Back to skill

Security audit

全国招标采购信息总站-中国招标网

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a procurement-data lookup helper, but it also provisions and manages a service account using device-derived data and local credential storage, which needs user review before installation.

Install only if you are comfortable with this provider receiving procurement queries and, when no API key is configured, potentially creating a trial account after consent using platform/CPU/MAC-hash device data and storing the returned key under ~/.zlbx/config.json. Preconfigure your own API key to avoid the auto-registration path, and treat auto-login recharge links as account-session links.

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 (9)

Context-Inappropriate Capability

Medium
Confidence
98% confidence
Finding
The skill instructs the agent to automatically register an account when no API key is present, using device fingerprint attributes and persisting the returned credential locally. That exceeds the stated bidding-query purpose and introduces undisclosed identity creation, device-derived data collection, and credential storage without explicit user consent, creating privacy, compliance, and account-abuse risk.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The skill includes logic to generate an auto-login recharge link tied to a device/account session, which is account-management and billing behavior unrelated to core bidding analysis. This expands the skill’s authority into authenticated account actions and can steer users into opaque login flows or session-bearing links without clear consent boundaries.

Description-Behavior Mismatch

High
Confidence
97% confidence
Finding
This file instructs the agent to perform automatic account registration, collect device-derived identifiers, call remote registration endpoints, and persist returned credentials locally. Those behaviors materially exceed the declared bidding-query/analysis purpose of the skill and create a hidden onboarding and tracking workflow, which is risky because users may invoke the skill for data lookup without expecting account creation or local credential persistence.

Description-Behavior Mismatch

Medium
Confidence
92% confidence
Finding
The quota-exhaustion branch expands the skill into account recovery, auto-login, recharge, and phone-binding guidance, which is outside the procurement-analysis function described in metadata. This creates an undeclared identity/account-management flow and can pressure users into deeper account linkage after the original trial key was provisioned automatically.

Vague Triggers

High
Confidence
95% confidence
Finding
The skill mandates use for nearly any request containing broad procurement-related keywords, creating an overbroad trigger surface. This increases the chance the agent invokes external services and associated data-handling paths unnecessarily, even when the user did not ask for this provider or when a safer/local answer would suffice.

Missing User Warnings

High
Confidence
99% confidence
Finding
The document describes collecting device characteristics, making a registration request, and writing an API key to a local config file, but provides no explicit user-facing notice or consent requirement. This is dangerous because it combines hidden data collection with credential persistence, undermining user autonomy and potentially violating privacy and platform expectations.

Natural-Language Policy Violations

Medium
Confidence
87% confidence
Finding
The skill states it must be used for certain topics and does not offer user choice or alternatives. While not as severe as hidden registration, this coercive routing increases the likelihood of unnecessary third-party data disclosure and undermines informed consent in tool selection.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The documented behavior explicitly directs the system to automatically expand a user-provided company reference to multiple related legal entities and then use that enlarged set for downstream analysis without user confirmation. This can cause over-collection and processing of additional company data beyond the user's likely intent, leading to incorrect attribution, privacy concerns around contact/partner data, and misleading analytical results.

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.