Security audit

solve-tech-problems

Security checks for vulnerabilities and agentic risk

Overview

This is a simple advisory engineering-response skill with minor scope drift but no code execution, persistence, credentials, or data-moving behavior.

Installers should understand that the skill may answer broader engineering-style requests unless the invoking agent enforces the materials-engineering-only description. It is low risk as a prompt-only skill, but publishers should tighten the wording if they want strict domain boundaries.

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 (2)

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The manifest says the skill is limited to materials engineering, but the body of the skill expands its scope to generic technical problem solving. This creates instruction-domain drift: an orchestrator or reviewer may approve the skill for a narrow, lower-risk use case while the runtime behavior encourages broader assistance, potentially including advice outside intended safety, policy, or competency boundaries.

Description-Behavior Mismatch

Medium

Confidence: 90% confidence
Finding: The core behavior and response template are written as a general engineering analysis framework rather than a materials-specific skill. Because the structure is reusable for many technical domains, the skill can be invoked to produce procedural guidance in unintended areas, increasing the chance of unsafe or policy-misaligned outputs due to overly broad applicability.

Static analysis

No suspicious patterns detected.