ClawHub
Back to skill

Security audit

输入验证

Security checks for vulnerabilities and agentic risk

Overview

This is a QA input-validation skill with some overbroad workflow wording, but no hidden execution, persistence, exfiltration, or destructive behavior was found.

Install only if you want a QA workflow gate that may ask clarification questions and may produce limited test-case guidance from incomplete requirements. Avoid giving it real customer, payment, ID, phone, or production data unless it is properly masked, and review any generated test cases that were produced after incomplete input.

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

Medium
Confidence
89% confidence
Finding
The skill states that failed input validation must return missing information and request clarification, but later instructs the agent to produce a minimal test case set when the user refuses to provide more details. That contradiction weakens the safety boundary of this gatekeeper skill and can cause downstream test generation from incomplete or ambiguous requirements, leading to misleading outputs and accidental handling of wrong scope or sensitive assumptions.

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
This skill is presented as an input-validation step, but its required output includes a full test case table. That scope expansion defeats the purpose of a validation-only guardrail and may cause the agent to begin substantive test design before prerequisites are met, bypassing workflow separation and increasing the chance of low-quality or unauthorized output generation.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The auto-activation condition is broad and ambiguous, covering vague cues like users saying information is unclear. In an agent environment, such triggers can cause the skill to activate unexpectedly, intercept unrelated requests, and steer the conversation into a rigid validation workflow, which increases the risk of denial-of-service-like friction and incorrect tool usage.

Static analysis

No suspicious patterns detected.