Security audit

Actor Identifier Tool Free

Security checks across malware telemetry and agentic risk

Overview

This is mostly a local Git repository analysis skill, but it contains conflicting instructions that could allow write-like actions or network diagnostics despite claiming to be read-only and offline.

Review this skill before installing. It appears intended for local, aggregate Git analysis, but users should enforce the documented read-only git command whitelist, use dry-run first, and avoid any save/export, modify/reset/import, or ping/network behavior unless they explicitly approve it.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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)

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: The skill repeatedly promises a read-only, no-write, no-network posture, but the technical implementation section introduces generic create/modify/delete operation modes. That contradiction weakens the security contract and can mislead an agent into invoking broader behaviors than users consented to, especially in tool-using environments with exec access.

Intent-Code Divergence

Medium

Confidence: 90% confidence
Finding: The output section states that results may be exported/saved, which conflicts with later claims of no writes. In an agent context, 'save' can reasonably be interpreted as writing files, creating artifacts, or persisting data outside the repository, expanding the skill's effective privileges beyond what the user expects.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The error-handling section recommends running ping/network diagnostics despite the skill claiming zero-network operation. This contradiction can normalize off-scope network commands and erode the stated local-only trust boundary, potentially causing unintended outbound connectivity or environment probing.

Intent-Code Divergence

Low

Confidence: 82% confidence
Finding: Stating that performance depends on network conditions contradicts the skill's all-local design and creates ambiguity about whether remote services or connectivity may be involved. While weaker than the other contradictions, it still undermines user understanding of the skill's trust and data-flow boundaries.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The capability description uses broad generic operation modes such as create/query/modify/delete/import/reset that are not necessary for a repository-analysis skill. Overbroad invocation scope increases the chance that an agent will generalize permissions and perform unintended actions under vague parameter names like input_params or config_options.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.