prepare-release 审计报告.

警告

• The skill directly executes `pnpm build:builtin-knowledge` and creates git branches/PRs without explicit user confirmation beyond the initial request. This could lead to unintended side effects if the agent is misused or the user input is misinterpreted.
• There is no explicit handling of prompt injection from commit messages or other repository content. While the skill only extracts `release-note` blocks, an attacker could craft commit messages to inject instructions into the release-note generation step (though the LLM is expected to follow the skill's rules).

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。