Security audit

Ai Writing Style Cloner

Security checks for vulnerabilities and agentic risk

Overview

This skill is transparent about cloning and saving author style profiles, but it gives broad command/file access and durable author-profile reuse without enough consent, access-control, or retention boundaries.

Review carefully before installing. Only use this with writing samples you have the right to analyze, avoid impersonating real people or brands, keep style_fingerprints private, delete profiles when no longer needed, and prefer running it where command execution is restricted or replaced with safer file APIs.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Context-Inappropriate Capability

Medium

Confidence: 92% confidence
Finding: The skill declares generic exec capability for a workflow that mainly needs LLM processing and simple file persistence. Unnecessary command-execution privileges materially increase the blast radius: if later prompts, wrappers, or implementations pass user-controlled values into shell commands, this can become command injection or arbitrary local command execution.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The skill is explicitly designed to clone a third party author's style from uploaded samples and persist a reusable per-author fingerprint, but it does not clearly warn about consent, copyright, privacy, or misuse risks. In this context, the missing policy guardrails make abuse more likely, including impersonation, unauthorized profiling, and retention of content-derived personal or proprietary information.

Ssd 3

Medium

Confidence: 94% confidence
Finding: The skill persists per-author style fingerprints and explicitly stores raw analysis artifacts for later reuse, creating a durable repository of content-derived profiles. Even if full samples are not stored verbatim, this retained natural-language metadata can expose private, proprietary, or identifying traits and becomes a target for unauthorized access or secondary use.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The preview flow returns saved fingerprints together with raw_analysis and timestamps on request, which can disclose previously derived author-profile data to anyone who can supply or guess an author_id. In a style-cloning skill, this is more dangerous because the data is specifically intended to capture distinctive traits that may be sensitive, proprietary, or usable for impersonation.

Static analysis

No suspicious patterns detected.