ClawHub
Back to skill

Security audit

Eval Skill

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed skill-review helper that reads a target skill and asks sub-agents to score it, with no executable code, persistence, credential handling, or hidden install behavior found.

Install this only if you want an agentic reviewer for skill packages. Invoke it explicitly on the intended skill directory, and keep credentials, private notes, or unrelated files out of that directory because the skill is designed to read full text files and send their contents to evaluator sub-agents.

SkillSpector

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

Vague Triggers

Medium
Confidence
88% confidence
Finding
The description "帮我评估一下这个 skill。" is broad and resembles ordinary user language rather than a narrowly scoped invocation phrase. That increases the chance the skill will trigger unintentionally during unrelated conversations, causing the agent to read files, spawn sub-agents, and process potentially sensitive workspace content without explicit user intent.

Natural-Language Policy Violations

Medium
Confidence
80% confidence
Finding
The skill content is written to enforce Chinese-language interaction and output format without offering language selection or fallback. While not a direct code-execution issue, this can mis-handle user intent, reduce operator visibility during review, and increase the chance that warnings or analysis results are misunderstood in multilingual environments.

Static analysis

No suspicious patterns detected.