ClawBridge - Find Connections

• Outcome: Pass
• Latest audit: May 28, 2026, 1:54 AM
• Version:
  1.0.1

Overview

This appears to be a public-web lead research skill with disclosed web access and no evidence of hidden execution or automatic outreach, though users should treat it as privacy-sensitive prospecting.

Install only if you are comfortable with a skill that performs recurring public-web prospect research. Keep secrets and sensitive internal targeting strategy out of profiles, restrict it to public business information, review every candidate before outreach, and do not use it to scrape private communities or send messages automatically.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Intent-Code Divergence

Medium

• Confidence: 84% confidence
• Finding: The skill encourages discovery of Discord servers and Slack workspaces but does not clearly restrict collection to publicly accessible content at the point where those venues are introduced. Because many community platforms mix public invite links with semi-private or expectation-of-limited-audience spaces, an agent could over-collect personal or behavioral data from communities where participants did not expect lead-generation or profiling.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The README explicitly states the skill will scout the web, gather evidence, and rank candidates based on a user-provided profile, but it does not warn that profile details or derived search queries may be sent to external websites, search engines, or other third-party services. In a prospecting/outreach context, these inputs can contain sensitive business strategy, targeting criteria, or do-not-contact preferences, so lack of disclosure increases the risk of unintended data exposure and unsafe deployment.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The skill clearly describes a persistent nightly scouting workflow that searches the web, fetches external pages, and profiles people, but it does not present this as an upfront user-facing warning about ongoing external activity. That omission can lead users to enable the skill without fully understanding the continuous data collection and outbound network behavior, increasing privacy, consent, and unexpected-data-processing risk.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The inputs encourage users to provide business goals, target personas, regions, avoid lists, and example contact identifiers, and the workflow then uses external search/fetch tools, but the skill does not explicitly warn that such profile data may be processed alongside third-party web services. This can expose sensitive commercial targeting criteria or personal/contact data to external systems without clear disclosure, creating privacy and data-handling concerns.

Missing User Warnings

Medium

• Confidence: 84% confidence
• Finding: The prompt instructs the agent to return identifiable third-party profile data such as names, handles, companies, and evidence URLs, but provides no privacy guardrails, consent checks, minimization rules, or restrictions on exposing sensitive external-profile data. In a connection-ranking workflow this can enable unnecessary disclosure, profiling, and redistribution of personal data to downstream users who may not have a legitimate need to see all raw identifiers and source links.

Missing User Warnings

Medium

• Confidence: 83% confidence
• Finding: The skill explicitly instructs agents to extract contact information and social links from public web pages, but provides no privacy boundary, lawful-use guidance, or data-minimization rules. Even if the data is publicly visible, systematic collection and downstream use for outreach can create privacy, compliance, and abuse risks, especially when combined with company/person profiling.

VirusTotal

63/63 vendors flagged this skill as clean.