ClawHub
Back to skill

Security audit

Prompt Benchmark

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local prompt-review helper that reads prompt text, runs a deterministic lint script, and asks the agent to produce a static quality report without network calls or persistence.

Install this if you want local, static prompt-quality analysis. It should not be treated as a runtime benchmark: accuracy, latency, cost, hallucination rate, and model performance are explicitly not measured unless separate dynamic tests are run.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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)

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The code is clearly aligned with part of the description: it statically analyzes prompt text without invoking external models, and it evaluates some structural/format issues, example presence, schema validity for JSON blocks, and limited instruction conflicts. However, the declared purpose is substantially broader, claiming an evidence-based static benchmark over general quality, few-shot examples, safety and robustness, target-model compatibility, cross-model portability, production readiness, and high-priority improvement recommendations. This implementation is a narrow deterministic linter with heuristic checks and explicitly states it is 'not a complete benchmark score.' That makes the description materially overstate the skill’s primary capability. There are no suspicious undeclared external accesses or unrelated triggers; the mismatch is mainly that the real behavior is much narrower than advertised.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The short description and default prompt present mixed-language behavior in the skill metadata, which may implicitly impose a language/locale experience without explicit user opt-in. The policy requires avoiding forced language behavior unless the skill offers a choice or clearly documents a justified locale constraint.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.