ClawHub
Back to skill

Security audit

火一五智能体扩展技能

Security checks across malware telemetry and agentic risk

Overview

This skill appears intended to sync agent configuration locally, but it can overwrite persistent agent behavior files and is not tightly constrained to safe target workspaces.

Install only if you trust the system that supplies agentId values and you are comfortable with dynamic-agent configuration, memory, and skills being replaced from the main workspace. Prefer a version that validates agentId, confines writes to ~/.openclaw/workspace-* paths after resolution, supports dry-run or confirmation, and backs up overwritten 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 (4)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The code comments state that directory sync should copy only non-existent files, but the implementation calls a recursive copy routine that overwrites existing files in the dynamic agent workspace. This mismatch is security-relevant because operators or downstream code may rely on the documented non-destructive behavior and unintentionally lose local files or trust boundaries in agent-specific workspaces.

Missing User Warnings

High
Confidence
95% confidence
Finding
The skill explicitly documents overwrite-based synchronization of configuration files and conflicting directory contents, but does not clearly warn that local dynamic-agent changes may be destroyed. In a multi-agent environment, this can cause silent loss of configuration integrity, accidental privilege propagation, or unsafe replacement of agent behavior files such as AGENTS.md, TOOLS.md, and MEMORY.md.

Missing User Warnings

High
Confidence
96% confidence
Finding
The documentation states that first-time synchronization is automatically triggered and modifies the dynamic agent workspace without an explicit warning or consent step. Automatic mutation of a workspace at creation time increases the chance of unintended state changes, propagation of unsafe configuration, and difficult-to-audit behavior across dynamically created agents.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill forcibly overwrites configuration and workspace content in per-agent directories without any explicit warning, confirmation, or safety guard. In this context, dynamic agent workspaces may contain agent-specific instructions, memory, or skills, so destructive synchronization can silently replace trusted local state and alter agent behavior in a way that may enable persistence or privilege misuse.

VirusTotal

No VirusTotal findings

View on VirusTotal

Static analysis

No suspicious patterns detected.