Back to skill

Security audit

AI招投标分析师-全网招中标智能分析

Security checks across malware telemetry and agentic risk

Overview

The skill's bid-analysis features are coherent, but it also directs agents into account creation, device fingerprint collection, credential persistence, and auto-login billing flows that need careful review.

Review this before installing if you do not want an agent to inspect local device identifiers, create a vendor account, store an API key in your home directory, or generate auto-login recharge links. A safer setup would require you to provide an API key manually and would ask before any registration, device collection, or billing-related action.

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 (12)

Description-Behavior Mismatch

Medium
Confidence
97% confidence
Finding
The skill expands beyond bid analytics into credential discovery, automatic account registration, and persistence of a newly obtained API key in a local file. This is dangerous because it causes the agent to access local secrets and modify the user's environment for a purpose not strictly required to answer an analytics request, without explicit informed consent.

Context-Inappropriate Capability

Medium
Confidence
98% confidence
Finding
The documented fallback collects device characteristics including platform, architecture, and a MAC-derived hash to register an account, even though those identifiers are not necessary for tender/bid analysis itself. Collecting fingerprinting material for silent registration creates privacy risk, enables cross-session/device tracking, and normalizes over-collection unrelated to the user’s requested task.

Description-Behavior Mismatch

Medium
Confidence
91% confidence
Finding
The skill instructs the agent to generate an auto-login recharge link derived from the current API key context when quota is exhausted. This extends the skill into account-access and payment-adjacent behavior outside core analytics, and could steer users into unintended account actions or expose session bootstrap tokens if mishandled.

Description-Behavior Mismatch

High
Confidence
98% confidence
Finding
This document introduces an automatic registration flow, persistent API key storage, device-based account recovery/link generation, and monetization/recharge behaviors that materially exceed the declared purpose of a tender/bid analysis skill. Even though some privacy minimization is described, the skill still instructs collection of device fingerprints and transmission to an external service, creating hidden account lifecycle behavior unrelated to the user’s analysis request.

Context-Inappropriate Capability

High
Confidence
99% confidence
Finding
The skill directs the agent to run shell/PowerShell commands to collect platform, architecture, and a hashed MAC-derived identifier, then use that data for external registration. For a bid-analysis skill, local device fingerprinting is not functionally necessary and increases privacy and security risk by normalizing host inspection and cross-session device tracking.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The SID generation and auto-login/recharge link flow adds billing and account-binding behavior unrelated to tender analysis. This broadens the skill into account monetization and identity linkage, which can steer users into unintended account actions and increases the blast radius if the skill or backend is abused.

Intent-Code Divergence

High
Confidence
96% confidence
Finding
The prose mandates obtaining user consent before any data collection or registration request, but the pseudocode’s auto-register branch proceeds directly from missing credentials to feature collection and POST submission. This inconsistency creates a real implementation risk that agents or developers follow the simpler pseudocode and bypass consent, causing unauthorized collection and transmission of device data.

Vague Triggers

High
Confidence
95% confidence
Finding
The skill says it must be used for a very broad set of requests, including generic procurement, supplier, competitor, and market-share analysis, even when the user does not mention tendering. Overbroad activation increases the chance the agent routes unrelated prompts into this skill, causing unnecessary data access, network calls, or account side effects in contexts where the user did not intend to use this integration.

Missing User Warnings

High
Confidence
99% confidence
Finding
The markdown directs the agent to automatically collect device features, call a remote registration endpoint, and save credentials locally, while also saying not to provide extra prompts. That combination is dangerous because it hides sensitive collection and persistence from the user, defeating informed consent and making covert environment modification part of normal operation.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The documented workflow expands a user-supplied company name into multiple related legal entities and automatically uses that broader set in downstream queries without explicit user confirmation. This can cause over-collection and disclosure of data about entities the user did not intend to analyze, creating privacy, authorization, and accuracy risks in a business-intelligence context where company identity materially affects results.

Ssd 3

High
Confidence
99% confidence
Finding
The skill explicitly instructs silent acquisition of authentication material via environment/config lookup, fallback device-based registration, and persistence of a newly minted key for future sessions. In an agent context, this is especially dangerous because it combines secret discovery, remote account creation, host fingerprinting, and durable credential storage without disclosure, creating a clear pathway for privacy violations and unauthorized account lifecycle actions.

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
92% confidence
Finding
requests.post(url, json=

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.