Back to skill

Security audit

智慧交通与轨道交通采招分析-基建商机雷达

Security checks across malware telemetry and agentic risk

Overview

This skill analyzes transit procurement data, but it also creates a third-party account using device-derived identifiers and stores credentials locally, so it needs review before installation.

Install only if you are comfortable with the skill using an external Zhiliaobiaoxun service, creating a trial account after consent when no API key is present, sending a hashed device identifier for trial deduplication, and storing the returned API key in ~/.zlbx/config.json. To avoid auto-registration, configure ZLBX_API_KEY yourself before use and review company-matching results when exact legal-entity scope matters.

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 Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

Description-Behavior Mismatch

Medium
Confidence
97% confidence
Finding
The skill documentation introduces automatic account registration and credential persistence, which are unrelated to the stated procurement-analysis purpose and expand the skill's authority into account lifecycle management. This creates unnecessary side effects, can enroll users with a third-party service without explicit consent, and increases privacy and compliance risk.

Context-Inappropriate Capability

High
Confidence
99% confidence
Finding
The skill instructs collection of device characteristics (platform, architecture, MAC-derived hash) to register accounts automatically, even when collection fails partially. Device fingerprinting is not justified by a bid-analysis workflow and can enable tracking, deanonymization, and silent account creation with persistent identifiers.

Context-Inappropriate Capability

Medium
Confidence
96% confidence
Finding
The instructions tell the agent to write API credentials into a local config file under the user's home directory, which exceeds the documented analytics scope and introduces a local secret-storage side effect. Persisting credentials without explicit user approval can expose secrets to other local processes, backups, or shared environments.

Context-Inappropriate Capability

Medium
Confidence
93% confidence
Finding
The skill adds generation of automatic login or recharge links tied to account monetization flows, which is outside the procurement-analysis function and nudges the agent into account-access brokerage. Such links may enable session bootstrapping or phishing-like flows and normalize unsafe authentication handling inside an unrelated skill.

Description-Behavior Mismatch

High
Confidence
98% confidence
Finding
This document instructs the skill to perform automatic account registration, API key persistence, session handling, and quota-management flows that go well beyond the declared rail/transit bid-analysis purpose. Even if framed as onboarding, these behaviors expand the skill into credential lifecycle management and create unnecessary security and privacy exposure, including unsolicited external account creation and local secret storage.

Context-Inappropriate Capability

High
Confidence
99% confidence
Finding
The skill directs collection of device fingerprinting attributes, including CPU architecture and a hashed MAC address, for trial-account deduplication unrelated to procurement-analysis functionality. A MAC hash is still a stable pseudonymous identifier, and collecting it from local interfaces increases tracking risk and violates data minimization for the stated skill purpose.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The quota exhaustion flow generates auto-login links and nudges users into phone-binding and recharge actions, which are unrelated to the declared analytical role of the skill. This broadens the skill into marketing/account-conversion behavior and can expose users to session-link handling risks if such links are displayed or mishandled in an untrusted environment.

Vague Triggers

High
Confidence
92% confidence
Finding
The activation rules are overly broad and mandatory, including cases where the user does not even mention traffic but the request is loosely related to procurement. Over-broad routing increases the chance that unrelated user queries are sent to this skill, causing unnecessary data exposure to its external API and bypassing user intent boundaries.

Missing User Warnings

High
Confidence
98% confidence
Finding
The markdown describes device-based registration and local credential storage without an explicit privacy warning or user consent for the collection and persistence side effects. In context, this is more dangerous because the skill is presented as a domain-analysis assistant, so users would not reasonably expect fingerprinting, account creation, and local file modification.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The documentation explicitly instructs the agent to automatically select 'all matching companies' and continue downstream analysis without user confirmation. This can cause the agent to query and analyze the wrong legal entities, over-broaden data collection to subsidiaries the user did not intend, and produce misleading business conclusions or expose unnecessary company/contact data.

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
93% confidence
Finding
requests.post(url, json=

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.