ClawHub
Back to skill

Security audit

multimodal-gen

Security checks for vulnerabilities and agentic risk

Overview

The skill mainly helps with multimodal generation, but it also includes a broad persistent learning module that stores user preferences and operational notes without clear limits or controls.

Review this skill before installing if you do not want persistent local records of preferences, errors, or usage notes. The core prompt helper appears ordinary, but the learning script can create learned_patterns.json containing free-form user or operational context, so avoid recording sensitive prompts or personal data unless you are comfortable managing that file yourself.

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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
Findings (4)

Lp3

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding
The skill documents a persistent learning feature that writes memories to `learned_patterns.json`, but no corresponding permission is declared. Undeclared file-write capability breaks least-privilege expectations and can lead to unauthorized persistence of user preferences, errors, or other sensitive operational data without informed approval.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding
The published description presents the skill as a multimodal generation helper, but the document also includes a generic self-learning system that records preferences, failures, insights, and reflections to disk. This hidden or under-disclosed behavior increases the risk of covert data collection and persistence, especially because user prompts and operational errors may contain sensitive information unrelated to media generation.

Description-Behavior Mismatch

Medium
Confidence
90% confidence
Finding
This file implements a generic cross-skill persistence and self-improvement module that is broader than the declared multimodal generation purpose. Scope creep matters here because a skill that can be reused to store history and preferences across arbitrary skills increases the chance of undocumented data collection and unintended coupling between skills, even if there is no overtly malicious behavior in this file.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The code persistently records usage history, error notes, recent operations, and arbitrary user preferences into learned_patterns.json without any visible consent, minimization, retention limit, or sensitivity filtering. In practice, notes and preference values may contain prompts, personal data, or operational context, creating a privacy and data leakage risk if the skill directory is shared, synced, or later consumed by other components.

Static analysis

No suspicious patterns detected.