Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly a disclosed process-plant diagnostics guide, but it mixes read-first safety claims with under-scoped publish, push, and export capabilities in an OT context.
Review this skill before installing in a production or sensitive plant environment. It appears intended for diagnostics and read-first analysis, not deception or persistence, but you should require explicit approval controls, dry-run defaults, destination limits, and audit logging before allowing historian push, UNS or stream publish, or data export tools to run against real plant data.
The primary skill instructions and operating guidance are written in Chinese, including the heading and startup guidance, without stating that the user may choose another language or that the skill is restricted to a Chinese-language environment. This creates a natural-language policy concern because it effectively imposes a locale/language assumption without opt-in.
The skill markets itself as read-first and MOC-gated, but the documented toolset includes write-adjacent capabilities such as historian_push, export_data, stream_publish, uns_publish, and stream_publish_event without consistently presenting them as exceptions subject to the same approval controls. In an OT/process-plant context, ambiguity about whether an agent may publish or export data can lead to unauthorized state changes, unintended data exfiltration, or unsafe operator trust in the skill's safety boundaries.
The manifest describes protocol access for HART/OPC-UA/Modbus and plant analytics such as RCA, data quality, and OEE. compliance_mapping, compliance_frameworks, compliance_dengbao_levels, compliance_report, and compliance_evidence_bundle introduce a governance/compliance capability that is not clearly tied to the stated instrumentation-focused purpose.
No suspicious patterns detected.
