agenthub 审计报告.

警告

• No explicit sandboxing or security restrictions are described for the spawned subagents—they execute commands in worktrees, which could be risky if not scoped carefully.
• The SKILL.md references external scripts (e.g., `dag_analyzer.py`, `references/agent-templates.md`) that are not provided in this excerpt; their safety and correctness cannot be verified from this submission alone.
• No guidance is given on limiting resource usage (e.g., maximum number of agents) or handling malicious task inputs that could lead to unintended actions.
• Quality score needs review

方法

本报告综合公开元数据、AI 审查输出、仓库活跃度、安装就绪度、OpenAgentSkill 事件、质量评分、信任检查和 Agent 安全门槛；它不是完整的源代码安全审计。