Security audit

calculuslobt

Security checks for vulnerabilities and agentic risk

Overview

This is a calculus tutoring skill with two visible leftover editing fragments in sample exams, but no harmful code or sensitive access.

Safe to install for calculus tutoring use, with the caveat that the publisher should remove the leftover internal editing/tool-call fragments from the two sample exam reference files to avoid confusing downstream agents or students.

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)

Intent-Code Divergence

Medium

Confidence: 99% confidence
Finding: The markdown includes internal assistant scratchpad and tool-call markup such as closing think tags and replacement instructions inside teacher-facing reference material. This can leak hidden workflow details, confuse downstream agents or users, and create prompt-injection surface by blending trusted instructional content with unintended control/meta text.

Intent-Code Divergence

Low

Confidence: 98% confidence
Finding: The file includes prompt-generation artefacts such as a closing </think> tag, editorial instructions, and tool-call markup inside what is presented as a finished exam document. In an agent skill, this can leak internal reasoning scaffolding or cause downstream systems to misinterpret instructions as trusted content, creating prompt-injection and integrity risks.

Static analysis

No suspicious patterns detected.