ClawHub
Back to skill

Security audit

iaiops-water

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a water-treatment monitoring guide, but it repeatedly says the profile is read-only while listing tools that can export, push, or publish operational data.

Review this before installing in any live plant environment. Confirm the actual loaded water profile cannot export, push, or publish plant data unless those actions are explicitly approved, scoped to known destinations, and logged. The artifact does not show malware or direct control writes, but the read-only labeling is not reliable as written.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill repeatedly claims the water edition is read-only, but the documented tool list includes capabilities such as historian_push, export_data, stream_publish, uns_publish, and stream_publish_event, which are not inherently read-only. In an ICS/water-treatment context, this mismatch can cause an agent or operator to trust the skill as safe for passive monitoring while it may expose paths for data exfiltration, state mutation in downstream systems, or unsafe automation side effects.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The workflow section states the edition surface is entirely read-only, yet the same document earlier enumerates tools with push/publish/export semantics. This creates a security-boundary deception problem: downstream policy engines, reviewers, or operators may permit use under a read-only assumption even though the available surface appears capable of transmitting or changing operational data paths.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The main instructional content and operational guidance are written in Chinese, and the file does not indicate that users may choose another language or that the locale restriction is required for a region-specific deployment. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.