ClawHub
Back to skill

Security audit

Prompt Version Control

Security checks across malware telemetry and agentic risk

Overview

The skill is not malware, but its documentation overstates important evaluation and sync features in ways that could mislead users managing production prompts.

Install only if you want a local demo-style prompt versioning helper. Do not rely on its A/B test quality scores, confidence labels, semantic summaries, or remote-sync documentation for production decisions unless you replace them with real evaluation and clearly configured git workflows. Keep proprietary prompts and test data out of shared repositories unless you have reviewed the .prompt contents and remote settings.

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 (2)

Intent-Code Divergence

High
Confidence
88% confidence
Finding
The skill describes LLM-assisted semantic diffing and A/B evaluation as if they are real, while the implementation later states that outputs are simulated offline with deterministic metrics and no LLM API calls. This can mislead operators into trusting fabricated quality signals, causing unsafe prompt promotions, ineffective rollbacks, or false assurance around evaluation results.

Vague Triggers

Medium
Confidence
88% confidence
Finding
The skill defines remote sync using highly generic commands, "prompt push / prompt pull", without namespacing, validation details, or clear transport constraints. In an agent or CLI-integrated environment, this ambiguity can cause command confusion, invoke unintended tooling, or trigger data synchronization to an unexpected remote target, which is especially risky because the skill handles prompt history and metrics that may contain sensitive intellectual property or test data.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.