Deep Researcher Skill

• Outcome: Review
• Latest audit: May 29, 2026 at 9:29 PM UTC
• Version:
  1.0.0

Overview

This research skill is mostly documentation, but it tells the agent to bypass paywalls by default, so it should be reviewed before installation.

Install only if you are comfortable reviewing and constraining its paywall behavior. Use it with open, licensed, or otherwise authorized sources; avoid entering sensitive API keys in plain chat; verify the external GitHub repository before cloning; and clarify how research history is stored or cleared if your host platform enables that feature.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (7)

Context-Inappropriate Capability

Medium

• Confidence: 98% confidence
• Finding: The skill explicitly instructs the agent to circumvent paywalls using archive mirrors, 12ft.io, r.jina.ai, and direct browser extraction. That goes beyond normal research assistance into access-circumvention behavior, creating legal, terms-of-service, and policy risk while normalizing unauthorized retrieval of restricted content.

Context-Inappropriate Capability

Medium

• Confidence: 97% confidence
• Finding: Describing the browser tool as being used 'for paywall bypass' assigns an access-circumvention purpose to a powerful tool. This increases the chance the agent will misuse browsing capabilities to obtain content the publisher intended to restrict.

Intent-Code Divergence

Medium

• Confidence: 96% confidence
• Finding: The guardrails say to 'Bypass paywalls before skipping,' which conflicts with the skill's otherwise quality- and ethics-oriented framing. Because guardrails are normative instructions, this makes circumvention a required behavior rather than an edge case, materially increasing misuse risk.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The paywall-bypass instructions are presented without any warning about legal, ethical, or contractual implications. That omission makes unsafe behavior appear standard and endorsed, which is especially risky in an agent skill intended for repeated use.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: Labeling the browser tool as intended 'for paywall bypass' gives users and the agent no warning that the behavior may be inappropriate or prohibited. This embeds risky use directly into prerequisite guidance, making misuse more likely from the start.

Ssd 3

Medium

• Confidence: 88% confidence
• Finding: The command 'Show my research history' implies retention and later exposure of prior user queries and findings, but the skill provides no data-handling, consent, retention, or access-control details. Research history can reveal sensitive interests, business plans, health, legal, or political topics and should not be stored or surfaced without explicit safeguards.

Ssd 3

Medium

• Confidence: 87% confidence
• Finding: Advertising 'research history' in version notes reinforces that the skill may persist prior queries and findings across sessions. Persistent tracking without a privacy notice, retention limits, or user controls creates unnecessary data exposure risk and could leak sensitive research activity.

VirusTotal

59/59 vendors flagged this skill as clean.