用途说明和使用频率"
}
}
}
步骤 3:性能评估
"disabled": truesettings.json 中配置)步骤 4:测试验证
# 重启 Claude Code 后测试 MCP 连接
# 执行一次实际调用验证功能正常
评估影响:
操作规范:
"disabled": true 而非直接删除_comment 中记录禁用原因和日期~/.claude/skills/ 下新增/修改/删除 skill 目录步骤 1:结构创建
~/.claude/skills/<skill-name>/
├── SKILL.md # 必需:skill 文档
├── scripts/ # 可选:执行脚本
└── resources/ # 可选:资源文件
步骤 2:SKILL.md 规范
---
name: skill-name
description: 简短描述(一句话)
version: 1.0.0
---
# Skill 名称
## Description
详细功能说明
## Usage
/skill-name [参数]
## Examples
示例用法
步骤 3:命名规范
my-skill)/skill-name(带斜杠前缀)SKILL.md(全大写)变更评估:
影响检查:
~/.claude/settings.json 中的 hooks 配置安全检查:
配置规范:
{
"hooks": {
"PreToolUse:ToolName": {
"command": "command",
"timeout": 1000,
"description": "Hook 用途说明"
}
}
}
验证步骤:
rules/创建 ~/.claude/scripts/validate-config.sh:
#!/bin/bash
# .claude 目录配置验证脚本
CLAUDE_DIR="${HOME}/.claude"
echo "=== .claude 配置验证 ==="
# 1. 检查 CLAUDE.md 交叉引用
echo "1. 检查 CLAUDE.md 交叉引用..."
grep -r "详见.*\.md" "$CLAUDE_DIR/CLAUDE.md" | while read line; do
file=$(echo "$line" | grep -oP '\.claude/rules/\K[^)` ]+')
[ -f "$CLAUDE_DIR/rules/$file" ] || echo "❌ 缺失: $file"
done
# 2. 检查 MCP 配置语法
echo "2. 检查 MCP 配置..."
if command -v jq &> /dev/null; then
jq empty "$CLAUDE_DIR/mcp.json" && echo "✓ MCP 配置语法正确"
fi
# 3. 检查 skills 目录结构
echo "3. 检查 skills 目录..."
for skill_dir in "$CLAUDE_DIR/skills"/*/; do
[ -f "$skill_dir/SKILL.md" ] || echo "❌ $(basename $skill_dir): 缺少 SKILL.md"
done
# 4. 性能检查
echo "4. 性能检查..."
lines=$(wc -l < "$CLAUDE_DIR/CLAUDE.md")
[ $lines -le 200 ] && echo "✓ CLAUDE.md 行数: $lines (≤ 200)" || echo "⚠ CLAUDE.md 行数: $lines (建议 ≤ 200)"
echo "=== 验证完成 ==="
# 备份 CLAUDE.md
cp ~/.claude/CLAUDE.md ~/.claude/CLAUDE.md.backup.$(date +%Y%m%d_%H%M%S)
# 备份 mcp.json
cp ~/.claude/mcp.json ~/.claude/mcp.json.backup.$(date +%Y%m%d_%H%M%S)
# 恢复最近的备份
cp ~/.claude/CLAUDE.md.backup.YYYYMMDD_HHMMSS ~/.claude/CLAUDE.md
/claude-dir-governance --validate
输出:
=== .claude 配置验证 ===
✓ 所有交叉引用有效
✓ MCP 配置语法正确
✓ 所有 skills 结构正确
✓ CLAUDE.md 行数: 146 (≤ 200)
=== 验证完成 ===
/claude-dir-governance --performance
输出:
性能指标:
- CLAUDE.md: 146 行 (目标 < 200)
- 启用的 MCP: 4 个
- 预估上下文: ~35K tokens (目标 < 40K)
| 问题 | 原因 | 解决方案 |
|------|------|---------|
| 交叉引用失效 | rules 文件被删除或重命名 | 更新 CLAUDE.md 中的引用路径 |
| MCP 配置语法错误 | JSON 格式不正确 | 使用 jq 验证语法 |
| Skills 缺少 SKILL.md | 创建目录���未添加文档 | 补充 SKILL.md 文件 |
| 性能下降 | CLAUDE.md 过长或 MCP 过多 | 拆分规则文件,禁用低频 MCP |
tools
SQL 分段验证、自我修复、结果导出与智能分析。流程:解析SQL → Dataphin MCP 验证元数据 → 自动修复 → 分段执行验证 → 导出 CSV → 智能分析(漏斗解读、异常识别、预判用户问题)。适用场景:"跑一下这个SQL"、"验证这个查询"、"帮我执行并导出"、"分析一下结果"等。
testing
Security-first vetting for OpenClaw skills. Use before installing any skill from ClawHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
development
A universal self-improving agent that learns from ALL skill experiences. Uses multi-memory architecture (semantic + episodic + working) to continuously evolve the codebase. Auto-triggers on skill completion/error with hooks-based self-correction.
data-ai
Standardize Jupyter notebooks (.ipynb) for interactive data analysis workflows. Enforces a mandatory cell manifest (M1-M8 + archetype chapters) with tags ([CONFIG]/[SETUP]/[FUNC]/[RUN]/[VIZ]/[EXPORT]), structured markdown sections, and output prefixes ([OK]/[WARN]/[SKIP]). Use when the user wants to standardize, clean up, or create a notebook from scratch. Two archetypes: problem-driven (question-answer analysis) and monitoring (dimension-based periodic reporting).