ClawHub
Back to skill

Security audit

story-review

Security checks across malware telemetry and agentic risk

Overview

The skill is a coherent fiction review helper, with one disclosed-but-inconsistent project tracking write users should be aware of.

This skill is reasonable to install for story planning and review. Use version control or ask the agent for read-only review when you do not want it to update the foreshadowing tracker, and be aware that full or lean mode may spawn already-deployed local reviewer agents.

SkillSpector

By NVIDIA
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
86% confidence
Finding
The skill claims to be a review workflow, but it instructs updating project tracking files during execution. That hidden state mutation can surprise users, corrupt project bookkeeping, or let a review action modify repository contents when the user expected read-only analysis.

Intent-Code Divergence

High
Confidence
95% confidence
Finding
The documentation explicitly says `story-review` does not modify files, yet the workflow directs writes to `追踪/伏笔.md`. This contradiction is dangerous because users or automation may grant the skill broader trust as a read-only reviewer, while it can actually persist changes to the repository and alter future workflow state.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.