Security audit

create-technology-intelligence-briefing-ip

Security checks across malware telemetry and agentic risk

Overview

This is a coherent technology-reporting skill, but its report builder executes a Python data file without a clear warning, so it should be reviewed before installation.

Install only if you are comfortable running the bundled Python renderer and any v2_data.py file it loads. Treat v2_data.py as executable code, review it before report generation, and keep outputs in a controlled working directory. The research connectors and file outputs are otherwise consistent with the skill’s stated purpose.

SkillSpector

By NVIDIA

Vulnerability Patterns

MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 94% confidence
Finding: The skill explicitly instructs the agent to read bundled files, use live patent/news/literature connectors, and create output artifacts on disk, which implies file-read, file-write, and network behavior without any declared permission model. This is dangerous because users and hosting systems may not have clear consent boundaries for external retrieval and filesystem writes, increasing the risk of unintended data access, exfiltration, or writes outside the expected workspace if the runtime does not enforce strict sandboxing.

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: The docstring presents the script as a safe static HTML builder from reviewed data, but the implementation later imports and executes v2_data.py as Python code. That mismatch is dangerous because users may trust the safety claim and supply or review data less carefully, leading to arbitrary code execution when the report is built.

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The documentation understates behavior by claiming the script only writes one HTML file, while the code also creates directories and executes the supplied input module. Misleading operational claims can cause unsafe deployment assumptions, especially in automation contexts where operators may not expect code execution from a reporting tool.

Description-Behavior Mismatch

High

Confidence: 100% confidence
Finding: load_data() uses importlib to load and execute a local Python file as input data, which means any code in v2_data.py runs with the privileges of the report-generation process. In the context of this skill, that is especially dangerous because the feature is supposed to process evidence-backed content, but instead it turns input data into executable code, enabling arbitrary local command execution, file access, or persistence.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The script executes a local Python data module without an obvious user-facing warning, so operators may treat the file as inert content rather than executable code. That increases the chance of accidental execution of malicious or unreviewed data, particularly in agent or automation pipelines where the input path may be produced by another component.

VirusTotal

61/61 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical

Code: suspicious.dynamic_code_execution
Location: scripts/build_report_v2.py:85