Security audit
Security checks for vulnerabilities and agentic risk
The skill is a plain analytics-reporting skill, but it handles sensitive business and financial metrics and tells the agent to send reports to an unclear recipient without enough scoping.
Before installing, confirm who Thomas is, what data may be included in reports, which channel is approved, and whether payment, lead, revenue, log, or trading data must be redacted. Define the 'Major Events' trigger so reports are generated only for intended events.
Natural-language policy review applies to all file types. The skill instructions and operational content are presented in German throughout, which effectively forces a specific language without any user opt-in or explanation that the skill is intentionally limited to a German-speaking context.
The instruction to send reports 'An Thomas' introduces a data-sharing action without specifying who Thomas is, what data may be included, or whether the recipient is authorized to receive potentially sensitive business metrics. Because the skill covers trading performance, app payments/errors, leads, and revenue, accidental disclosure of confidential or personal data is a realistic risk.
The trigger 'Nach Major Events' is underspecified and can cause the skill to run after undefined or overly broad events, leading to unexpected report generation or data processing. In a skill that aggregates trading, app, freelance, and marketing data, this ambiguity increases the chance of unauthorized or excessive access, especially if automated systems interpret 'major events' differently.
No suspicious patterns detected.
