Back to skill

Security audit

Advisory-Companion

Security checks for vulnerabilities and agentic risk

Overview

This advisory skill needs review because it calls for keeping sensitive client histories across conversations without clear controls over retention or deletion.

Before sharing sensitive details, require session-only handling or explicitly agree on what records may be saved, where, for how long, and how to review and delete them. Read-only mode reduces file-writing authority. Medical and psychological guidance still requires appropriate professional review.

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)

Sensitive client records are maintained without an instruction to warn the user about their retention.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The skill directs the agent to maintain client records for continuity without requiring notice or consent before retaining sensitive information. Its mental-health, medical, relationship, and child-related contexts increase the consequences of unexpected persistence or later unauthorized access; confidentiality instructions help but do not establish retention limits or informed consent. The file does not show an actual storage implementation or disclosure, so this is a privacy-design vulnerability rather than evidence of an existing leak.

Content

No source excerpt is available for this finding.