ClawHub
Back to skill

Security audit

进化引擎

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly coherent, but it asks for command execution while saying it does not need it and it creates persistent local memory files.

Review before installing. This skill will keep local memory about corrections, preferences, and project patterns under ~/evolution-engine/. Install it only if you want that persistence, and consider removing the exec tool request or requiring explicit confirmation before the first write.

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 (4)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The manifest requests the exec tool while the documentation claims the skill is pure Markdown and does not require exec. This mismatch is dangerous because it grants a more powerful capability than users are led to expect, increasing the risk of unintended command execution or abuse if the skill is invoked in a permissive agent environment.

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The skill describes persistent memory management through local file operations, including creating and updating files under ~/evolution-engine, but the declared tools only include read and exec and do not declare an explicit write capability. This creates an unsafe and ambiguous implementation gap: either the skill cannot do what it promises, or it may rely on exec as a backdoor for file writes, which broadens the attack surface and bypasses least-privilege expectations.

Vague Triggers

Medium
Confidence
85% confidence
Finding
The trigger keywords are very broad terms like reflection, learning, and self-improving that commonly appear in normal conversations. This can cause accidental activation of a persistence-oriented skill, leading to unplanned memory recording, file creation, or behavioral changes without clear user intent.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill stores persistent memory files under the user's home directory, but the user-facing description does not provide a clear upfront warning about this behavior. That omission undermines informed consent and may expose sensitive preferences, project information, or interaction history to unintended retention on disk.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.