ClawHub
Back to skill

Security audit

Deep Research

Security checks across malware telemetry and agentic risk

Overview

This skill openly sends research questions to a local research API and shows no hidden persistence, credential access, destructive behavior, or unrelated data collection.

Install only if you run or trust the Agentic RAG-CN service on localhost:18888. Avoid using it with secrets, credentials, proprietary material, or regulated personal data unless you understand how that local service logs, stores, and forwards research queries.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding
The skill clearly instructs the agent to perform network operations against a local HTTP service, but no permissions are declared. Undeclared network capability is risky because it hides data-flow and runtime behavior from review systems and users, and it may allow unintended transmission of user queries to another service.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding
The skill claims broad multi-source research behavior, but the implementation shown only forwards the user's question to a local service and trusts that service to do the work. This mismatch is dangerous because reviewers and users cannot verify what sources are actually queried, what additional actions the local service performs, or whether the promised fallback and source constraints exist at all.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The script sends user research queries to a local API over plain HTTP and does not provide any warning or consent prompt that the input is being transmitted to another service. While the endpoint is localhost, this still creates confidentiality and trust concerns because sensitive prompts may be forwarded to a separate process without explicit user awareness, and local traffic may be exposed in some environments through proxies, container networking, port forwarding, or a malicious local service binding to that port.

External Transmission

Medium
Category
Data Exfiltration
Content
Agentic RAG-CN must be running locally. Check with:

```bash
curl -s http://localhost:18888/health
```

If unavailable, fall back to built-in `web_search` tool.
Confidence
90% confidence
Finding
curl -s http://localhost:18888/health ``` If unavailable, fall back to built-in `web_search` tool. ## Usage Call the API with a POST request: ```bash curl -s -X POST http://localhost:18888/api/ana

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.