Security audit

patent-inventor-profile-report

Security checks across malware telemetry and agentic risk

Overview

The skill is not malicious, but it needs review because conflicting data-collection instructions could make formal inventor ratings and credit-style recommendations misleading.

Review this skill before installing if reports may be used for credit, hiring, investment, or other consequential decisions. Require full paginated data collection for all metrics, treat any sample-based output as incomplete, and avoid retaining generated report/progress files with raw personal names unless that is acceptable for your environment.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Description-Behavior Mismatch

High

Confidence: 98% confidence
Finding: The skill metadata explicitly requires that when patent counts exceed 100, the agent must paginate and return full data rather than use sampling. This file instead instructs the agent to analyze only the latest 100 records and base core metrics such as A14, IPC, citations, and legal status on that sample, creating a material integrity gap that can systematically misrepresent an inventor's profile. In a reporting skill, this is dangerous because downstream decisions may rely on incomplete or biased analytics while the output still appears authoritative and complete.

Description-Behavior Mismatch

High

Confidence: 97% confidence
Finding: The display-mode table hard-codes that any result set above 100 patents should still use only the latest 100 as the shown sample, directly conflicting with the stated requirement for full paginated return above 100 items. This inconsistency can cause the agent to suppress large portions of relevant data while preserving the appearance of policy compliance, making reports incomplete and potentially misleading. Because the skill is designed to generate polished HTML profiles, the presentation quality can amplify trust in inaccurate results.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The A14 dominance-rate verification is defined as a sample-based estimate when A1 exceeds 100, rather than a full-data computation. Since A14 appears to be a core credibility metric about inventor contribution, estimating it from a recency-biased subset can skew rankings or assessments, especially for prolific inventors with changing collaboration patterns over time. The skill context increases risk because the output is a formal talent-evaluation report, not an informal exploratory summary.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The spec instructs saving a progress file using a personal name in the filename, which unnecessarily embeds personally identifiable information into filesystem artifacts. This increases the chance of unintended disclosure through logs, directory listings, backups, shared sessions, or debugging output, especially in a multi-user or retained-session environment.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The design requires a real-time log panel during report generation but provides no safeguards against logging names, institutions, patent query terms, record contents, or intermediate analysis results. Real-time logs are commonly copied into UI output, browser history, monitoring systems, or stored artifacts, which can expose sensitive personal or proprietary data beyond the intended report audience.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.