Security audit
Security checks for vulnerabilities and agentic risk
This skill provides AI governance guidance and simple offline tools, with minor limitations in its bundled audit and reporting.
Suitable as an offline governance reference, not a security certification or enforcement system. Optional verification replaces bundled report files. Its self-audit, scores, and outdated manifest hash do not establish independent assurance; verify substantive governance claims before relying on them. This review inspected the supplied files without executing the scripts or authenticating external publisher sources.
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).
- 编码智能体专用沙箱（禁网络/禁写敏感路径）
### 2.3 越权与破坏面
- Agent 自主执行 `rm`、`git push --force`、改生产配置
- 超出仓库范围读写其他目录
- 缓释：最小权限工作区 + 命令白名单 + 高危操作人工确认（两阶段提交）
Without declared permissions the skill's intent is opaque and cannot be validated.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run_tool(args, timeout=30):
try:
r = subprocess.run([PY, os.path.join(TOOLS, "aibuild_toolkit.py")] + args,
capture_output=True, text=True, timeout=timeout)
return r.returncode, r.stdout or ""
except Exception as e:
Opening the fixed output path in write mode truncates any existing results file. The module docstring describes local verification and the console output reports scores, but neither discloses that a persistent report will be created or replaced.