ClawHub
Back to skill

Security audit

Competitive Data Science

Security checks for vulnerabilities and agentic risk

Overview

The skill is a competition-focused data science reference with example modeling code, and the noted broad wording is a scoping quality issue rather than a security concern.

Use this as a reference skill for Kaggle-style or similar competitions. Do not apply its competition-optimization advice directly to production ML systems without separate validation, monitoring, security, and deployment review.

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (3)

Intent-Code Divergence

Low
Confidence
97% confidence
Finding
The inline comment at L215 says 'Rank-based fusion', which implies converting predictions to ranks before combining them. However, the code from L216-L225 directly computes a weighted average of the prediction values using model scores as weights, so the documentation contradicts the implemented method.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The manifest description says the skill should be used when 'working on data science competitions' and then lists broad activities like data processing, feature engineering, and model building. Those activities are common to many non-competition data science tasks, so the trigger scope is not specific enough to clearly distinguish when this skill should or should not activate.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The 'When to Use' bullets describe general situations like 'Improving an existing baseline solution' or 'Need advanced feature engineering techniques,' which are common across many ML workflows. Because the file does not define concrete trigger phrases or tighter competition-only constraints there, the activation conditions remain overly broad.

Static analysis

No suspicious patterns detected.