Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only coaching skill that encourages persistence and broader investigation, with no code, persistence, credential access, or hidden execution.
Install this if you want the agent to be more persistent and proactive when stuck. Be aware that it nudges the agent to investigate and try alternatives before asking, so for tightly scoped or sensitive tasks you should state explicit limits in your request.
The skill declares that it applies to 'all task types' and activates under broad, subjective conditions like being 'stuck' or 'about to give up.' That can cause the skill to override normal task scoping and push the agent into behavior that is inappropriate for sensitive, regulated, or user-bounded contexts, increasing the chance of unsafe autonomy or instruction drift.
Large portions of the skill are presented with Chinese headings, quotations, and terminology without stating that the user can choose their preferred language or locale. This may conflict with language/locale policy expectations if the skill steers interaction style toward a specific language without explicit opt-in.
