Security audit

Code Analysis Toolkit

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a coherent Git history analysis guide, but it gives conflicting privacy claims about LLM-backed processing of repository and contributor data.

Review this before installing for private or workplace repositories. Treat Git history, author names, emails, reports, baselines, and consent logs as sensitive. Do not rely on the 'fully local/no external transmission' claim unless your agent platform and model configuration actually keep prompts and tool outputs local. Keep .code-analysis out of version control unless intentionally shared, and restrict access to consent and audit files.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Intent-Code Divergence

High

Confidence: 96% confidence
Finding: The documentation explicitly claims the tool is fully local and does not transmit data externally, while also declaring an LLM API dependency provided by the agent platform. This can mislead users into processing sensitive Git history and personal data under a false privacy assumption, causing unintended disclosure to external model providers or hosted agent services.

Description-Behavior Mismatch

Medium

Confidence: 88% confidence
Finding: The skill markets itself as local, privacy-protective Git history analysis, but later requires LLM-backed processing through the agent environment. That mismatch can cause users to expose repository history, contributor identities, and internal engineering patterns in contexts where they expected only local analysis.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The skill instructs users to create and maintain consent logs containing names, email addresses, timestamps, scope, expiry, and revocation state, but it does not prominently warn that personally identifiable information will be stored on disk. In a team-analysis context, silent persistence of this data increases privacy and compliance risk, especially on shared workstations or repositories.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.