Back to skill

Security audit

企业背调助手-给个公司名出份报告

Security checks across malware telemetry and agentic risk

Overview

The skill’s company-reporting purpose is mostly coherent, but it needs review because it can auto-register accounts, fingerprint the device, persist credentials and reports, and expose signed links or contact data.

Review before installing. Use this skill only if you are comfortable with a vendor-backed company-intelligence workflow that may create a trial account, send a hashed device identifier after consent, store an API key locally, save reports to disk, and include signed platform links or contact information in outputs. Prefer preconfiguring ZLBX_API_KEY to avoid auto-registration, avoid sharing exported HTML unless you have checked it for signed links and contact data, and use contact lookup only for legitimate business purposes.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (15)

Lp3

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding
The skill explicitly instructs local file reads from `~/.zlbx/config.json` and local file writes to `~/zlbx-company-intel-files/`, yet no corresponding permissions are declared. This creates a capability/permission mismatch that can bypass user expectations and platform controls, especially because it involves reading credential material and persisting generated artifacts on disk.

Description-Behavior Mismatch

High
Confidence
98% confidence
Finding
This document embeds a full auto-registration, device-fingerprinting, credential persistence, and auto-login workflow inside a skill whose declared purpose is company due-diligence reporting. That is a scope expansion from business analysis into local system inspection, account creation, and secret handling, which creates unnecessary privacy and security risk and could cause an agent to perform sensitive actions unrelated to the user’s requested research task.

Context-Inappropriate Capability

High
Confidence
99% confidence
Finding
The skill instructs the agent to collect device fingerprints, including architecture and a SHA256-derived MAC identifier, then transmit them to a remote service to deduplicate trial accounts. Even with hashing and a consent prompt, this is unrelated to company-background analysis and turns the skill into a mechanism for local host enumeration and tracking, increasing privacy risk and creating an avenue for unauthorized system interrogation by an agent.

Context-Inappropriate Capability

Medium
Confidence
92% confidence
Finding
The document directs the agent to inspect environment variables and local config files for API keys and later persist credentials to a user config path. Handling secrets and local credential state is outside the stated intelligence-reporting function and increases the blast radius of the skill by normalizing secret discovery and storage behavior within a content-analysis tool.

Context-Inappropriate Capability

Medium
Confidence
94% confidence
Finding
The skill generates auto-login and recharge links, including phone-binding guidance, when balance is exhausted. This is unrelated to company due-diligence output and extends the skill into account-lifecycle manipulation, which can mislead users, create phishing-like behavior patterns, and encourage the agent to facilitate sensitive authentication flows.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The template instructs the agent to create a local file by invoking an external Python script as a default behavior, even though the skill is described as producing a report in conversation. This expands the skill's side effects from content generation to code execution and filesystem writes, which can surprise users, increase attack surface, and create opportunities for unsafe path handling or unintended local data exposure.

Vague Triggers

High
Confidence
91% confidence
Finding
The activation rules are intentionally overbroad: they require the skill to trigger whenever a user wants to 'understand a company' even if they did not ask for background checks. That can cause the agent to route ordinary queries into a workflow that performs external API calls, potential auto-registration prompts, local file access, and report generation beyond the user's likely intent.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill explicitly supports retrieving and displaying project contact phone numbers, including full numbers for paid accounts, but does not include a clear privacy, purpose-limitation, or anti-abuse warning. In a background-check/intelligence skill, this increases the risk of misuse for unsolicited contact, scraping, or doxxing of individuals whose data is only partially contextualized as business contacts.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The instructions require creating an HTML file and revealing its filesystem path without warning users in the skill description that local file creation will occur. This undermines user expectations and transparency, and may leak environment details such as directory structure, usernames, or host conventions through the disclosed absolute path.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The workflow explicitly instructs the agent to output company page URLs "原样输出" even when they contain an `sk` parameter. If `sk` is an access-bearing token, echoing it into reports can leak authenticated links to end users, logs, screenshots, or downstream systems, enabling unintended access or token reuse.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The workflow requires attaching announcement links with embedded `sk` values "原样带 sk". Reproducing tokenized URLs in output broadens token exposure through chat history, exports, and sharing, which can grant unauthorized access to bid data or allow session-style replay if the token is reusable.

Missing User Warnings

High
Confidence
92% confidence
Finding
The contact workflow instructs showing phone numbers exactly as returned, including full numbers for paid accounts, and frames this as a product upsell without any privacy or lawful-use guardrails. This creates a clear path for disclosure of personal contact data, increasing risk of privacy violations, spam, targeted harassment, or misuse of scraped procurement contacts.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The script renders contact details and account-scoped URLs directly into a self-contained HTML report that is intended for offline opening, printing, and forwarding. Even though values are escaped for XSS safety, this still creates a confidentiality risk because sensitive business/contact data can be redistributed outside the original access-control context without a strong warning, minimization, or access check at render time.

Missing User Warnings

Low
Confidence
85% confidence
Finding
The script writes the generated report to disk automatically in a predictable local directory, creating a persistent copy of potentially sensitive due-diligence content. This increases exposure through local file compromise, shared workstations, backups, sync services, or accidental forwarding, especially given the report's stated use for printing and sharing.

External Transmission

Medium
Category
Data Exfiltration
Content
> ### ⚠️ 请求体必须用 JSON 序列化函数生成,不要手拼字符串
>
> 用 `json.dumps(payload)` / `requests.post(url, json=payload)` / `JSON.stringify(payload)`,
> 或 `curl -d @file`;**不要用字符串拼接,也不要用 Python 的 `str(dict)`**
> (后者产出单引号,服务端会报 `Expecting property name enclosed in double quotes`)。
>
Confidence
88% confidence
Finding
requests.post(url, json=

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.