ClawHub
Back to skill

Security audit

Prompt Safe

Security checks across malware telemetry and agentic risk

Overview

This is a local prompt-assembly helper with disclosed memory and token-budget behavior, but users should treat its token safety and memory privacy guidance as guardrails rather than guarantees.

Install only if you are comfortable with a heuristic prompt-sizing helper. Validate token counts against your actual model before relying on it near context limits, and pair any long-term memory use with explicit user consent, review, deletion, and sensitive-data limits.

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 (2)

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The class docstring promises a guarantee against API failure from token overflow, but the implementation relies on rough character-based token estimation rather than the target model’s real tokenizer and overhead rules. In a prompt-assembly skill, this can cause downstream callers to trust unsafe sizing logic, resulting in oversized requests, dropped context, or operational denial of service when API calls fail unexpectedly.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The document explicitly allows storing identity and preference data in long-term memory, but it does not require user consent, retention limits, minimization, or any privacy notice. In a memory orchestration skill, this increases the chance that personally identifiable or sensitive preference data will be persistently retained and reused beyond user expectations, creating privacy and compliance risk.

VirusTotal

61/61 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.