Security audit

Catalog

Security checks for vulnerabilities and agentic risk

Overview

This is a simple catalog skill that runs a bundled local script to return fixed service prices, with no evidence of data access, network activity, persistence, or deception.

Before installing, understand that this skill answers service and price questions by running a bundled local Node.js script. The inspected script is simple and benign, but users who want maximum transparency may prefer the catalog data to be stored as static text instead of executed code.

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

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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)

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The trigger condition 'when the user asks about services/prices' is broad, which can cause the skill to run the local command in contexts the user did not clearly intend. While the command is fixed and low risk here, ambiguous invocation increases unnecessary command execution and can create surprising tool use.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The skill directs the agent to execute a local command without any user-facing disclosure or warning, which reduces transparency around tool use. Even though the command appears fixed and benign, silent local execution can erode consent boundaries and becomes more dangerous if the script later changes or gains access to sensitive local resources.

Description-Behavior Mismatch

Low

Confidence: 84% confidence
Finding: The manifest description suggests a minimal, simple catalog capability, while the skill documentation instructs the agent to invoke the Exec Tool and run a local Node.js program. Executing local code is a materially broader behavior than a plain 'hello world' catalog description implies.

Context-Inappropriate Capability

Low

Confidence: 80% confidence
Finding: A simple catalog skill would typically present predefined service and price information; invoking an execution tool to run a local program is a stronger capability than the stated purpose alone justifies. Because the manifest does not disclose this execution behavior, the capability is contextually surprising.

Static analysis

No suspicious patterns detected.