Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed research-and-logging helper for AI agent papers, with some broad activation wording users should control intentionally.
Install this only if you want an agent to read project context and maintain a local AI-agent research log. Invoke it intentionally in the projects where that logging is desired, and review research/harness-log.md if project documents may contain sensitive information.
The trigger phrases include very broad, everyday terms such as '研究论文', '找最新文章', '读paper', and '分析论文', which can cause the skill to activate during ordinary research requests unrelated to this specific harness workflow. In an agent environment, overbroad activation can unexpectedly cause web searches, file reads, and log writes, increasing the risk of unintended tool use and context leakage.
The claim that the skill is suitable for '任何运行中的 AI Agent 系统' without clear activation boundaries makes its scope effectively universal, encouraging invocation across unrelated projects and contexts. Because the skill instructs the agent to inspect local project documents and write persistent logs, ambiguous scope raises the chance of unintended cross-context behavior and unauthorized modification of workspace files.
