Back to skill

Security audit

Geoskill: Blue Carbon Assessment

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a disclosed geospatial assessment tool, but it can produce synthetic results while labeling them as file-based assessment output.

Review carefully before installing. Treat current outputs as demo or screening outputs only, and do not rely on results from --ecosystem-raster until real raster loading is implemented or the tool fails closed instead of substituting synthetic data. If enabling remote downloads, confirm what service is contacted, where cached data is stored, and how to clear it.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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
Findings (6)

Lp3

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding
The skill documentation indicates file read/write behavior, but no permissions are declared for those capabilities. This creates a transparency and policy-enforcement gap: users or a hosting agent may approve the skill without understanding that it can access local files and write outputs, increasing the risk of unintended data exposure or filesystem modification.

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The skill file documents automatic remote downloads from Microsoft Planetary Computer, but this networked behavior is omitted from the manifest description. Hidden remote data access changes the trust model because a user may expect an offline local-analysis skill while the tool actually initiates external connections and ingests untrusted remote content.

Intent-Code Divergence

High
Confidence
99% confidence
Finding
When a user supplies an ecosystem raster, the pipeline logs that file-based mode is being used but then silently falls back to synthetic data. This produces materially false assessment outputs while presenting them as if they were derived from the user’s input, creating an integrity vulnerability that can mislead downstream decisions, reports, and audits.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill documents automatic remote download and local caching behavior without an explicit warning that network access will occur and data may be stored on disk. This is dangerous because users may unknowingly transmit query metadata such as AOI, bbox, and date range to a third-party service, and cached data may persist locally beyond the session.

Unpinned Dependencies

Low
Category
Supply Chain
Content
geoskill-data-fetcher>=0.2.0

# Third-party dependencies used by this skill
numpy
Confidence
95% confidence
Finding
numpy

Unpinned Dependencies

Low
Category
Supply Chain
Content
# Requirements for blue-carbon-assessment
# Shared data-download library (Planetary Computer / NASA POWER / OSM)
geoskill-data-fetcher>=0.2.0

# Third-party dependencies used by this skill
numpy
Confidence
97% confidence
Finding
geoskill-data-fetcher>=0.2.0

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.