ClawHub

Deepagents Implementation

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only skill for building Deep Agents; it describes powerful optional patterns but does not install code or perform actions by itself.

Install only if you want implementation guidance for Deep Agents. When copying examples, keep filesystem roots narrow, avoid home or system directories, require approval for shell commands and file writes, keep secrets in environment or secret managers, and define how persistent memories are scoped and cleared.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

Missing User Warnings

Medium
Confidence
84% confidence
Finding
The example explicitly instructs the agent to save important findings to persistent memory and check that memory on future tasks, but it provides no warning about retention, reuse of prior user data, or privacy implications. In an agent-building skill, this can normalize storing sensitive user content by default and can lead developers to deploy cross-session memory without consent, minimization, or deletion controls.

Missing User Warnings

High
Confidence
93% confidence
Finding
The example promotes a FilesystemBackend rooted at a real project directory and then states the agent can read/write real files and execute shell commands, without any warning about destructive actions, privilege scope, or sandboxing. In this skill context, users are likely to copy this pattern directly, which can grant an LLM-backed agent high-impact access to source code, secrets, and the host environment.

VirusTotal

67/67 vendors flagged this skill as clean.

View on VirusTotal