unisound-academic-material-generation

• Outcome: Review
• Latest audit: Jul 16, 2026 · 8:39 PM UTC
• Version: 1.0.2

Overview

This skill performs the medical drafting function it describes, but users should treat submitted medical or business content as being sent to the disclosed remote model API.

Install only if you are allowed to send the entered topic, key messages, evidence points, references, and any extracted document content to the disclosed remote medical model API. Avoid PHI, patient identifiers, unpublished study data, or confidential commercial material unless your organization has approved that endpoint and processing arrangement.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 95% confidence
• Finding: The skill advertises and operationally depends on capabilities including file ingestion, OCR/document parsing, shell-invoked external tools, and outbound network access, yet it declares no permissions or trust boundaries. This creates a real security issue because users and the hosting platform are not clearly informed that local files may be processed and content transmitted to an external API endpoint, increasing the chance of overbroad access and unsafe deployment assumptions.

Context-Inappropriate Capability

Medium

• Confidence: 76% confidence
• Finding: The script executes local external parsers/converters (LibreOffice, pdftotext, tesseract) on supplied files, significantly expanding the attack surface beyond simple text generation. In a skill context, this is dangerous because untrusted documents can trigger vulnerabilities in those tools or cause resource exhaustion, and the behavior is not obviously necessary from the high-level skill description.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill accepts PDFs, Office documents, spreadsheets, text files, and images, performs text extraction/OCR, and then sends generated content through a remote medical-model API, but it does not provide a clear privacy notice, data-classification warning, or restrictions on sensitive medical/patient information. In a medical affairs context, this is especially dangerous because users may upload regulated or confidential material containing PHI, patient details, unpublished study data, or proprietary medical information that could be transmitted off-box without informed consent or sanitization.

Missing User Warnings

Medium

• Confidence: 80% confidence
• Finding: The code silently performs subprocess-based conversion and OCR without any user-visible disclosure or consent boundary. In practice this matters because it can surprise operators with execution of heavyweight local binaries on untrusted inputs, increasing trust, compliance, and attack-surface concerns in a medical-content workflow.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The code sends user-supplied topic, audience, key messages, evidence points, and references to an external LLM API endpoint, but this file contains no explicit user-facing disclosure, consent gate, or data-classification check before transmission. In a medical/pharma academic drafting context, inputs may contain confidential medical, scientific, or commercial information, so silent outbound transfer increases privacy, confidentiality, and compliance risk.

VirusTotal

62/62 vendors flagged this skill as clean.

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical

• Code: suspicious.dynamic_code_execution
• Location: scripts/run.py:195