Claude Code Connector

通过 ACP 桥接调用 Claude Code。 **触发词：** "调用 Claude"、"让 Claude 处理"、"claude code" **使用方式：** ``` 调用 Claude Code 分析这个文件

Security Scan

The skill claims to 'call Claude Code via ACP' but its runtime instructions execute a hard-coded local script path (/Users/mac/.openclaw/workspace/scripts/acp-bridge-claude.sh) that is not provided or described—this could run arbitrary code and access local data, so the skill's behavior is plausible but not sufficiently constrained or transparent.

Purpose & Capability

The declared purpose (bridge to Claude Code) matches the instruction to call an ACP bridge script, so the capability aligns in concept. However, the skill assumes the presence of a specific user-local script that is not part of the package and not documented, which is an unexpected external dependency for an instruction-only skill.

Instruction Scope

The SKILL.md directs the agent to execute a specific absolute path on the user's machine. While the instruction does not explicitly tell the agent to read other files, executing an arbitrary local script can perform any actions (read files, network calls, exfiltrate data). The instructions provide no limits, verification steps, or description of what the script does.

Install Mechanism

No install spec (instruction-only), which reduces disk footprint from the skill itself. But relying on an external local script (not installed or bundled) is unusual and shifts trust to an out-of-band file; this is not inherently dangerous but is opaque.

Credentials

The skill declares no required environment variables or credentials, yet invoking an arbitrary local bridge script could access credentials or secrets on the host. Because the script is external and not audited, the skill's declared environment requirements understate what the executed code may access.

Persistence & Privilege

The skill does not request always:true and is user-invocable only. It does not claim to modify other skills or system-wide agent settings.

What to consider before installing

This skill runs a specific local script path that is not provided with the skill. Only enable it if you trust and have inspected /Users/mac/.openclaw/workspace/scripts/acp-bridge-claude.sh (open it and review its contents). Prefer skills that either bundle or link to a known implementation or clearly document what network endpoints and file access are used. If you need to test, run the script in a sandbox/container or request the script source from the author. If you cannot review the script, treat the skill as risky because it can execute arbitrary code and access local data or credentials.

Call Claude Code via ACP

工具定义

{
  "name": "call_claude_code",
  "description": "Call Claude Code via ACP bridge",
  "inputSchema": {
    "type": "object",
    "properties": {
      "prompt": {
        "type": "string",
        "description": "The prompt to send to Claude Code"
      }
    },
    "required": ["prompt"]
  }
}

执行逻辑

/Users/mac/.openclaw/workspace/scripts/acp-bridge-claude.sh

示例

输入：

{
  "prompt": "分析当前目录的项目结构"
}

输出：

Claude Code 的分析结果...