ClawHub
Back to skill

Security audit

Review Common Core

Security checks for vulnerabilities and agentic risk

Overview

This skill is a shared review checklist package with disclosed local validation scripts and no evidence of hidden or harmful behavior.

Installers should treat this as shared review documentation plus optional local validation scripts. Review the scripts before operational use, run them only on intended checklist JSON files, and do not rely on suggested-item warnings as hard blockers unless your workflow explicitly enforces that.

SkillSpector

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 (3)

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding
The metadata says this skill is a non-standalone shared reference engine, but the documented directory structure includes standalone validation scripts that read local input via file paths, globs, or stdin and return process exit codes. This mismatch can mislead users and security controls about the skill's actual execution surface, causing under-review of code paths that process local files and potentially broad input sets.

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The reference Python validator does not robustly match the documented JSON shape: some `details` fields are objects while the code assumes iterables of objects, and it also relies on `description` being present inside each check although the example output omits it. In practice this can cause exceptions or silently miss failed mandatory items, allowing incomplete reviews to be marked as acceptable and weakening the review gate.

Intent-Code Divergence

Medium
Confidence
97% confidence
Finding
The prose says any suggested-item failure prevents a pass, but the sample implementation only emits warnings and returns `PASS_WARN` instead of failure. That mismatch creates a policy-enforcement gap where operators may believe suggested checks are gating requirements when the code will still pass the review, reducing the effectiveness of the control process.

Static analysis

No suspicious patterns detected.