Claude Code Controller

控制和管理 Claude Code 编码助手，支持项目感知编码、代码审查、重构和功能实现。使用 ACP 运行时在隔离会话中执行 Claude Code 任务，或在主会话中管理配置和项目上下文。

Security Scan

The skill's instructions and files match a Claude Code controller, but the package metadata omits required environment/config declarations and it references home/system paths and example connectors that raise proportionality and scope questions.

Purpose & Capability

The skill's name, description, SKILL.md, and the provided launch script consistently implement a controller for Anthropic Claude Code (installing/using the claude CLI, configuring ~/.claude, running tasks). However the metadata claims no required env vars or config paths while the runtime docs and script clearly require ANTHROPIC_API_KEY and rely on ~/.claude files — this discrepancy is unexpected and should be justified.

Instruction Scope

SKILL.md instructs the agent/user to install and run the claude CLI, create and edit project-level ~/.claude/settings.json, read logs (~/.claude/logs/latest.log) and run claude in interactive/ACP sessions. Those actions are consistent with a controller, but they grant the skill access to user home config, logs, and (via MCP examples) potentially databases or filesystem bridges. The instructions also reference absolute/root-like paths in docs (e.g., /root/.openclaw/workspace) which implies assumptions about environment privileges.

Install Mechanism

There is no automatic install spec in the skill bundle (instruction-only). SKILL.md recommends installing @anthropic-ai/claude-code via npm (a normal public registry package). The included shell script is simple and does not download arbitrary archives or external payloads. No high-risk download URLs or extract steps are present.

Credentials

The metadata declares no required environment variables, yet SKILL.md and scripts require ANTHROPIC_API_KEY and reference other env examples (ANTHROPIC_MODEL, DATABASE_URL in MCP examples). This mismatch is disproportionate and could lead to implicit credential use. The skill also expects to read/write ~/.claude config and logs — access that should be explicitly declared and limited.

Persistence & Privilege

The skill does not request always:true and does not include install-time mechanisms that modify other skills. The launch script writes a project-local .claude/settings.json if missing; that is normal for a tooling helper. No evidence the skill attempts to change system-wide agent config or persist credentials beyond normal config files.

What to consider before installing

Before installing or enabling this skill: - Expect it to require an Anthropic API key (ANTHROPIC_API_KEY) and to read/write ~/.claude/* files; the metadata currently does not declare these — treat that as a red flag and confirm with the author. - Inspect and if needed sandbox the included scripts (scripts/launch-claudecode.sh) before running; it will create project/.claude/settings.json and invoke the claude CLI. - Only install @anthropic-ai/claude-code from the official registry and verify package provenance. - Review any MCP/database examples carefully: configuration can launch helper processes (npx ...) that may access databases or filesystem paths — don’t provide DATABASE_URL or broad ALLOWED_PATHS unless you trust the skill. - Avoid placing sensitive keys in project files; keep ANTHROPIC_API_KEY in a secure secret store and verify the skill does not leak it to unexpected endpoints. - If you need higher assurance, run the skill in an isolated environment (container or VM) and require the author/source to provide signed or repository-traceable code and a declared list of required env vars/config paths.

本技能提供对 Anthropic Claude Code 的完整控制，让你能够在 OpenClaw 环境中高效使用 Claude 进行代码开发。

核心能力

1. 项目感知编码 - Claude Code 理解你的项目结构和上下文
2. 代码审查 - 自动审查 PR、提供改进建议
3. 重构辅助 - 安全地重构现有代码
4. 功能实现 - 从需求描述到完整实现
5. 调试帮助 - 定位和修复 bug
6. 文档生成 - 自动生成代码文档

使用方式

模式选择

模式	用途	命令
ACP 隔离会话	长时间编码任务、需要独立上下文	/claudecode <任务描述>
主会话管理	配置管理、快速查询、项目设置	/claudecode-config <操作>

基本命令

# 启动 Claude Code 进行编码任务
/claudecode 帮我实现一个用户登录功能，使用 JWT 认证

# 代码审查
/claudecode 审查这个 PR 的代码质量，检查潜在问题

# 重构代码
/claudecode 重构这个模块，提高可维护性和性能

# 调试帮助
/claudecode 帮我找出这个 bug 的原因并修复

配置管理

设置 Claude Code

1. 安装 Claude Code CLI (如未安装):

npm install -g @anthropic-ai/claude-code

2. 配置 API 密钥:

# 设置环境变量
export ANTHROPIC_API_KEY="your-api-key"

# 或在 ~/.claude/config.json 中配置

3. 验证安装:

claude --version

项目配置

在项目根目录创建 .claude/settings.json:

{
  "model": "claude-sonnet-4-5-20250929",
  "allowedTools": ["bash", "edit", "write", "read"],
  "maxTurns": 50,
  "permissionMode": "auto"
}

工作流程

1. 代码开发流程

需求分析 → 方案设计 → 代码实现 → 测试验证 → 代码审查

示例对话:

用户：帮我创建一个 REST API，包含用户 CRUD 操作
Claude: 我将使用 Express.js 创建 API，包含以下端点：
  - POST /users - 创建用户
  - GET /users/:id - 获取用户
  - PUT /users/:id - 更新用户
  - DELETE /users/:id - 删除用户
  开始实现...

2. 代码审查流程

代码加载 → 静态分析 → 问题识别 → 建议生成 → 修复实施

审查要点:

• 代码风格和一致性
• 潜在 bug 和边界情况
• 性能优化机会
• 安全漏洞检查
• 测试覆盖率

3. 重构流程

代码分析 → 识别改进点 → 制定重构计划 → 安全重构 → 验证测试

常见重构:

• 提取函数/类
• 简化条件逻辑
• 优化数据结构
• 改进命名和文档

最佳实践

提示词技巧

1. 具体明确: 清晰描述需求和约束
2. 提供上下文: 说明项目背景和技术栈
3. 分步执行: 复杂任务分解为小步骤
4. 验证输出: 审查生成的代码是否符合预期

安全注意事项

1. 权限控制: 限制 Claude Code 的文件访问范围
2. 代码审查: 始终审查生成的代码后再提交
3. 敏感信息: 不要将 API 密钥等敏感信息放入代码
4. 版本控制: 使用 git 跟踪所有更改

故障排除

常见问题

问题	解决方案
API 密钥错误	检查 ANTHROPIC_API_KEY 环境变量
命令未找到	确认已安装 @anthropic-ai/claude-code
权限被拒绝	检查 .claude/settings.json 中的 allowedTools
超时错误	增加 maxTurns 配置或使用更小的任务

日志和调试

# 查看 Claude Code 日志
cat ~/.claude/logs/latest.log

# 调试模式运行
claude --debug

高级功能

多模型支持

{
  "models": {
    "default": "claude-sonnet-4-5-20250929",
    "complex": "claude-opus-4-5-20250929",
    "quick": "claude-haiku-4-5-20250929"
  }
}

自定义工具

创建自定义工具脚本放在 ~/.claude/tools/:

#!/bin/bash
# ~/.claude/tools/run-tests.sh
npm test "$@"

项目模板

使用预定义模板快速启动项目:

claude --template express-api
claude --template react-app
claude --template python-package

与 OpenClaw 集成

ACP 会话使用

当使用 ACP 运行时，Claude Code 将在隔离会话中执行:

用户请求 → OpenClaw → ACP 会话 → Claude Code → 结果返回

优势:

• 独立上下文，不污染主会话
• 可配置超时和模型
• 支持长时间运行任务

主会话管理

在主会话中管理配置和项目设置:

• 查看和修改配置
• 管理项目上下文
• 查看历史记录
• 快速查询和统计

参考资源

• Claude Code 官方文档
• OpenClaw ACP 文档
• 技能开发指南