OpenClaw Subagent Toolset

Security checks across malware telemetry and agentic risk

Overview

This is a Markdown-only helper for choosing sub-agent tool prompts; it has a minor internal inconsistency but no hidden code, persistence, or data transfer behavior.

Treat this as a prompt-template reference, not a real sandbox. Fix or account for the Secretary glob_search contradiction, and enforce actual tool permissions through platform controls before relying on these subsets for sensitive work.

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: 97% confidence
Finding: The Secretary subset contradicts itself by allowing glob_search while also forbidding glob_search and grep_search, creating ambiguous policy guidance for sub-agents. In a skill whose purpose is to constrain tool use via prompt instructions, contradictory rules can cause inconsistent enforcement, accidental over-permission, and make it easier for a model or attacker-crafted task to justify use of a supposedly forbidden capability.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal