Back to skill

Security audit

ai-search-visibility

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign: the scanners found no reported executable code or persistence concerns, but flagged a possible privacy disclosure gap when user questions are sent to external AI services; the scan was partial.

Before installing, consider whether you would use the workflow with confidential buyer questions, since the flagged workflow may send them to external AI services without a privacy warning. The verdict is benign, but Skillspector marked its analysis partial because some local path-like references were unresolved.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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 (1)

The assistant-sampling workflow sends user-provided buyer questions to third-party AI services without a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

Step 4 instructs the agent to submit the buyer questions to ChatGPT, Perplexity, Claude and Gemini, directly or through ask_ai_models. Although the services are named, the skill does not warn that user-provided questions will leave the current environment or advise against including confidential business information; announcing the question count does not provide that disclosure.

Content

No source excerpt is available for this finding.