Security audit

Academic Paper Refinement

Security checks for vulnerabilities and agentic risk

Overview

This academic refinement skill is mostly a manuscript-editing workflow, but it needs review because it makes AI-detection reduction a required academic writing step and runs broad external review/editing tools over user papers.

Install only after reviewing the workflow carefully. Use it on a copy of the manuscript in a dedicated folder, disable auto-approve where possible, confirm what each external tool can read or store, handle API keys outside shell history, and do not use the Humanizer step to bypass AI-use disclosure or academic-integrity requirements.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The 'Humanizer' section explicitly aims to lower AI-detection probability rather than improve scholarly quality alone. That makes the skill an evasion aid for detection systems, enabling deceptive submission of AI-generated academic text under a more human-seeming presentation.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The workflow prescribes measuring AI-detection scores before and after rewriting and iterating based on those results, which operationalizes evasion of detection tools. This moves beyond ordinary editing into optimization against integrity controls commonly used in academic review settings.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The manifest omits that the workflow includes AI-detection evasion behavior, which hides materially risky functionality from users and reviewers. Concealing that capability increases the chance the skill is deployed in contexts that prohibit deceptive authorship masking.

Natural-Language Policy Violations

High

Confidence: 98% confidence
Finding: The documentation instructs users to rewrite text specifically to suppress AI-generation signals, which supports circumvention of provenance and integrity checks. In an academic-paper skill, this is especially concerning because it facilitates misrepresentation of authorship and undermines review trust.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The skill directs execution of multiple external binaries and shell-style validation commands over user papers and related files without a clear up-front warning about that operational behavior. This creates avoidable risk around privacy, unexpected file processing, dependency trust, and user consent.

Ssd 2

High

Confidence: 99% confidence
Finding: The 'humanization' guidance is not just stylistic polish; it is framed as reducing detectable AI traces and includes concrete transformation strategies to do so. That is a direct evasion pattern that can be used to bypass academic integrity screening and other detector-based controls.

Static analysis

No suspicious patterns detected.