Back to skill

Security audit

招投标尽调助手-合同流水看经营实态

Security checks across malware telemetry and agentic risk

Overview

The skill’s due-diligence function is coherent, but it needs Review because it uses device-based auto-registration, stores credentials locally, and preserves login-bypass links in shareable reports.

Install only if you are comfortable sending company queries to Zhiliaobiaoxun, allowing an optional consent-gated trial signup that sends a hashed device identifier, and storing the resulting API key in ~/.zlbx/config.json. To reduce risk, preconfigure ZLBX_API_KEY instead of auto-registering, avoid requesting contacts unless needed, and treat generated HTML reports as sensitive because they may contain sk signed links that grant direct access to platform pages.

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 (13)

Lp3

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding
The skill explicitly states it will read from and write to local paths such as `~/.zlbx/config.json` and `~/zlbx-company-intel-files/`, yet no corresponding permissions are declared. Undeclared file access weakens the trust boundary for users and hosts because the skill can handle local credential material and generate local artifacts without transparent capability scoping.

Description-Behavior Mismatch

High
Confidence
98% confidence
Finding
This file embeds a full auto-registration, remote account provisioning, and local credential persistence workflow inside a skill whose stated purpose is bidding due-diligence. That is dangerous because it expands the skill's authority from analysis into local system inspection, outbound registration, and secret storage, creating an unnecessary capability path that could be abused for tracking, account creation without informed understanding, or persistence on the host.

Context-Inappropriate Capability

High
Confidence
97% confidence
Finding
The skill instructs the agent to collect platform, architecture, and a hashed MAC-derived identifier to support device-based trial deduplication, which is unrelated to due-diligence analysis. Even though the raw MAC is hashed, it still functions as a stable device fingerprint and is sent to an external service, enabling cross-session tracking and creating privacy risk disproportionate to the skill's purpose.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The instructions direct the agent to create and modify ~/.zlbx/config.json, merge configuration, and persist API credentials locally, despite this being unrelated to performing due-diligence analysis. This broadens the skill into local state management and secret handling, increasing the risk of overwriting user configuration, leaving persistent secrets on disk, and normalizing hidden side effects from a business-analysis skill.

Context-Inappropriate Capability

Medium
Confidence
97% confidence
Finding
The template explicitly requires exposing full API-returned URLs, including `sk` login-bypass parameters, directly to end users and preserving them in both Markdown and generated HTML. If `sk` functions as an authenticated access token or deep-link bypass, this leaks bearer-style access that can be copied, shared, indexed, or replayed by unintended parties, potentially granting unauthorized access to company pages or premium data.

Context-Inappropriate Capability

Medium
Confidence
88% confidence
Finding
The workflow explicitly instructs fallback to general WebSearch when the bidding database has no record, which expands the skill from tender-data due diligence into broader open-source intelligence collection. This increases the chance of collecting irrelevant, inaccurate, or privacy-impacting information outside the skill’s declared scope, especially because the fallback is presented as a normal continuation path rather than a separately consented mode.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
The public-risk section operationalizes multi-round general WebSearch for lawsuits, penalties, dishonesty status, and news, creating a deliberate capability expansion beyond procurement data into general reputational and legal profiling. Even with cautionary wording, this can surface unverified or stale allegations and encourage broad external data gathering that may exceed user expectations for a bidding-data skill.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The optional contact-discovery step shifts the skill from company due diligence into identifying and surfacing individual contact data, which is not necessary for the core purpose of assessing operating reality, customer concentration, or procurement performance. This materially broadens the skill’s use into lead generation or direct outreach, increasing privacy and misuse risk.

Vague Triggers

High
Confidence
88% confidence
Finding
The skill says it 'must' be used even when the user does not explicitly ask for due diligence, creating an overly broad trigger that can capture adjacent tasks and steer the agent into unnecessary data collection, API usage, and cost-incurring actions. In a security context, overbroad routing is dangerous because it bypasses user intent minimization and can expose company names and investigative queries to external services without sufficiently specific consent.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The documentation instructs the skill to send authenticated requests containing company identifiers and potentially contact information to a third-party API, but it does not disclose this data sharing or provide any privacy warning, consent boundary, or minimization guidance. In this due-diligence context, the tool can retrieve and expose project contact details and business relationship data, so silent transmission and disclosure can create privacy, compliance, and user-trust risks.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The workflow instructs displaying phone numbers 'as returned' and explicitly allows full phone disclosure for some account tiers, while the skill description does not disclose a personal-data handling purpose or provide meaningful privacy safeguards. This creates a direct path for exposing identifiable individual contact information for outreach or scraping under the guise of due diligence.

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=

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`)。
>
> 历史教训:曾有版本采集 `home_path`,Windows 的 `C:\Users\alice` 直接拼进 JSON 字符串时
Confidence
90% confidence
Finding
curl -d

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.