ClawHub

Deep Research

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only deep research workflow that is clear about planning, user approval, and delegated research execution.

Before installing, confirm that the companion deep-research-executor skill or tool is trusted, since final behavior depends on it. Avoid including confidential business data, private documents, credentials, or personal information in research plans unless you are comfortable passing that context to a spawned research agent.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (2)

Context-Inappropriate Capability

Medium
Confidence
88% confidence
Finding
The skill explicitly instructs spawning a new sub-agent session, which expands capability from planning into delegated execution. Agent spawning is security-relevant because it can bypass tighter control in the main session, increase autonomy, and make it easier to perform broader actions than the user may have clearly approved.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
Although the document repeatedly says Phase 2 requires explicit user approval, the execution section directly tells the agent to launch the sub-agent and does not operationally enforce that gate. In practice, this mismatch can cause an implementation to proceed into autonomous research execution without a clear approval checkpoint, undermining user intent and authorization boundaries.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal