Security audit
Security checks for vulnerabilities and agentic risk
This is a planning-checklist skill that may trigger too broadly, but it does not contain code, credential access, persistence, network behavior, or destructive instructions.
Install this if you want the assistant to pause and question plans before starting important work. Be aware that generic start words like "go" or "开始" may invoke the checklist when you expected immediate execution; narrower trigger wording would improve reliability.
The skill auto-activates on very broad phrases like “开始做”, “开始吧”, “go”, and “执行”, which are common in ordinary conversation and can be said without intent to invoke this skill. That creates unintended routing/control-flow hijacking risk: the assistant may enter a confirmation workflow when the user simply wants progress, delaying work or altering behavior without explicit consent.
The trigger-command section repeats the same broad automatic triggers, reinforcing that the skill should run on generic start words and after another skill step completes. This increases the chance of accidental or cascading activation across workflows, which can cause unwanted chaining into implementation, agent-teams, or refactoring paths without sufficiently explicit user intent.
Conditions like “multiple modules/team collaboration,” “clear deadline or quality requirements,” and “technical path unclear” are subjective and apply to many normal tasks. Because the boundaries are unclear, the skill can activate too often and override normal assistant behavior, creating reliability and prompt-routing issues that an attacker or even routine phrasing could exploit to force unnecessary workflow transitions.
No suspicious patterns detected.
