Agent Batch Guard

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only batching guide with some risky wording and one incorrect configuration checklist item, but no hidden execution, credential access, or destructive behavior.

Safe to install as a batching guide, but review any scripts an agent generates from it before running them. Keep device targets, accounts, output paths, and data scope explicit; do not use the randomized-delay advice to evade service rules or anti-abuse systems; and ignore or correct the checklist item that says to use compaction rolling/maxTurns.

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 (1)

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
The skill contains internally inconsistent configuration guidance: it correctly states earlier that OpenClaw only supports compaction.mode="safeguard", but the checklist later recommends setting compaction to rolling with maxTurns ≤ 40. This can mislead operators into applying invalid or unsupported settings, causing failed hardening, configuration errors, or a false sense that session-growth protections are in place when they are not.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal