Security audit
Security checks for vulnerabilities and agentic risk
This HR diagnosis skill is mostly purpose-aligned, but it handles sensitive HR and applicant data with unclear retention and scope boundaries.
Review this skill before installing if you expect it to process real employee or applicant data. Use it only with explicit user direction, avoid storing case summaries for restricted matters such as layoffs or labor disputes, and verify any missing reference files or knowledge-base behavior before relying on it for sensitive HR decisions.
The manifest description directs use in Chinese ('Use when user asks to 深度分析问题...'), and the operational prompts and outputs throughout the file are prescribed in Chinese without any opt-in or language selection. This can violate language/locale policy because the skill appears to force a specific language rather than offering user choice.
The implicit triggers are broad terms like 'why,' 'recently,' or 'not right,' which overlap with ordinary HR conversation and can cause the skill to prompt for deep diagnostic mode unnecessarily. In this skill's context, unintended activation is more dangerous because the diagnostic flow is designed to collect and process highly sensitive HR data such as performance, compensation, and labor-dispute information.
The document states Level 4 restricted data must not be stored, but the later allowance to store desensitized case summaries for Level 3+ flows creates ambiguity about whether restricted-case derivatives may be retained. In an HR context involving layoffs, disputes, and compliance issues, ambiguous retention rules can lead to unauthorized persistence of highly sensitive information or insufficiently anonymized summaries.
The candidate-evaluation module materially expands the skill from organization diagnosis into external hiring assessment, including resume auditing, deception-oriented questioning, and capability grading. That scope expansion increases the chance the agent handles more sensitive personal data and makes higher-stakes judgments without explicit manifest scoping, separate safeguards, or tighter consent boundaries.
The skill is presented as an HR deep organization-diagnosis system, but its expert catalog includes an administrative expert for office environment, vendors, and event planning, plus AI application/transformation experts for tool selection and end-to-end AI value-chain redesign. Those capabilities are not obviously necessary to perform the core diagnosis functions described in the manifest and expand the skill into adjacent consulting domains.
No suspicious patterns detected.
