ClawHub
Back to skill

Security audit

Azure智能体框架工具-免费版

Security checks for vulnerabilities and agentic risk

Overview

This is mostly an Azure Agent Framework guide, but its permissions and activation scope are inconsistent enough that users should review it before installing.

Install only if you want a Chinese-language Azure AI Foundry Agent Framework guide and are comfortable with an agent that may use command execution for setup. Review Azure credential scope, cloud resources, hosted tools, and package installs before following examples; avoid letting it activate for unrelated general chat or LLM tasks.

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (3)

Scope Creep

Medium
Confidence
97% confidence
Finding
The manifest declares `allowed-tools: read exec` while a later `tools` section includes `write`, creating a permission mismatch that can mislead policy enforcement and reviewers about the skill's actual capabilities. In an agent ecosystem, ambiguous capability declarations increase the risk that the skill is granted broader filesystem access than expected or that downstream tooling enforces the wrong permission set.

Intent-Code Divergence

Medium
Confidence
88% confidence
Finding
The skill is presented as a documentation-only guide, but later text claims the skill itself supports create/query/modify/delete/export/save-style operations. That mismatch can cause an invoking agent or user to overtrust the skill as an active executor and perform riskier actions, especially when combined with `exec` and the implied `write` capability.

Vague Triggers

Medium
Confidence
83% confidence
Finding
The trigger condition is extremely broad—covering general AI model calls, chat, agent orchestration, and LLM applications—so the skill may activate in many unrelated contexts. Over-broad auto-selection increases the chance that an agent invokes a high-capability skill unnecessarily, exposing users to unintended command execution, network use, or unsafe guidance.

Static analysis

No suspicious patterns detected.