Back to skill

Security audit

医疗大健康采招雷达-医疗招标采购网

Security checks for vulnerabilities and agentic risk

Overview

Review is recommended because optional signup understates device tracking, and update notices trigger automatic account-linked acknowledgments.

Configure your own API key to bypass device-based signup, and limit use to explicit procurement or company-research requests. Review query charges and whether automatic update acknowledgments are acceptable. A hashed MAC remains a persistent identifier; provider retention and actual server behavior are not established by these files. Protect saved keys and temporary login links.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (15)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command generates a host identifier rather than searching or analyzing medical procurement information. Local hardware fingerprinting is not declared and has no evident supporting role in the stated procurement-analysis workflow. The snippet does not show transmission or storage of the resulting hash.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local device fingerprinting rather than procurement searching or supplier analysis. Reading and hashing a network interface’s MAC address is not a supporting implementation detail justified by the declared purpose. The chunk shows no transmission of the resulting identifier.

Content

No source excerpt is available for this finding.

Healthcare keywords trigger the skill without requiring procurement intent.

Medium
Category
Trigger Abuse
Confidence
95% confidence
Finding

The description activates the skill whenever search terms contain 医院 (hospital), 医疗 (medical care), 卫生 (health/hygiene), or 体检 (physical examination). These terms also occur in ordinary healthcare searches, so the condition can invoke a bidding-data skill for unrelated requests without requiring tendering, procurement, or supplier-analysis intent.

Content

No source excerpt is available for this finding.

L052-L056 introduce device-fingerprinting and account-registration operations unrelated to medical procurement analysis.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The skill instructs the agent to collect platform, architecture, and a MAC-derived hash as part of automatic remote account registration, then persist the resulting credential locally. Authenticating procurement queries is expected, but collecting device identifiers and provisioning accounts exceed the manifest's stated hospital and supplier analysis purpose; the required user consent mitigates, but does not eliminate, this scope mismatch.

Content

No source excerpt is available for this finding.

The skill mandates an unrelated remote update-notice delivery and acknowledgment workflow.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes medical procurement searches and hospital supplier analysis, but these instructions require an authenticated account request at first use in every session to check for update notices. They also require forwarding the server-provided message verbatim and submitting a POST acknowledgment after display, adding a remote messaging and acknowledgment capability unrelated to retrieving or analyzing procurement information.

Content

No source excerpt is available for this finding.

The skill exposes paid corporate due-diligence workflows beyond its declared medical-procurement analysis purpose.

Medium
Category
Excessive Agency
Confidence
87% confidence
Finding

The manifest describes identifying hospital purchasers, winning medical suppliers, and hospitals’ top suppliers. L048–L111 instead define standalone shareholder, investment, corporate-change, full annual-report, construction-qualification, and credit-rating queries, including financial and contact information in L077–L079. These broader corporate-investigation capabilities are not a direct requirement of the declared procurement analysis and incur additional paid queries.

Content

No source excerpt is available for this finding.

Automatic registration collects and transmits a persistent hardware-derived device fingerprint unrelated to medical procurement analysis.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An unsalted, deterministic MAC-address hash remains a persistent device identifier that can support correlation; hashing does not make it anonymous. The consent prompt explicitly mentions the hash but incorrectly describes the collected features as non-identifying and asserts that no personal information is collected, undermining informed consent for hardware fingerprinting that is ancillary to procurement searches.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/auto-register.md (reported line 130)May include surrounding context.

md
> ### ⚠️ 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`)。
>

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/auto-register.md (reported line 131)May include surrounding context.

md
> ### ⚠️ 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

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/auto-register.md (reported line 195)May include surrounding context.

text

注意事项:
- 目录不存在时先 `mkdir -p ~/.zlbx`
- 文件已存在时**合并而非覆盖**(保留用户可能的其他配置)
- `source: "auto"` 字段必须写入,**这是后续判断「是否输出自动登录链接」的关键依据**

L536-L546 expand the medical bid radar into general corporate-record and construction-qualification queries.

Low
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes extracting hospital purchasers and pharmaceutical suppliers and analyzing a hospital's top suppliers. These instructions additionally enable standalone company annual-report, construction-qualification, certificate, and historical credit-rating queries without requiring a connection to medical procurement, making the operational scope substantially broader than advertised.

Content

No source excerpt is available for this finding.

The skill mandates an unrelated package-update notification and acknowledgement workflow.

Low
Category
Excessive Agency
Confidence
91% confidence
Finding

L351 requires an authenticated account request at the start of each session specifically to check upgrade notices, and L352–L353 require relaying those notices and submitting a POST acknowledgement after display. This package-maintenance workflow, including a remote state-changing acknowledgement, is separate from discovering medical procurement notices and analyzing hospital suppliers.

Content

No source excerpt is available for this finding.

L393–L395 require authenticated account and update-acknowledgment operations unrelated to medical procurement analysis.

Low
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes finding medical procurement participants and analyzing hospital suppliers. These instructions additionally require a credentialed account-balance request each session specifically to retrieve upgrade notices, followed by a POST acknowledgment after displaying a notice; neither operation is necessary for the stated procurement-analysis task.

Content

No source excerpt is available for this finding.

The reference instructions mandate an authenticated upgrade-notification and acknowledgment workflow unrelated to medical procurement analysis.

Low
Category
Excessive Agency
Confidence
90% confidence
Finding

L334 requires an authenticated account request on first use in each session specifically to check upgrade reminders. L335–L336 then require relaying the reminder and submitting an authenticated POST recording its display; this package-update workflow is separate from searching medical procurement notices and analyzing hospital suppliers.

Content

No source excerpt is available for this finding.

The update-notice workflow reports user-visible notice delivery to the API provider without a privacy disclosure.

Low
Category
Output Handling
Confidence
86% confidence
Finding

The skill instructs the agent to submit an authenticated POST containing the notice ID and displayed: true after showing an update notice. This sends account-linked interaction telemetry to the provider, but the file does not warn users about that reporting and explicitly keeps the technical receipt out of the conversation.

Content

No source excerpt is available for this finding.