Security audit

model-router

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a local model-tier recommendation tool, but its requested Edit authority and wording about executing model calls are broader than its stated no-filesystem/no-network design.

Review before installing because the skill asks for Edit even though its actual router only needs to read input and print recommendations. If installed, treat it as advisory unless you explicitly want it to change model routing or execute downstream model calls.

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 (9)

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The trigger list includes broad natural-language phrases such as cost-saving and model-selection concerns that can overlap with ordinary conversation, increasing the chance the skill is invoked when the user did not actually request routing behavior. In an agentic system, mis-triggering can cause inappropriate workflow selection, unnecessary policy application, or disclosure of internal guidance in contexts where it is not relevant.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 88% confidence
Finding: Allowing activation based on implicit quality/cost context instead of explicit user intent creates confused-deputy risk: the agent may switch workflows or alter model routing without the user deliberately requesting it. In context, this skill affects how tasks are processed and could downgrade quality, change privacy posture, or interfere with expected controls if triggered opportunistically.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The natural-language triggers are broad enough to activate on ordinary discussion about cost savings rather than a clear user request to invoke routing logic. In an agent ecosystem, overbroad activation can cause unintended skill execution, context capture, or workflow steering, especially when the skill influences model selection and downstream processing behavior.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The documentation says the system will "Execute the call and monitor output quality," which describes active invocation of models and runtime observation. That contradicts the declared interface and permissions in this same skill file, which present the skill as a local recommendation/router component with network, filesystem, and subprocess access all disabled.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The line states "执行调用并监控输出质量," indicating the skill actively carries out model calls and watches output quality at runtime. This conflicts with the manifest permissions and interface, which do not permit the external actions such execution would normally require.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The trigger list is broad and includes common cost-optimization phrases that can plausibly appear in ordinary conversation, increasing the chance the skill activates when the user did not explicitly request model-routing behavior. In an agent setting, unintended activation can steer execution, recommendations, or downstream model selection in ways the user did not intend, creating reliability and policy-bypass risk rather than direct code-execution risk.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: Natural-language strings throughout the file, including the module description, model tier names, capability labels, and CLI output, are hard-coded in Chinese. This creates a locale policy concern because users are not given an opt-in or alternative language despite the tool also recognizing English task keywords.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 78% confidence
Finding: The document is structured as two mandatory language sections, English and Chinese, with no indication that language should be selected based on user preference. This can be a natural-language locale policy issue because it hardcodes language presentation rather than offering choice or documenting a justified locale constraint.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 85% confidence
Finding: All user-facing natural-language instructions in this file are written in Chinese, and there is no indication that the skill supports other languages or that Chinese is a justified locale-specific requirement. This can violate language/locale policy if the skill is expected to be usable without forcing a specific language absent user opt-in.
Content: No source excerpt is available for this finding.