ClawHub
Back to skill

Security audit

skills-eval

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent skill-auditing guide, but it includes under-scoped examples for executing discovered tools and broad triggers that could activate it outside clear audit tasks.

Review this before installing if you work with untrusted skills or third-party tool declarations. Use it only in trusted development workspaces, skip or sandbox the active tool-execution and benchmarking examples, and consider narrowing triggers so it only activates for explicit skill-audit requests.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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
Findings (4)

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The benchmarking example actively executes an arbitrary binary supplied as tool_path via subprocess.run. In an auditing/benchmarking skill, this expands scope from passive analysis into code execution, which can run untrusted programs and trigger side effects on the host system if the path is attacker-controlled or insufficiently constrained.

Context-Inappropriate Capability

Medium
Confidence
88% confidence
Finding
The scalability section describes feeding generated datasets into a tool runner, which again introduces active execution capability beyond simple skill auditing. If used on untrusted tools or with unsafe runner logic, this can invoke harmful behavior, consume excessive resources, or process attacker-influenced input in a privileged environment.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The trigger list contains broad, common terms such as "evaluation," "improvement," and "skills," which can cause the skill to activate in many unrelated contexts. Over-broad activation increases the chance of unintended tool guidance or workflow injection into conversations that did not request this skill, potentially confusing behavior and expanding the skill's effective attack surface.

Unvalidated Output Injection

High
Category
Output Handling
Content
# Test tool runs with --help
            try:
                subprocess.run([tool_path, '--help'],
                              capture_output=True,
                              timeout=5)
                results.tools_functional.append(tool)
Confidence
92% confidence
Finding
subprocess.run([tool_path, '--help'], capture_output

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.