ClawHub
Back to skill

Security audit

research-assistant

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed local research indexer that fetches user-supplied URLs and stores extracted data locally, with privacy and network caveats but no hidden or destructive behavior found.

Install only if you are comfortable with a local research index that can fetch any URL you give it and keep extracted information on disk. Do not use it on confidential documents, private network URLs, pages with PII, or URLs containing usernames, passwords, tokens, or email-like userinfo unless you plan to manually clean memory/research afterward.

SkillSpector

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 (3)

Intent-Code Divergence

Medium
Confidence
88% confidence
Finding
The code explicitly states that email addresses are not extracted to avoid PII leakage, but the URL regex will still capture full URLs, including credentials or email-like userinfo segments such as https://user@example.com@host. Because extracted entities and summaries are persisted to disk and later searchable, this can unintentionally retain and expose sensitive personal or credential-like data despite the stated safeguard.

Missing User Warnings

Medium
Confidence
86% confidence
Finding
The skill fetches arbitrary user-supplied URLs over HTTP/HTTPS, which can cause outbound requests to attacker-chosen destinations without any confirmation or restriction. In an agent environment, this creates SSRF-style risk, internal network probing, and privacy leakage because the host running the skill may have access to internal services or identifying network metadata unavailable to the user.

Ssd 3

Medium
Confidence
95% confidence
Finding
The skill explicitly instructs agents to extract content, persist it indefinitely under memory/research, and link key findings into MEMORY.md, creating a durable data-retention path for user-provided or fetched information. In this skill's context, that is especially risky because the same document also states there is no retention limit, no built-in deletion, and extracted material may include sensitive or personal data from files or arbitrary URLs.

Static analysis

No suspicious patterns detected.