Back to skill

Security audit

企业查询助手(免费版)

Security checks for vulnerabilities and agentic risk

Overview

This skill is mainly a company lookup helper, but it asks for broad command/file authority and contains unclear or conflicting capability claims that users should review before installing.

Install only if you are comfortable with an agent running Node-based commands for company lookups and sending company or person identifiers to an external business-information API. Treat the SEO, generic file-processing, and risk-screening claims as unresolved scope issues, and avoid using it for sensitive due-diligence workflows unless the publisher clarifies the actual available tools, API provider, and data handling.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The manifest and description position the skill as a company-information lookup tool, but also advertise unrelated SEO and broad automation use cases. This scope inflation can cause agents to invoke the skill in contexts where its actual behavior and data flows are not appropriate, increasing the chance of misuse and policy bypass through misclassification.

Description-Behavior Mismatch

High
Confidence
95% confidence
Finding
The documentation states the free edition does not support risk screening, yet the query-dimension table exposes multiple risk-related endpoints such as executed-person, dishonesty, anomaly, and administrative-penalty lookups. This contradiction can mislead users and agents into accessing more sensitive due-diligence data than advertised, undermining trust boundaries and compliance expectations.

Description-Behavior Mismatch

High
Confidence
95% confidence
Finding
The free-version limitations claim advanced risk and history capabilities are locked, but earlier sections document tool categories that appear to expose those same areas. Such contradictory access claims can lead agents or users to over-trust the stated restrictions while still attempting sensitive lookups, creating a deceptive capability surface.

Intent-Code Divergence

Medium
Confidence
86% confidence
Finding
The capability-coverage statement asserts that all declared scenarios—including SEO optimization and ranking use cases—are supported, while the documented functions only implement enterprise lookup workflows. This can misroute agents into using the skill outside its intended domain and weaken tool-selection safety controls based on metadata accuracy.

Intent-Code Divergence

High
Confidence
93% confidence
Finding
The free-version limitations conflict with the documented query-dimension table by saying risk-screening features are locked while still listing risk-related tools. This inconsistency creates ambiguity around what sensitive data can be queried and may result in unauthorized or unexpected access attempts by downstream agents.

Context-Inappropriate Capability

Medium
Confidence
89% confidence
Finding
The documentation introduces generic file-reading and file-writing capabilities even though the stated purpose is company-information querying. Broad file I/O claims expand the perceived authority of the skill and can encourage unsafe agent behavior, especially since the manifest also includes write access not justified by the core use case.

Context-Inappropriate Capability

Medium
Confidence
77% confidence
Finding
The discover_tools method exposes a broader capability than the stated company-query purpose by allowing generic tool discovery based on an arbitrary keyword. In an agent setting, this can expand the reachable tool surface and enable capability enumeration or misuse beyond the intended least-privilege scope.

Context-Inappropriate Capability

Medium
Confidence
69% confidence
Finding
The implementation delegates work to an external Node.js script via subprocesses, introducing an execution boundary and dependency on local runtime and script integrity that is broader than the end-user purpose suggests. If the script path, runtime environment, or underlying script is tampered with, the skill could execute unintended code with the agent's privileges.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The invocation guidance says to use the skill for common SEO tasks unrelated to company lookup. Overbroad triggers increase the likelihood that an agent will select this skill in inappropriate contexts, causing unnecessary external calls, disclosure of user queries, and privilege expansion through irrelevant tool use.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The examples encourage external API-backed commands using a built-in shared key without clearly warning that company or person identifiers will be transmitted to a third-party service. This is a privacy and compliance issue because users and agents may submit sensitive investigative queries without informed consent or data-handling disclosure.

Unvalidated Output Injection

High
Category
Output Handling
Content
self.runtime, self.script_path, "call", tool_name,
            "--params", json.dumps(params, ensure_ascii=False)
        ]
        result = subprocess.run(
            cmd, capture_output=True, text=True, timeout=30, encoding="utf-8"
        )
        if result.returncode == 0:
Confidence
71% confidence
Finding
While this call is not shell injection in the Python layer, it accepts an arbitrary tool_name and forwards attacker-controlled params to an external script that can invoke different tools. In an agent context, this creates a tool-surface expansion and indirect command/capability invocation risk if callers can influence tool_name or if the Node layer lacks strict allowlisting.

Static analysis

No suspicious patterns detected.