Security audit

短篇网文写作-免费版

Security checks for vulnerabilities and agentic risk

Overview

This short-story writing skill does not show malicious behavior, but it asks agents to run an unspecified local Python script with vague triggers, so users should review it before installing.

Install only if you are comfortable with the agent having read/write and command-execution capability for this writing workflow. Before using it, verify what short_story.py is supposed to be, avoid running it from directories containing unrelated scripts with that name, and provide only story files or prompts you are comfortable processing through the agent's LLM.

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Behavioral ASTexec() Call, eval() Call, Dynamic Import
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 (4)

Intent-Code Divergence

Medium

Confidence: 91% confidence
Finding: The documentation presents conflicting trust and deployment assumptions: it says the skill is local-only and needs no extra API key, while also declaring an LLM API dependency and including network-related troubleshooting. This can mislead users and host agents about data flow, causing sensitive prompts or files to be handled under false assumptions about offline execution and external transmission.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: An overly broad natural-language trigger can cause the agent to invoke this skill for loosely related writing requests without clear user intent. Because the skill advertises file read/write and command execution capabilities, accidental activation increases the chance of unintended local file access or command execution in response to ordinary prompts.

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The trigger conditions section is incomplete and unclear, leaving the agent to infer when the skill should run. In a skill with execution and filesystem permissions, ambiguous activation logic raises the risk of inappropriate routing, unexpected behavior, and unsafe use of privileged capabilities.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill description emphasizes creative writing features but does not warn users that the skill has local file read/write and command execution tools. This omission creates a consent and transparency problem: users may provide sensitive content or approve use without understanding that the skill can access or modify local files and run commands.

Static analysis

No suspicious patterns detected.