Security audit

知识图谱工具(专业版)

Security checks for vulnerabilities and agentic risk

Overview

The skill is broadly aligned with a team knowledge-graph tool, but it asks agents to handle secrets, shared memory, and command execution with unclear boundaries.

Review this carefully before installing in an agent that can access real credentials, shared team folders, or conversation history. If used, require explicit user confirmation before vault access, memory import, writes, exports, or any exec-backed action; keep vault files and workspace paths tightly scoped; and do not rely on the documented vault secrecy claims for raw secret retrieval commands.

SkillSpector

Vulnerability Patterns

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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The document makes absolute secrecy claims about the vault while also documenting a command that emits the raw secret for downstream shell use. Even if intended for piping, exposing plaintext to process arguments, shell history, logs, or other tooling breaks the stated security model and can mislead users into unsafe handling of sensitive values.

Intent-Code Divergence

Low

Confidence: 80% confidence
Finding: The permission model is described inconsistently: one section says other agents are read-only with no write permission, while another says any agent can use the reader and authorized agents can write via configuration. Ambiguous authorization semantics are dangerous in a multi-agent shared-memory tool because operators may assume stronger isolation than is actually implemented, leading to unintended writes or data exposure.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The skill claims automatic activation based on broad terms like knowledge, collaboration, visualization, and memory-import concepts. In an agent environment with exec/read/grep/glob available, overly broad triggers can cause the skill to engage in unrelated conversations, increasing the chance of unauthorized file access, command execution, or unintended ingestion of sensitive conversational data.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The stated trigger condition covers generic needs like data analysis, reporting, statistics, and visualization, far beyond a narrowly bounded knowledge-graph skill. Because this skill also advertises memory import and command execution, such vague scope can cause unnecessary invocation in sensitive contexts and broaden the path for over-collection or unsafe actions.

Static analysis

No suspicious patterns detected.