Security audit

管理体系文件审核技能

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a coherent local document-audit helper that reads user-provided documents and generates a local Markdown report.

Install only if you are comfortable letting the agent read the audit documents you provide and save a local Markdown report that may include confidential process, compliance, or risk details. Keep provided paths limited to the intended document set and store generated reports in an appropriate workspace.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 95% confidence
Finding: The skill invokes a local report-generation script that writes a Markdown file, yet no explicit permission model or disclosure is declared for file creation. This can lead to unexpected persistence of potentially sensitive audit content on disk, especially in environments where users assume the skill is analysis-only.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 91% confidence
Finding: The documented purpose is management-system analysis, but the described implementation includes report generation and persistence while not evidencing that substantive analysis is actually performed by code. This mismatch can mislead users about what the skill does, causing over-trust in generated outputs and unexpected handling of local data.

Description-Behavior Mismatch

Low

Confidence: 81% confidence
Finding: Claiming the skill is a 'pure document/script skill' with no external risk downplays the fact that it executes a local Python script and writes files. While not inherently malicious, such understatement reduces informed consent and can cause users to expose sensitive content to local storage unintentionally.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The TRACE section presents the skill as low-risk and without external dependency concerns, but the skill actually performs local execution and file output. Security-trust labels that overstate safety can cause users or reviewers to skip scrutiny, increasing the chance of unsafe deployment or unintended data retention.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: Allowing users to provide arbitrary file paths for the agent to read, without a clear warning or restriction, creates a local file exposure risk. Users may unintentionally cause the skill to access sensitive files outside the intended document set, and the content could then be processed or reflected in outputs.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The report-generation step writes a timestamped Markdown report to disk, but users are not clearly told that analysis results will be persisted locally. Because audit content may include confidential compliance gaps, organization names, process details, or file excerpts, silent persistence increases data leakage and retention risk.

Static analysis

No suspicious patterns detected.